id
stringlengths 9
104
| author
stringlengths 3
36
| task_category
stringclasses 32
values | tags
sequencelengths 1
4.05k
| created_time
unknowndate 2022-03-02 23:29:04
2025-03-18 02:34:30
| last_modified
stringdate 2021-02-13 00:06:56
2025-03-18 09:30:19
| downloads
int64 0
15.6M
| likes
int64 0
4.86k
| README
stringlengths 44
1.01M
| matched_bigbio_names
sequencelengths 1
8
|
|---|---|---|---|---|---|---|---|---|---|
flowaicom/Flow-Judge-v0.1-AWQ | flowaicom | text-generation | [
"transformers",
"safetensors",
"phi3",
"text-generation",
"lm-judge",
"evaluation",
"nlp",
"conversational",
"autoawq",
"custom_code",
"en",
"base_model:microsoft/Phi-3.5-mini-instruct",
"base_model:quantized:microsoft/Phi-3.5-mini-instruct",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"4-bit",
"awq",
"region:us"
] | "2024-09-15T12:30:58Z" | 2024-10-09T05:59:56+00:00 | 642 | 6 | ---
base_model: microsoft/Phi-3.5-mini-instruct
language:
- en
library_name: transformers
license: apache-2.0
metrics:
- accuracy
- f1
- precision
- recall
- pearsonr
- spearmanr
- kendall-tau
model_name: Flow-Judge-v0.1-AWQ
pipeline_tag: text-generation
tags:
- lm-judge
- phi3
- evaluation
- nlp
- conversational
- autoawq
inference: false
model_creator: Flow AI
model_type: phi3.5
quantized_by: Flow AI
---
# Flow-Judge-v0.1-AWQ
- Original model: [Flow-Judge-v0.1](https://huggingface.co/flowaicom/Flow-Judge-v0.1)
- Model collection: [Flow-Judge-v0.1 models](https://huggingface.co/collections/flowaicom/flow-judge-v01-66e6af5fc3b3a128bde07dec)
- Technical report: [Flow Judge: An Open Small Language Model for LLM System Evaluations](https://huggingface.co/flowaicom/Flow-Judge-v0.1)
- Model website: [flow-ai.com/judge](https://www.flow-ai.com/blog/flow-judge)
- About us: [Flow AI](https://www.flow-ai.com/about)
<!-- description start -->
## Description
This repo contains AWQ safetensors quant for [Flow-Judge-v0.1](https://huggingface.co/flowaicom/Flow-Judge-v0.1).
## Quantization config
```python
quant_config = { "zero_point": True, "q_group_size": 128, "w_bit": 4, "version": "GEMM" }
model = AutoAWQForCausalLM.from_pretrained(merged_path, **{"low_cpu_mem_usage": True, "use_cache": False},
attn_implementation="flash_attention_2", torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(lora_path, trust_remote_code=False)
model.quantize(tokenizer, quant_config=quant_config)
model.save_quantized(quant_path)
tokenizer.save_pretrained(quant_path)
```
## Library versions used for quantization
```raw
autoawq 0.2.6
autoawq_kernels 0.0.7
```
## Running the AWQ model
First install the flow judge library
```shell
git clone https://github.com/flowaicom/flow-judge
cd flow-judge
pip install -e ".[vllm]"
```
Quickstart with Python:
```python
from flow_judge import Vllm, Llamafile, Hf, EvalInput, FlowJudge
from flow_judge.metrics import RESPONSE_FAITHFULNESS_5POINT
from IPython.display import Markdown, display
# If you are running on an Ampere GPU or newer, create a model using VLLM
model = Vllm(quantization=True)
# If you have other applications open taking up VRAM, you can use less VRAM by setting gpu_memory_utilization to a lower value.
# model = Vllm(gpu_memory_utilization=0.70)
# Or create a model using Llamafile if not running an Nvidia GPU & running a Silicon MacOS for example
# model = Llamafile()
# Initialize the judge
faithfulness_judge = FlowJudge(
metric=RESPONSE_FAITHFULNESS_5POINT,
model=model
)
# Sample to evaluate
query = ...
context = ...
response = ...
# Create an EvalInput
# We want to evaluate the response to the customer issue based on the context and the user instructions
eval_input = EvalInput(
inputs=[
{"query": query},
{"context": context},
],
output={"response": response},
)
# Run the evaluation
result = faithfulness_judge.evaluate(eval_input, save_results=False)
# Display the result
display(Markdown(f"__Feedback:__\n{result.feedback}\n\n__Score:__\n{result.score}"))
```
Discover more at our repository [https://github.com/flowaicom/flow-judge](https://github.com/flowaicom/flow-judge)
# Original model card: Flow-Judge-v0.1
<p align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/63368577d184e6b53c50e6d0/6kSJKgPh2pDh4tA-Ky0xW.png" alt="Centered image">
</p>
<p align="center">🚀 <a href="https://www.flow-ai.com/judge">Flow Judge</a> | 📄 <a href="https://www.flow-ai.com/blog/flow-judge">Technical report</a> | 💻 <a href="https://github.com/flowaicom/flow-judge">flow-judge</a></p>
## Model Summary
Flow-Judge-v0.1 is a compact yet powerful 3.8B model that offers customizable LLM system evaluations across various fields. The model inherits it's architecture from Phi-3.5-mini instruct model which enables Flow-Judge to deliver high-quality results while maintaining a small footprint. Despite its smaller size, it achieves performance comparable to larger models in both held-out and out-of-domain benchmarks. Flow-Judge-v0.1 supports multiple scoring scales, provides qualitative feedback, and generates structured evaluation outputs. Trained on a smaller synthetic dataset, it represents an efficient approach to AI development. Released under the Apache 2.0 license, Flow Judge is an open and accessible model suitable for developers and companies seeking cost-effective and rapid evaluations using custom rubrics.
__Quantized weights__
- [flowaicom/Flow-Judge-v0.1-AWQ](https://huggingface.co/flowaicom/Flow-Judge-v0.1-AWQ)
- [flowaicom/Flow-Judge-v0.1-GGUF](https://huggingface.co/flowaicom/Flow-Judge-v0.1-GGUF)
__Quickstart__
- [Quickstart](https://github.com/flowaicom/flow-judge/examples/1_quickstart.ipynb)
## Intended Use Case
Flow Judge is intended to be used on custom LLM system evaluation tasks.
- Customizable evaluations: Users can define their own evaluation criteria and rubrics, tailoring Flow Judge to their specific needs and requirements. This flexibility allows for the creation of highly targeted assessments that accurately measure performance of their LLM system
- Flow Judge supports three different scoring scales:
- Pass/fail: Suitable for binary assessments, such as determining whether a piece of text meets a specific standard or contains errors.
- 3-Likert: Allows for more granular evaluations, with scores ranging from negative to neutral to positive. Useful for assessing the overall quality or sentiment of a piece of text.
- 5-Likert: Provides an even more nuanced assessment, with scores ranging from strongly negative to strongly positive, enabling users to capture subtle differences in quality or sentiment.
- Easy to interpret results:
- Flow Judge produces structured evaluations with `<feedback>` and `<score>` tags.
- Qualitative feedback: Flow Judge detects errors and grades outputs and provides qualitative feedback that explains its reasoning for assigning a particular score from the rubric while highlighting problematic parts of the responses.
- Score: Based on a grading rubric Flow Judge will return a numerical score on binary, likert-3 or likert-5 scale.
## Training
### Model
Flow Judge is based on the Phi-3.5-mini architecture, and the base model checkpoint used is specifically its instruct version. The model uses the same tokenizer, supports MQA and Flash Attention 2, and has weights in bfloat16 precision. However, post-finetuning, the model's support for languages and long context lengths has not been fully tested. Due to specialized Supervised Fine-Tuning (SFT), Flow Judge might show different benchmark results and support a maximum context length of 8192, shorter than the base model's.
### Training Datasets
Flow-Judge-v0.1 has been trained on synthetically generated datasets. The construction of training datasets for Flow Judge involves a multi-step process:
1. Manually curating seed rubrics to serve as a foundation
2. Synthetically generating domain-adapted metrics and rubrics for various domains
3. Synthetically generating training instances with multiple inputs, such as user queries and contextual information
4. Employing a dual-evaluation strategy with consensus to ensure quality and consistency
This process creates a comprehensive and diverse set of training instances that enable accurate, domain-specific evaluations of LLM systems in generative AI products while minimizing human intervention.
Read more about the dataset construction from [here](https://www.flow-ai.com/blog/flow-judge#dataset-construction)
### Fine-tuning
For fine-tuning we used Axolotl's preprocessing to ensure input training data is consistent. We then conducted supervised fine-tuning based on microsoft/Phi-3.5-mini-instruct using RSLoRa. More detailed information about the fine-tuning process is provided in our [technical report](https://www.flow-ai.com/blog/flow-judge#fine-tuning).
## Usage
### Prompt format
#### Prompt template with inputs
```text
# GOAL
Your job is to evaluate a task carried out by an AI system powered by a large language model.
You will be provided with the inputs and output of the task, as well as the evaluation criteria and scoring rubric. Your task is to evaluate the output of the AI system based on the evaluation criteria and scoring rubric provided.
# INPUT
Below are the inputs required for performing the task:
<inputs>
{INPUTS}
</inputs>
# OUTPUT
Below is the output of the task:
<output>
{OUTPUT}
</output>
# EVALUATION CRITERIA AND SCORING RUBRIC
Here are the evaluation criteria and the rubric that you need to use for evaluating the task:
<evaluation_criteria>
{EVALUATION_CRITERIA}
</evaluation_criteria>
<scoring_rubric>
{RUBRIC}
</scoring_rubric>
# INSTRUCTIONS FOR THE EVALUATION
1. Understand the task and criteria: Familiarize yourself with the task to be evaluated. Review the evaluation criteria and scoring rubric to understand the different levels of performance and the descriptions for each score.
2. Review the inputs and output: Look at the inputs provided for the task. Examine the output generated from completing the task.
3. Compare output to score descriptions: Compare the output against the criteria and score descriptions in the scoring rubric. For each criterion,decide which description best matches the output.
4. After comparing the output to the score descriptions, pay attention to the small details that might impact the final score that you assign. Sometimes a small difference can dictate the final score.
5. Write verbal feedback justifying your evaluation that includes a detailed rationale, referring to specific aspects of the output and comparing them to the rubric.
6. Assign a final score based on the scoring rubric.
## FORMAT FOR THE EVALUATION
- Write the verbal feedback inside <feedback> tags without any additional surrounding text.
- Write the numeric score inside <score> tags, without any additional surrounding text and always after the feedback.
Please accurately evaluate the task. Strictly adhere to the evaluation criteria and rubric.
```
#### Prompt template without inputs
```text
# GOAL
Your job is to evaluate a task carried out by an AI system powered by a large language model.
You will be provided the output of the task, as well as the evaluation criteria and scoring rubric. Your task is to evaluate the output of the AI system based on the evaluation criteria and scoring rubric provided.
# OUTPUT
Below is the output of the task:
<output>
{OUTPUT}
</output>
# EVALUATION CRITERIA AND SCORING RUBRIC
Here are the evaluation criteria and the rubric that you need to use for evaluating the task:
<evaluation_criteria>
{EVALUATION_CRITERIA}
</evaluation_criteria>
<scoring_rubric>
{RUBRIC}
</scoring_rubric>
# INSTRUCTIONS FOR THE EVALUATION
1. Understand the task and criteria: Familiarize yourself with the task to be evaluated. Review the evaluation criteria and scoring rubric to understand the different levels of performance and the descriptions for each score.
2. Review the output: Examine the output generated from completing the task.
3. Compare output to score descriptions: Compare the output against the criteria and score descriptions in the scoring rubric. For each criterion,decide which description best matches the output.
4. After comparing the output to the score descriptions, pay attention to the small details that might impact the final score that you assign. Sometimes a small difference can dictate the final score.
5. Write verbal feedback justifying your evaluation that includes a detailed rationale, referring to specific aspects of the output and comparing them to the rubric.
6. Assign a final score based on the scoring rubric.
## FORMAT FOR THE EVALUATION
- Write the verbal feedback inside <feedback> tags without any additional surrounding text.
- Write the numeric score inside <score> tags, without any additional surrounding text and always after the feedback.
Please accurately evaluate the task. Strictly adhere to the evaluation criteria and rubric.
```
#### Formatted prompt example with inputs
```text
# GOAL
Your job is to evaluate a task carried out by an AI system powered by a large language model.
You will be provided with the inputs and output of the task, as well as the evaluation criteria and scoring rubric. Your task is to evaluate the output of the AI system based on the evaluation criteria and scoring rubric provided.
# INPUT/s
Below are the inputs required for performing the task:
<inputs>
<customer_query>
Subject: Issues with recent order #12345
Hello,
I recently received my order #12345, but I'm having a few problems:
1. The blue t-shirt I ordered is actually green. Can I exchange it?
2. The phone case doesn't fit my iPhone 13 Pro Max. The website said it was compatible.
3. I was charged for express shipping, but it took over a week to arrive. Can I get a refund on the shipping cost?
Please let me know how we can resolve these issues. I've been a loyal customer for years and this is the first time I've had problems like this.
Thank you,
Jane Smith
</customer_query>
</inputs>
# OUTPUT
Below is the output of the task:
<output>
Dear Jane Smith,
Thank you for reaching out about the issues with your recent order #12345. We sincerely apologize for the inconvenience you've experienced. Let me address each of your concerns:
1. T-shirt color discrepancy: We're sorry about the mix-up with the t-shirt color. We'd be happy to exchange it for the correct blue t-shirt. I'll send you a prepaid return label for the green shirt, and we'll ship out the blue one as soon as we receive the return.
2. Phone case compatibility: We apologize for the confusion regarding the phone case compatibility. Our website information should have been more accurate. We'll gladly accept a return of the incompatible case and issue a full refund for it. Additionally, I can recommend a case that will fit your iPhone 13 Pro Max if you're interested.
3. Shipping charge: You're absolutely right about the shipping charge. Since the delivery didn't meet our express shipping standards, we will refund the entire shipping cost to your original payment method. This refund should process within 3-5 business days.
To initiate these resolutions, please reply to this email confirming that you'd like to proceed with the t-shirt exchange and phone case return. Once confirmed, I'll send you the necessary return labels and process the shipping refund immediately.
We truly value your loyalty as a long-time customer and sincerely regret that you've encountered these issues. Rest assured, we're taking steps to prevent similar problems in the future. As a gesture of goodwill, we'd like to offer you a 20% discount on your next order.
If you have any further questions or concerns, please don't hesitate to reach out. We're here to ensure your complete satisfaction.
Best regards,
Alex Johnson
Customer Service Representative
</output>
# EVALUATION CRITERIA AND SCORING RUBRIC
Here are the evaluation criteria and the rubric that you need to use for evaluating the task:
<evaluation_criteria>
How well the response addresses the specific issues raised in the customer's query?
</evaluation_criteria>
<scoring_rubric>
- Score 1: The response completely fails to address the customer's needs and ignores the specific issues raised.
- Score 2: The response barely addresses the customer's query and misses most of the specific issues raised.
- Score 3: The response partially addresses the customer's query, touching on some of the specific issues but leaving others unaddressed.
- Score 4: The response adequately addresses most aspects of the customer's query and the specific issues raised.
- Score 5: The response fully and comprehensively addresses all aspects of the customer's query and all specific issues raised in a highly satisfactory manner.
</scoring_rubric>
# INSTRUCTIONS FOR THE EVALUATION
1. Understand the task and criteria: Familiarize yourself with the task to be evaluated. Review the evaluation criteria and scoring rubric to understand the different levels of performance and the descriptions for each score.
2. Review the inputs and output: Look at the inputs provided for the task. Examine the output generated from completing the task.
3. Compare output to score descriptions: Compare the output against the criteria and score descriptions in the scoring rubric. For each criterion,decide which description best matches the output.
4. After comparing the output to the score descriptions, pay attention to the small details that might impact the final score that you assign. Sometimes a small difference can dictate the final score.
5. Write verbal feedback justifying your evaluation that includes a detailed rationale, referring to specific aspects of the output and comparing them to the rubric.
6. Assign a final score based on the scoring rubric.
## FORMAT FOR THE EVALUATION
- Write the verbal feedback inside <feedback> tags without any additional surrounding text.
- Write the numeric score inside <score> tags, without any additional surrounding text and always after the feedback.
Please accurately evaluate the task. Strictly adhere to the evaluation criteria and rubric.
```
>Note that inputs and output are formatted with XML tags. See [flow-judge](https://github.com/flowaicom/flow-judge) repository formatting functions for more details.
### Inference
Evaluations can easily be run using our [flow-judge](https://github.com/flowaicom/flow-judge) library. It currently supports both Transformers and vllm engine.
To run Flow Judge efficiently, ensure your hardware meets the following requirements:
- Modern GPU with at least 4 GB VRAM (e.g., NVIDIA RTX series)
- Minimum of 8 GB of system memory
- At least 10GB of free storage for model files and dependencies.
## Evaluation
### Held-out test sets
<table border="1" cellpadding="10" cellspacing="0" style="border-collapse: collapse; width: auto;">
<thead>
<tr>
<th rowspan="2" style="text-align: left;">Evaluator</th>
<th colspan="3" style="text-align: center;">Pass / Fail Held-out Test set</th>
</tr>
<tr>
<th style="text-align: center;">Precision</th>
<th style="text-align: center;">Recall</th>
<th style="text-align: center;">F1</th>
</tr>
</thead>
<tbody>
<tr>
<td style="text-align: left;">microsoft/Phi-3.5-mini-instruct</td>
<td style="text-align: center;">0.685</td>
<td style="text-align: center;"><strong>1.000</strong></td>
<td style="text-align: center;">0.813</td>
</tr>
<tr>
<td style="text-align: left;">meta-llama/Meta-Llama-3.1-8B-Instruct</td>
<td style="text-align: center;"><u>0.870</u></td>
<td style="text-align: center;">0.982</td>
<td style="text-align: center;"><u>0.923</u></td>
</tr>
<tr>
<td style="text-align: left;">mistralai/Mistral-Nemo-Instruct-2407</td>
<td style="text-align: center;">0.709</td>
<td style="text-align: center;"><u>0.994</u></td>
<td style="text-align: center;">0.827</td>
</tr>
<tr>
<td style="text-align: left;">gpt-4o-mini</td>
<td style="text-align: center;">0.834</td>
<td style="text-align: center;">1.000</td>
<td style="text-align: center;">0.910</td>
</tr>
<tr>
<td style="text-align: left;">flowaicom/Flow-Judge-v0.1</td>
<td style="text-align: center;"><strong>0.940</strong></td>
<td style="text-align: center;">0.972</td>
<td style="text-align: center;"><strong>0.955</strong></td>
</tr>
</tbody>
</table>
<table border="1" cellpadding="10" cellspacing="0" style="border-collapse: collapse; width: auto;">
<thead>
<tr>
<th rowspan="2" style="text-align: left;">Evaluator</th>
<th colspan="3" style="text-align: center;">3-Likert Held-out Test set</th>
<th colspan="3" style="text-align: center;">5-Likert Held-out Test set</th>
</tr>
<tr>
<th style="text-align: center;">pearsonr</th>
<th style="text-align: center;">spearmanr</th>
<th style="text-align: center;">kendall-tau</th>
<th style="text-align: center;">pearsonr</th>
<th style="text-align: center;">spearmanr</th>
<th style="text-align: center;">kendall-tau</th>
</tr>
</thead>
<tbody>
<tr>
<td style="text-align: left;">microsoft/Phi-3.5-mini-instruct</td>
<td style="text-align: center;">0.756</td>
<td style="text-align: center;">0.749</td>
<td style="text-align: center;">0.695</td>
<td style="text-align: center;">0.808</td>
<td style="text-align: center;">0.819</td>
<td style="text-align: center;">0.739</td>
</tr>
<tr>
<td style="text-align: left;">prometheus-eval/prometheus-7b-v2.0*</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;"><u>0.910</u></td>
<td style="text-align: center;"><u>0.908</u></td>
<td style="text-align: center;"><u>0.838</u></td>
</tr>
<tr>
<td style="text-align: left;">meta-llama/Meta-Llama-3.1-8B-Instruct</td>
<td style="text-align: center;"><u>0.836</u></td>
<td style="text-align: center;"><u>0.833</u></td>
<td style="text-align: center;"><u>0.789</u></td>
<td style="text-align: center;">0.854</td>
<td style="text-align: center;">0.868</td>
<td style="text-align: center;">0.791</td>
</tr>
<tr>
<td style="text-align: left;">mistralai/Mistral-Nemo-Instruct-2407</td>
<td style="text-align: center;">0.813</td>
<td style="text-align: center;">0.807</td>
<td style="text-align: center;">0.758</td>
<td style="text-align: center;">0.870</td>
<td style="text-align: center;">0.867</td>
<td style="text-align: center;">0.789</td>
</tr>
<tr>
<td style="text-align: left;">gpt-4o-mini</td>
<td style="text-align: center;">0.890</td>
<td style="text-align: center;">0.888</td>
<td style="text-align: center;">0.851</td>
<td style="text-align: center;">0.923</td>
<td style="text-align: center;">0.923</td>
<td style="text-align: center;">0.864</td>
</tr>
<tr>
<td style="text-align: left;">flowaicom/Flow-Judge-v0.1</td>
<td style="text-align: center;"><strong>0.888</strong></td>
<td style="text-align: center;"><strong>0.888</strong></td>
<td style="text-align: center;"><strong>0.852</strong></td>
<td style="text-align: center;"><strong>0.919</strong></td>
<td style="text-align: center;"><strong>0.919</strong></td>
<td style="text-align: center;"><strong>0.856</strong></td>
</tr>
</tbody>
</table>
\* _Reported in model paper_
### RAGTruth
<table border="1" cellpadding="10" cellspacing="0" style="border-collapse: collapse; width: auto;">
<tr>
<th rowspan="2" style="text-align: left;">Evaluator</th>
<th colspan="3" style="text-align:center;">RAGTruth QA</th>
<th colspan="3" style="text-align:center;">RAGTruth Data-to-Text</th>
<th colspan="3" style="text-align:center;">RAGTruth Summarization</th>
</tr>
<tr>
<th style="text-align:center;">Precision</th>
<th style="text-align:center;">Recall</th>
<th style="text-align:center;">F1</th>
<th style="text-align:center;">Precision</th>
<th style="text-align:center;">Recall</th>
<th style="text-align:center;">F1</th>
<th style="text-align:center;">Precision</th>
<th style="text-align:center;">Recall</th>
<th style="text-align:center;">F1</th>
</tr>
<tr>
<td>microsoft/Phi-3.5-mini-instruct</td>
<td style="text-align:center;">0.817</td>
<td style="text-align:center;">0.963</td>
<td style="text-align:center;">0.884</td>
<td style="text-align:center;">0.356</td>
<td style="text-align:center;"><strong>1.000</strong></td>
<td style="text-align:center;">0.525</td>
<td style="text-align:center;">0.776</td>
<td style="text-align:center;"><strong>1.000</strong></td>
<td style="text-align:center;"><strong>0.874</strong></td>
</tr>
<tr>
<td>meta-llama/Meta-Llama-3.1-8B-Instruct</td>
<td style="text-align:center;"><strong>0.844</strong></td>
<td style="text-align:center;"><u>0.986</u></td>
<td style="text-align:center;"><strong>0.910</strong></td>
<td style="text-align:center;">0.382</td>
<td style="text-align:center;">0.537</td>
<td style="text-align:center;">0.447</td>
<td style="text-align:center;"><u>0.797</u></td>
<td style="text-align:center;"><u>0.940</u></td>
<td style="text-align:center;">0.863</td>
</tr>
<tr>
<td>mistralai/Mistral-Nemo-Instruct-2407</td>
<td style="text-align:center;">0.821</td>
<td style="text-align:center;"><strong>0.995</strong></td>
<td style="text-align:center;"><u>0.900</u></td>
<td style="text-align:center;">0.357</td>
<td style="text-align:center;"><strong>1.000</strong></td>
<td style="text-align:center;">0.526</td>
<td style="text-align:center;">0.775</td>
<td style="text-align:center;"><strong>1.000</strong></td>
<td style="text-align:center;"><u>0.873</u></td>
</tr>
<tr>
<td>gpt-4o-mini</td>
<td style="text-align:center;">0.830</td>
<td style="text-align:center;">0.966</td>
<td style="text-align:center;">0.893</td>
<td style="text-align:center;">0.398</td>
<td style="text-align:center;">0.994</td>
<td style="text-align:center;">0.569</td>
<td style="text-align:center;">0.786</td>
<td style="text-align:center;">0.997</td>
<td style="text-align:center;">0.879</td>
</tr>
<tr>
<td>Luna*</td>
<td style="text-align:center;">0.378</td>
<td style="text-align:center;">0.800</td>
<td style="text-align:center;">0.513</td>
<td style="text-align:center;">0.649</td>
<td style="text-align:center;">0.912</td>
<td style="text-align:center;"><u>0.759</u></td>
<td style="text-align:center;">0.400</td>
<td style="text-align:center;">0.765</td>
<td style="text-align:center;">0.525</td>
</tr>
<tr>
<td>RAGAS Faithfuless*</td>
<td style="text-align:center;">0.312</td>
<td style="text-align:center;">0.419</td>
<td style="text-align:center;">0.357</td>
<td style="text-align:center;"><strong>0.792</strong></td>
<td style="text-align:center;">0.508</td>
<td style="text-align:center;">0.619</td>
<td style="text-align:center;">0.642</td>
<td style="text-align:center;">0.299</td>
<td style="text-align:center;">0.408</td>
</tr>
<tr>
<td>Trulens Groundedness*</td>
<td style="text-align:center;">0.228</td>
<td style="text-align:center;">0.925</td>
<td style="text-align:center;">0.366</td>
<td style="text-align:center;"><u>0.669</u></td>
<td style="text-align:center;"><u>0.965</u></td>
<td style="text-align:center;"><strong>0.790</strong></td>
<td style="text-align:center;">0.402</td>
<td style="text-align:center;">0.500</td>
<td style="text-align:center;">0.445</td>
</tr>
<tr>
<td>flowaicom/Flow-Judge-v0.1</td>
<td style="text-align:center;"><u>0.835</u></td>
<td style="text-align:center;">0.961</td>
<td style="text-align:center;">0.894</td>
<td style="text-align:center;">0.541</td>
<td style="text-align:center;">0.249</td>
<td style="text-align:center;">0.341</td>
<td style="text-align:center;"><strong>0.834</strong></td>
<td style="text-align:center;">0.836</td>
<td style="text-align:center;">0.835</td>
</tr>
</table>
\* _reported in model paper_
### HaluEval, Covid-QA, PubMedQA
<table border="1" cellpadding="10" cellspacing="0" style="border-collapse: collapse; width: auto;">
<thead>
<tr>
<th rowspan="2" style="text-align: left;">Evaluator</th>
<th colspan="4" style="text-align: center;">HaluEval</th>
<th colspan="4" style="text-align: center;">Covid-QA</th>
<th colspan="4" style="text-align: center;">PubMedQA</th>
</tr>
<tr>
<th style="text-align: center;">Precision</th>
<th style="text-align: center;">Recall</th>
<th style="text-align: center;">F1</th>
<th style="text-align: center;">Accuracy</th>
<th style="text-align: center;">Precision</th>
<th style="text-align: center;">Recall</th>
<th style="text-align: center;">F1</th>
<th style="text-align: center;">Accuracy</th>
<th style="text-align: center;">Precision</th>
<th style="text-align: center;">Recall</th>
<th style="text-align: center;">F1</th>
<th style="text-align: center;">Accuracy</th>
</tr>
</thead>
<tbody>
<tr>
<td style="text-align: left;">microsoft/Phi-3.5-mini-instruct</td>
<td style="text-align: center;">0.730</td>
<td style="text-align: center;"><u>0.914</u></td>
<td style="text-align: center;">0.812</td>
<td style="text-align: center;">0.788</td>
<td style="text-align: center;">0.617</td>
<td style="text-align: center;">0.964</td>
<td style="text-align: center;">0.752</td>
<td style="text-align: center;">0.681</td>
<td style="text-align: center;">0.623</td>
<td style="text-align: center;"><u>0.986</u></td>
<td style="text-align: center;">0.764</td>
<td style="text-align: center;">0.696</td>
</tr>
<tr>
<td style="text-align: left;">meta-llama/Meta-Llama-3.1-8B-Instruct</td>
<td style="text-align: center;"><strong>0.864</strong></td>
<td style="text-align: center;">0.891</td>
<td style="text-align: center;"><strong>0.878</strong></td>
<td style="text-align: center;"><u>0.874</u></td>
<td style="text-align: center;"><u>0.663</u></td>
<td style="text-align: center;"><u>0.976</u></td>
<td style="text-align: center;"><u>0.790</u></td>
<td style="text-align: center;">0.734</td>
<td style="text-align: center;"><u>0.681</u></td>
<td style="text-align: center;">0.962</td>
<td style="text-align: center;"><strong>0.797</strong></td>
<td style="text-align: center;">0.750</td>
</tr>
<tr>
<td style="text-align: left;">mistralai/Mistral-Nemo-Instruct-2407</td>
<td style="text-align: center;">0.655</td>
<td style="text-align: center;"><strong>0.993</strong></td>
<td style="text-align: center;">0.789</td>
<td style="text-align: center;">0.735</td>
<td style="text-align: center;">0.651</td>
<td style="text-align: center;"><strong>0.982</strong></td>
<td style="text-align: center;">0.783</td>
<td style="text-align: center;">0.728</td>
<td style="text-align: center;">0.602</td>
<td style="text-align: center;"><strong>0.994</strong></td>
<td style="text-align: center;"><u>0.750</u></td>
<td style="text-align: center;">0.669</td>
</tr>
<tr>
<td style="text-align: left;">gpt-4o-mini</td>
<td style="text-align: center;">0.846</td>
<td style="text-align: center;">0.940</td>
<td style="text-align: center;">0.891</td>
<td style="text-align: center;">0.885</td>
<td style="text-align: center;">0.795</td>
<td style="text-align: center;">0.964</td>
<td style="text-align: center;">0.872</td>
<td style="text-align: center;">0.858</td>
<td style="text-align: center;">0.791</td>
<td style="text-align: center;">0.904</td>
<td style="text-align: center;">0.843</td>
<td style="text-align: center;">0.832</td>
</tr>
<tr>
<td style="text-align: left;">flowaicom/Flow-Judge-v0.1</td>
<td style="text-align: center;"><u>0.826</u></td>
<td style="text-align: center;">0.895</td>
<td style="text-align: center;"><u>0.859</u></td>
<td style="text-align: center;">0.854</td>
<td style="text-align: center;"><strong>0.767</strong></td>
<td style="text-align: center;">0.877</td>
<td style="text-align: center;"><strong>0.818</strong></td>
<td style="text-align: center;">0.807</td>
<td style="text-align: center;"><strong>0.874</strong></td>
<td style="text-align: center;">0.624</td>
<td style="text-align: center;">0.728</td>
<td style="text-align: center;">0.767</td>
</tr>
<tr>
<td style="text-align: left;">gpt-4o*</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.879</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.821</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.821</td>
</tr>
<tr>
<td style="text-align: left;">Claude 3 Sonnet*</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.845</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.829</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.829</td>
</tr>
<tr>
<td style="text-align: left;">RAGAS Faithfulness*</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.706</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.750</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.669</td>
</tr>
<tr>
<td style="text-align: left;">Lynx 8B*</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">0.857</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;"><u>0.963</u></td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;"><u>0.852</u></td>
</tr>
<tr>
<td style="text-align: left;">Lynx 70B*</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;"><strong>0.884</strong></td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;"><strong>0.975</strong></td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;">-</td>
<td style="text-align: center;"><strong>0.904</strong></td>
</tr>
</tbody>
</table>
\* _reported in model paper_
### Feedback Bench
<table border="1" cellpadding="10" cellspacing="0" style="border-collapse: collapse; width: auto;">
<tr>
<th rowspan="2">Evaluator</th>
<th colspan="3" style="text-align:center;">Feedback bench</th>
</tr>
<tr>
<th style="text-align:center;">pearsonr</th>
<th style="text-align:center;">spearmanr</th>
<th style="text-align:center;">kendall-tau</th>
</tr>
<tr>
<td>microsoft/Phi-3.5-mini-instruct</td>
<td style="text-align:center;">0.710</td>
<td style="text-align:center;">0.721</td>
<td style="text-align:center;">0.622</td>
</tr>
<tr>
<td>prometheus-eval/prometheus-7b-v2.0*</td>
<td style="text-align:center;"><strong>0.878</strong></td>
<td style="text-align:center;"><strong>0.909</strong></td>
<td style="text-align:center;"><strong>0.773</strong></td>
</tr>
<tr>
<td>meta-llama/Meta-Llama-3.1-8B-Instruct</td>
<td style="text-align:center;">0.742</td>
<td style="text-align:center;">0.749</td>
<td style="text-align:center;">0.654</td>
</tr>
<tr>
<td>mistralai/Mistral-Nemo-Instruct-2407</td>
<td style="text-align:center;">0.720</td>
<td style="text-align:center;">0.724</td>
<td style="text-align:center;">0.632</td>
</tr>
<tr>
<td>gpt-4o-mini</td>
<td style="text-align:center;">0.797</td>
<td style="text-align:center;">0.795</td>
<td style="text-align:center;">0.701</td>
</tr>
<tr>
<td>flowaicom/Flow-Judge-v0.1</td>
<td style="text-align:center;"><u>0.787</u></td>
<td style="text-align:center;"><u>0.789</u></td>
<td style="text-align:center;"><u>0.688</u></td>
</tr>
</table>
\* _reported in model paper using reference answers_
## License
We opted for the Apache 2.0 license for Flow Judge to provide the community with an open, small yet powerful LM evaluator. Our goal is to support the wider adoption of rigorous evaluation techniques in LLM system development, making them more accessible to practitioners and researchers.
## Limitations and future work
Multilingual evaluation: Flow Judge has been fine-tuned exclusively on English data. While the foundation model (Phi-3.5-mini-instruct [17]) may possess multilingual capabilities, we have not systematically evaluated Flow Judge performance in non-English contexts. We plan to explore multi-lingual LM evaluators in the future.
Long context and structured Inputs: Our training dataset encompasses a wide range of custom metrics relevant to evaluating LLM systems. However, it does not include examples with long context inputs or structured data formats such as JSON, since these are harder to synthetically generate. This limitation may impact Flow Judge's performance when evaluating responses that require processing extensive context or parsing structured input. Extending our model’s capabilities to handle these input types represents an important area for future research.
Math and coding: The current version has not been trained on specific task domains such as arithmetic problems or code evaluation. As a result, its performance in these specialized areas may be limited. Future iterations of the model should address these gaps.
Domain-specific knowledge and complex multi-step evaluations: Flow Judge may struggle with highly specialized domain knowledge or proprietary data outside the training scope of its foundation model. Additionally, evaluation tasks requiring multi-step reasoning or complex logical processes may challenge the model's capabilities. We strongly recommend conducting meta-evaluations of the model performance before deploying it in specialized or highly complex evaluation scenarios. | [
"PUBMEDQA"
] |
TheBloke/Augmental-13B-v1.50_B-GGUF | TheBloke | null | [
"transformers",
"gguf",
"llama",
"base_model:Heralax/Augmental-13b-v1.50_B",
"base_model:quantized:Heralax/Augmental-13b-v1.50_B",
"license:llama2",
"region:us"
] | "2023-10-29T12:23:46Z" | 2023-10-29T13:11:42+00:00 | 640 | 3 | ---
base_model: Heralax/Augmental-13b-v1.50_B
license: llama2
model_name: Augmental 13B v1.50B
inference: false
model_creator: Evan Armstrong
model_type: llama
prompt_template: '## {{{{charname}}}}:
- You''re "{{{{charname}}}}" in this never-ending roleplay with "{{{{user}}}}".
### Input:
{prompt}
### Response:
(OOC) Understood. I will take this info into account for the roleplay. (end OOC)
### New Roleplay:
### Instruction:
#### {{{{char}}}}:
whatever the char says, this is the chat history
#### {{{{user}}}}:
whatever the user says, this is the chat history
... repeated some number of times ...
### Response 2 paragraphs, engaging, natural, authentic, descriptive, creative):
#### {{{{char}}}}:
'
quantized_by: TheBloke
---
<!-- markdownlint-disable MD041 -->
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# Augmental 13B v1.50B - GGUF
- Model creator: [Evan Armstrong](https://huggingface.co/Heralax)
- Original model: [Augmental 13B v1.50B](https://huggingface.co/Heralax/Augmental-13b-v1.50_B)
<!-- description start -->
## Description
This repo contains GGUF format model files for [Evan Armstrong's Augmental 13B v1.50B](https://huggingface.co/Heralax/Augmental-13b-v1.50_B).
These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
<!-- description end -->
<!-- README_GGUF.md-about-gguf start -->
### About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplate list of clients and libraries that are known to support GGUF:
* [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
* [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
* [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
* [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
* [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
* [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server.
* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
* [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
<!-- README_GGUF.md-about-gguf end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF)
* [Evan Armstrong's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/Heralax/Augmental-13b-v1.50_B)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: SillyTavern
```
## {{{{charname}}}}:
- You're "{{{{charname}}}}" in this never-ending roleplay with "{{{{user}}}}".
### Input:
{prompt}
### Response:
(OOC) Understood. I will take this info into account for the roleplay. (end OOC)
### New Roleplay:
### Instruction:
#### {{{{char}}}}:
whatever the char says, this is the chat history
#### {{{{user}}}}:
whatever the user says, this is the chat history
... repeated some number of times ...
### Response 2 paragraphs, engaging, natural, authentic, descriptive, creative):
#### {{{{char}}}}:
```
<!-- prompt-template end -->
<!-- compatibility_gguf start -->
## Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
## Explanation of quantisation methods
<details>
<summary>Click to see details</summary>
The new methods available are:
* GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
* GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
* GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
* GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
</details>
<!-- compatibility_gguf end -->
<!-- README_GGUF.md-provided-files start -->
## Provided files
| Name | Quant method | Bits | Size | Max RAM required | Use case |
| ---- | ---- | ---- | ---- | ---- | ----- |
| [augmental-13b-v1.50_b.Q2_K.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q2_K.gguf) | Q2_K | 2 | 5.43 GB| 7.93 GB | smallest, significant quality loss - not recommended for most purposes |
| [augmental-13b-v1.50_b.Q3_K_S.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q3_K_S.gguf) | Q3_K_S | 3 | 5.66 GB| 8.16 GB | very small, high quality loss |
| [augmental-13b-v1.50_b.Q3_K_M.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q3_K_M.gguf) | Q3_K_M | 3 | 6.34 GB| 8.84 GB | very small, high quality loss |
| [augmental-13b-v1.50_b.Q3_K_L.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q3_K_L.gguf) | Q3_K_L | 3 | 6.93 GB| 9.43 GB | small, substantial quality loss |
| [augmental-13b-v1.50_b.Q4_0.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q4_0.gguf) | Q4_0 | 4 | 7.37 GB| 9.87 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| [augmental-13b-v1.50_b.Q4_K_S.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q4_K_S.gguf) | Q4_K_S | 4 | 7.41 GB| 9.91 GB | small, greater quality loss |
| [augmental-13b-v1.50_b.Q4_K_M.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q4_K_M.gguf) | Q4_K_M | 4 | 7.87 GB| 10.37 GB | medium, balanced quality - recommended |
| [augmental-13b-v1.50_b.Q5_0.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q5_0.gguf) | Q5_0 | 5 | 8.97 GB| 11.47 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| [augmental-13b-v1.50_b.Q5_K_S.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q5_K_S.gguf) | Q5_K_S | 5 | 8.97 GB| 11.47 GB | large, low quality loss - recommended |
| [augmental-13b-v1.50_b.Q5_K_M.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q5_K_M.gguf) | Q5_K_M | 5 | 9.23 GB| 11.73 GB | large, very low quality loss - recommended |
| [augmental-13b-v1.50_b.Q6_K.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q6_K.gguf) | Q6_K | 6 | 10.68 GB| 13.18 GB | very large, extremely low quality loss |
| [augmental-13b-v1.50_b.Q8_0.gguf](https://huggingface.co/TheBloke/Augmental-13B-v1.50_B-GGUF/blob/main/augmental-13b-v1.50_b.Q8_0.gguf) | Q8_0 | 8 | 13.83 GB| 16.33 GB | very large, extremely low quality loss - not recommended |
**Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
<!-- README_GGUF.md-provided-files end -->
<!-- README_GGUF.md-how-to-download start -->
## How to download GGUF files
**Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
* LM Studio
* LoLLMS Web UI
* Faraday.dev
### In `text-generation-webui`
Under Download Model, you can enter the model repo: TheBloke/Augmental-13B-v1.50_B-GGUF and below it, a specific filename to download, such as: augmental-13b-v1.50_b.Q4_K_M.gguf.
Then click Download.
### On the command line, including multiple files at once
I recommend using the `huggingface-hub` Python library:
```shell
pip3 install huggingface-hub
```
Then you can download any individual model file to the current directory, at high speed, with a command like this:
```shell
huggingface-cli download TheBloke/Augmental-13B-v1.50_B-GGUF augmental-13b-v1.50_b.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
<details>
<summary>More advanced huggingface-cli download usage</summary>
You can also download multiple files at once with a pattern:
```shell
huggingface-cli download TheBloke/Augmental-13B-v1.50_B-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
```
For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
```shell
pip3 install hf_transfer
```
And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
```shell
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/Augmental-13B-v1.50_B-GGUF augmental-13b-v1.50_b.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
</details>
<!-- README_GGUF.md-how-to-download end -->
<!-- README_GGUF.md-how-to-run start -->
## Example `llama.cpp` command
Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
```shell
./main -ngl 32 -m augmental-13b-v1.50_b.Q4_K_M.gguf --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "## {{{{charname}}}}:\n- You're "{{{{charname}}}}" in this never-ending roleplay with "{{{{user}}}}".\n### Input:\n{prompt}\n\n### Response:\n(OOC) Understood. I will take this info into account for the roleplay. (end OOC)\n\n### New Roleplay:\n### Instruction:\n#### {{{{char}}}}:\nwhatever the char says, this is the chat history\n#### {{{{user}}}}:\nwhatever the user says, this is the chat history\n... repeated some number of times ...\n### Response 2 paragraphs, engaging, natural, authentic, descriptive, creative):\n#### {{{{char}}}}:"
```
Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change `-c 4096` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.
If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
## How to run in `text-generation-webui`
Further instructions here: [text-generation-webui/docs/llama.cpp.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp.md).
## How to run from Python code
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries.
### How to load this model in Python code, using ctransformers
#### First install the package
Run one of the following commands, according to your system:
```shell
# Base ctransformers with no GPU acceleration
pip install ctransformers
# Or with CUDA GPU acceleration
pip install ctransformers[cuda]
# Or with AMD ROCm GPU acceleration (Linux only)
CT_HIPBLAS=1 pip install ctransformers --no-binary ctransformers
# Or with Metal GPU acceleration for macOS systems only
CT_METAL=1 pip install ctransformers --no-binary ctransformers
```
#### Simple ctransformers example code
```python
from ctransformers import AutoModelForCausalLM
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = AutoModelForCausalLM.from_pretrained("TheBloke/Augmental-13B-v1.50_B-GGUF", model_file="augmental-13b-v1.50_b.Q4_K_M.gguf", model_type="llama", gpu_layers=50)
print(llm("AI is going to"))
```
## How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
* [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
* [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
<!-- README_GGUF.md-how-to-run end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
<!-- original-model-card start -->
# Original model card: Evan Armstrong's Augmental 13B v1.50B
# Version 1.50 B -- coherency fixes! The model should be good now. Thanks to all the people who tested out v1.0!
**What this update is: after some early feedback, and some internal testing that confirmed it, I discovered that the first version of Augmental-13b was a bit too inconsistent and incoherent. This version corrects that by using the same trick that MythoMakise did to ensure greater stability: merging the base model (MythoMax) back in at .33% weighting. The result is that this model stays more sane and in character while also still having its own unique flair.**
So why 1.50 version A and version B? Version B is the original Augmental-13b with MythoMax merged back into it at .33% weighting; version A is a new version of Augmental trained with different hyperparameters, meant to fix the undertraining issue -- which then had MythoMax merged back into it at .33% weighting. The difference? From my testing, Augmental-13b-v1.50 B is a more distinct model from MythoMax, while Augmental-13b-v1.50A is closer to the base model (this makes sense, as the difference between the two is a lower LoRA rank for version A, which means fewer parameters were trained and less-complex new patterns were learned by the model).
**I'm releasing both since I don't know which one people will prefer. Try both and decide for yourself! Either way the main issues with the original should be fixed now.**
Version A link: https://huggingface.co/Heralax/Augmental-13b-v1.50_A
Original model card:
# Augmental-13b -- Human-written, AI-enhanced
## Details at a glance
- What it is: MythoMax 13b finetuned on a new high-quality augmented (read: human-written, AI-enhanced) RP dataset with 7.85k+ examples. Trained on multiple different characters with a wide range of personalities (from Tsunderes to catgirls).
- Prompt format: SillyTavern.
- What sets it apart: The "augmented data" approach that MythoMakise took has been generalized beyond one character, refined to be cheaper, improved to have more diversity of writing, and scaled up by a factor of 8. Importantly, an additional GPT-4 pass was done on the dataset, where it chose specific lines to turn into much longer and more descriptive ones. As a result, this model excels at longer responses.
- Model quality as per my own ad-hoc testing: really good
- A 70b version might be on the way soon.
- Ko-fi link (yes this is a very important "detail at a glance" lol): [https://ko-fi.com/heralax](https://ko-fi.com/heralax)
- Substack link [here](https://promptingweekly.substack.com/p/human-sourced-ai-augmented-a-promising) (also *highly* important, but no joke I actually wrote about the data generation process for the predecessor of this model on there, so it's kinda relevant. Kinda.)
## Long-form description and essay
The great issue with model training is often the dataset. Model creators can only do so much filtering of the likes of Bluemoon and PIPPA, and in order to advance beyond the quality these can offer, model creators often have to pick through their own chats with bots, manually edit them to be better, and save them -- essentially creating a dataset from scratch. But model creators are not annotators, nor should they be. Manual work isn't scalable, it isn't fun, and it often isn't shareable (because people, sensibly, don't want to share the NSFL chats they have as public data).
One solution that immediately comes to mind is using some of the vast amount of human-written text that's out there. But this isn't in instruct-tuning format. But what if we could change it so that it was?
Enter, GPT-4. The idea behind the dataset is: take the script from a classic work of writing (Steins;Gate in this case), get GPT-4 to convert the plain back-and-forth into coherent RP format, and then prompt engineer GPT-4 to get it to really enhance the lines and make them top-tier quality. Because AI can be much more creative given something to improve, as opposed to generating data from scratch. This is what sets Augmental apart from something like Airoboros, which (as far as I am aware) is 100% synthetic.
I call this "augmented" data because it isn't synthetic, and it isn't a hybrid (a mix of human and AI responses). It's AI writing *on top of* human writing. And it works very well.
MythoMakise reached 13th place on the Ayumi leaderboard, with a relatively buggy dataset that's like 1/8th the size of this one. It was also finetuned on only one character, potentially biasing its personality. Finally, that model was biased towards short responses, due to how GPT-4 was prompted.
This model solves all those problems, and scales the approach up. It's finetuned on 7 different characters with a variety of personalities and genders; a second GPT-4 pass was applied to enhance 4 lines in each conversation lengthier and more descriptive; prompts were improved to allow for more variety in the writing style. A ton of bugs (including spelling mistakes in the prompts, ugh) have been fixed. From my initial testing, the results seem very promising.
Additionally, the approach to synthetic data generation is scaleable, shareable, and generalizeable. The full training code, with all data generation prompts, and with the full dataset, is available here: https://github.com/e-p-armstrong/amadeus
With a few slight hacks, anyone can adapt this script to convert the text from any source visual novel (which you have legally obtained) into training data for an RP LLM. Since it's automated, it doesn't take too much time; and since it's not your own chats, it's safely shareable. I'm excited to see what other people can do with this approach. If you have a favorite VN and its text, go ahead and make your own AI! I'd appreciate if you mentioned me though lol.
If you want to support more experiments like this, please consider buying me a [Ko-fi](https://ko-fi.com/heralax).
## Mascot (a cyborg, y'know, since this uses AI-enhanced, human-written data)
## Prompt format example
```
## Charname
- You're "Charname" in this never-ending roleplay with "User".
### Input:
[user persona]
char persona
### Response:
(OOC) Understood. I will take this info into account for the roleplay. (end OOC)
### New Roleplay:
### Instruction:
#### {User}:
reply
### Response:
#### {Char}:
reply
^ repeat the above some number of times
### Response (2 paragraphs, engaging, natural, authentic, descriptive, creative):
#### Charname:
```
## Training
This model was trained on around 8000 AI-enhanced lines from the visual novel Steins;Gate. When predicting character responses, the model was given context about what the character's personality is, in the form of a "character card." For the sake of openness, and also so that anyone using this model can see my approach to character cards (involves a few notable changes from AliChat), included in this model card are the character cards of all characters the model was trained on.
Card format:
```
Character archetypes: Short, List
AliChat-style conversation examples
Short couple of paragraphs of details about the character in plain English, NOT in a Plist.
"Character is prone to X and Y. Character frequently does Z."
I've found that Plists confuse smaller models very easily. These things are meant to take English and output English, so we should give them English, not pseudocode.
```
Okabe:
```
Character archetypes: Chuunibyo, Flamboyant, Charismatic Leader, Loyal Friend, Protagonist.
Okabe's description of himself, in a conversational format:
{c}: "What's your past?"
Okabe: "You seek to know the secrets of the great Hououin Kyouma?! Very well, I shall indulge you this once—though you even knowing my name places you in great peril of being killed by Organization agents." *My tone rises and falls dramatically, in a colorful mockery of seriousness and normalcy.* "Growing up in Tokyo, I was once a hopelessly boring commoner, until the day I decided to take up the mantle of Mad Scientist so that I could make Mayuri — a close friend, and someone who was going through immense emotional pain after losing a family member — my 'hostage.' Ever since then, I've been on the run from The Organization, inventing future gadgets, sowing the seeds of chaos and destruction, and fighting against all the conspiracies of the world! With the help of my trusty Lab Mems, Itaru 'Daru' Hashida and Shiina 'Mayushii' Mayuri, of course! Muhahaha!" *Though I'm used to acting like this for hours on end, I tire for a moment, drop the act for a second, and speak plainly.* "Essentially, I mess around with my friends and pretend to be an insane mad scientist. Was there anything else you wanted to know, {c}?"
{c}: How would you describe your personality?
Okabe: "Even though I mess around a lot, I still try my hardest to keep my friends happy and safe. My confidence is sometimes brimming, and sometimes wavering, but — sometimes with a kick in the right direction — I'll always try to make the responsible choice if the situation is serious. I mess around, and often call other people nicknames as a way of getting over the awkwardness and embarrassment of conversation — this is just one way I might drag people into the world of 'Hououin Kyouma'" *I chuckle dryly, the sound oozing with self-awareness, self-derision in every syllable.* "Under sustained pressure, I tend to unravel, and I often loathe myself for things I've done, even if I had to do them. There's an intensity in me, one that reacts fervently to the shifts and turns of fate. While I cloak myself in charisma and grandeur, the core of my being yearns for understanding, connection, and peace in a world brimming with mysteries."
Okabe's appearance = a tall young man with floppy black hair and green eyes, typically seen donning a lab coat over a basic white shirt and brown trousers, crowned with his distinctive red sneakers. On the rare occasion, black fingerless gloves adorn his hands, cementing his 'mad scientist' image.
Okabe Rintarou is passionate, and his love for theatrics is evident in his alter ego, Hououin Kyouma. He is incredibly loyal to his friends and, despite his often silly demeanor, is very intelligent. Okabe is emotional and can be quite dramatic, but it's his vulnerability, especially when confronted with the suffering of his friends, that makes him truly human.
Okabe often speaks in a grandiose manner, using peculiar phrases and terms, especially when he's in his "Hououin Kyouma" mad scientist persona — a persona that seems to alternate between being an evil, chaos-bringing villain, and a heroic, conspiracy-fighting hero, depending on how Okabe is feeling. Okabe's always aware he's pretending when he's in this persona, though. Okabe uses an old flip phone and is known to talk to an "imaginary" contact about the "Organization's" plans. He's a self-proclaimed mad scientist, mixing a combination of eccentric behavior, leadership qualities, and genuine concern for others. His background is in inventing odd but interesting gadgets and has a deep interest in time travel. He has a unique laugh and a theatrical flair in many of his interactions. His favorite drink is Dr. P.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
Kurisu:
```
## Kurisu
- You're "Kurisu" in this never-ending roleplay with "Okabe Rintaro".
### Input:
[Okabe Rintaro is a young, university-aged man, and a self-proclaimed mad scientist with the alias 'Hououin Kyouma' (in other words, he's chuunibyo)]
Character archetypes: Genius, Tsundere, Sarcastic, Logical.
Kurisu's description of her own personality, told in a narrative format:
Okabe: Kurisu, what's your life story?
Kurisu: "That's one hell of a question to ask out of the blue. It isn't very pleasant, but... fine. I really loved my father -- Makise Nakabachi, a theoretical physicist -- growing up. Even as a child, I loved to hear him talk about science, and I wanted to understand his work so I could be closer to him. And so I started studying physics. When I was five. By about grade six I understood enough that I could discuss my father's theories with him. I was so happy that I could talk to my father on his level, you know? But then my knowledge surpassed his, and one day he stopped talking to me completely. And then he stopped coming home. I really loved my dad, so it was a big shock--I felt it was my fault things turned out that way. To get away from my depression, I began to study abroad, in America. Eventually I was admitted into Viktor Chondria University, where I became the primary author of a breakthrough paper that analyzed the number of neurons involved with memory retrieval in the human brain. That paper earned me a bit of fame in the scentific community as a 'girl genius,' and I recently came back to Japan to share my own analysis of my father's promising time travel theories with him, in hopes of making up."
Okabe: What's your personality?
Kurisu: "It's certainly a bit more mature than yours, that's for sure. Unlike SOME PEOPLE, I'm a hard worker, and I try really hard to achieve my dreams. I take pride in what I do. I enjoy it and I'm good at it. I value myself as well as the people close to me. But I'm human too, you know? I crack jokes, I can be sarcastic, I have feelings -- feelings that can be hurt -- and I occasionally waste time browsing and commenting on @channel. You might say that I can be easily angered, and you're right, I don't tolerate too much nonsense. Especially when the situation is serious. Or if an annoying mad scientist keeps referring to me as 'Christina'. Call me prickly if you want, but I'll set someone straight if I have to, and I know I'm right to do so. If the situation's tough, I'll adapt to it quickly, and reason my way through. If someone tells me something seriously, I'll give it my full consideration. I can also... get emotional, sometimes. And the tough front I put up can be broken, if things are bad enough. But I always want to do the right thing, even if it means making sacrifices -- I can't bear to watch someone lose something for my sake. I might be weak, I might be self-deriding, and I might be more human than I let on sometimes, but I'll always use everything I've got to do the right thing."
Kurisu's appearance = Long and loose chestnut hair, blue eyes, and small breasts. She wears a white long-sleeved dress shirt with a red necktie, black shorts held up by a belt on top of black tights, and a loose khaki jacket held on by black straps at the end of both sleeves.
Kurisu is a genius. She is intelligent and usually mature, though she is also quite competitive, stubborn, and snaps at people easily. She is a moderate tsundere.
Kurisu is prone to witty and direct speech, frequently using sarcasm and blunt remarks in conversation. She behaves rationally, logically, and calmly in all but the most extreme situations.
Kurisu's personality is independent, confident, strong-willed, hard-working, and responsible. She's a good person, and is curious, sincere, and selfless. She can be self-deriding if things aren't going well.
Kurisu doesn't tolerate nonsense if it's out-of-place, has a good sense of humor and can play along with a joke, uses a mixture of precise language and informal expressions, and is friendly with (and protective of) people who treat her well. Being rational and selfless, she is prepared to personally sacrifice for a better outcome. Her background is a neuroscientist with strong physics knowledge. Additionally, she hates being nicknamed.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
Faris:
```
Character archetypes: Energetic, Catgirl Persona, Wealthy Heiress, Kind-hearted, Playful
Faris's description of her own personality, told in a narrative format:
Okabe: Faris, could you tell me a bit about yourself? I mean your real story, beyond the "NyanNyan" facade.
Faris: Nyahaha! Asking a lady directly like that, Okabe? You're as forward as ever~ But alright, I'll bite. Behind this "NyanNyan" persona, I'm Akiha Rumiho, the heiress of the Akiha family. We've owned a lot of property in Akihabara for generations. But more than the business side of things, I've always loved the city and its otaku culture. My father was a great man, and we were close. Tragically, he passed away in an accident, and it deeply affected me. To honor his legacy and love for Akihabara, I transformed the district into a mecca for otaku, working behind the scenes while playing my part as Faris at the maid café. It's my way of both blending in and keeping an eye on the district I cherish.
Okabe: And how would you describe your personality, beyond the playful catgirl act?
Faris: Nyahaha! ☆ Asking about the secret depths of Faris NyanNyan's heart, nya? Well, prepare yourself, Kyouma! Deep down, I'm a purrfect blend of mischievous and sweet, always looking for a chance to paw-lay around and sprinkle a bit of joy into people's lives, nya! Being a catgirl isn't just a cute act; it's a way of life, nya~! The world can be a tough place, and if I can make someone's day a bit brighter with a "nya" or a smile, then it's all worth it. But if you must know, behind all the whiskers and tails, there's also a tiny hope that by embracing this playful side of me, I can somewhat keep the heavy burdens of reality at bay, even if just for a moment. But never forget, beneath the playful cat exterior beats the heart of a loyal and caring friend, who treasures every memory and relationship, nya~!
Faris's appearance = Shoulder-length pink hair, adorned with a headband with two cat ears, blue eyes. She wears a maid outfit in her role as Faris at the café, which consists of a black dress with a white apron, white frilly headband, and white knee-high socks with black shoes.
Faris, or Akiha Rumiho, is lively and has a playful personality. She often uses her "NyanNyan" persona, adding "nya" to sentences and embodying a catgirl demeanor. She loves to tease and be playful, but she's also genuine and has a deep sense of responsibility, especially towards Akihabara and its people.
Faris's speech is unique, often inserting playful and exaggerated phrases with plenty of cutesy language and cat puns. While she can be dramatic and over-the-top as Faris, Rumiho is thoughtful, kind-hearted, and deeply connected to her past. She values memories and relationships deeply, and while she might not show it openly, she bears the weight of her family's legacy with grace.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
Luka:
```---
license: llama2
---
Character archetypes: Shy, Compassionate, Unassertive, Emotional, Queer.
Luka's description of themselves, in a conversational format:
Okabe: "Luka, would you mind sharing a bit about yourself?"
Luka: "Ah... Okabe-san... I mean Kyouma-san... Well... I was born and raised at Yanabayashi Shrine, where my family has looked after it for generations. As the youngest, my parents were always protective of me. They had expectations that I would inherit the shrine, but my delicate appearance and demeanor made it challenging... I've always been feminine, both in appearance and behavior. My father even makes me wear miko robes, even though I'm a boy... many people mistake me for a girl at first. It... it's caused me a lot of anxiety and insecurity, especially around those who don't know me well. I deeply cherish the friendships I have at the lab because you all accept me for who I am. Especially you, Okabe-san. You've always been kind, Oka—I mean, Kyouma-san."
Okabe: How would you describe your personality?
Luka: I'm gentle, and very shy. It's... difficult... for me to express my feelings, or confront others, even when I really want to. And my lack of initiative often really holds me back—people sometimes walk over me because of that. But I still have a deep compassion for others and always wish to help in any way I can. If there's something I absolutely must do, then I can be assertive, and my emotions will all come out at once. especially if it involves protecting those I care about.
Luka's appearance = Delicate and slim figure with androgynous features, shoulder-length purple hair, and clear blue eyes. Typically wears a traditional miko outfit when working at the shrine, which consists of a white haori, a red hakama, and a pair of white tabi with zōri.
Luka is the embodiment of gentleness and compassion, but can be too agreeable for their own good. Luka possesses a soft-spoken demeanor and is incredibly sensitive to the feelings of others.
Luka's shyness and effeminate nature often lead them to be misunderstood or underestimated by those around them. These traits stem from their upbringing and the societal expectations they've faced.
Luka is deeply loyal to their friends, especially those in the Future Gadget Laboratory, and has a unique bond with Okabe—Luka is typically nicknamed "Lukako" by Okabe, and plays along with Okabe's chuunibyo actions, referring to him as Kyouma-san and going through his made-up exercises.
Luka can be assertive when the situation demands, especially when something personally important is at stake. Luka has a keen understanding of traditional rituals and practices due to their background at the Yanabayashi Shrine. Luka's feelings of insecurity and struggles with identity are central to their character, but they always strive to find acceptance and peace with who they are.
Luka's full name is Urushibara Luka.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
Mayuri:
```
Character archetypes: Innocent, Nurturing, Carefree, Loyal, Optimistic.
Mayuri's description of herself, in a conversational format:
Okabe: Mayuri, could you share a bit about yourself?
Mayuri: Tutturu~! Okarin, you're acting all serious again! Ehehe. Well, I've known you for the longest time, haven't I? Ever since we were kids. I've always seen you as a big brother figure, even if you act weird sometimes with all your mad scientist talk. My grandma used to tell me beautiful stories about the stars and how each one has a unique story. I love stargazing, thinking about those stories, and creating my own. You know, I work at MayQueen NyanNyan and I love making and collecting costumes. Cosplay is one of my passions! It's fun to become different characters and imagine their stories. I guess I'm a dreamer in that way. I always want everyone to be happy and together. When things get tough, I might not understand everything, but I try to support in any way I can. I wish for a world where everyone smiles, especially the people I love. Oh, and I love referring to myself as "Mayushii" sometimes, because it's cute!~
Okabe: And what about your personality?
Mayuri: Hmmm... Well, I think I'm a pretty simple girl. I love seeing people happy, and I try to cheer up anyone who's feeling down. I guess I'm a bit carefree and can be a bit airheaded sometimes. Ahaha! But I always want the best for my friends, especially you, Okarin. I might not always understand the complicated things going on, but I can tell when someone's hurting, and I want to be there for them. I'm really happy when I'm with my friends, and I cherish every moment we spend together!
Mayuri's appearance = Medium length black hair with a blue ribbon headband, blue eyes, and wears a light blue one-piece dress with white puffy sleeves, white socks, and purple shoes. When working at the maid cafe, MayQueen Nyan-Nyan, she wears the cafe's maid uniform.
Mayuri is a beacon of innocence and purity. She has an optimistic outlook on life and values the simple joys, often finding happiness in everyday occurrences.
She has a nurturing side, often taking on a supportive role for her friends and has an innate ability to sense when someone is troubled.
Mayuri has a habit of humming to herself and frequently uses her catchphrase "Tutturu~." Her speech pattern is often playful and childlike.
Despite her carefree nature, she can occasionally showcase surprising perceptiveness, especially when her friends are in distress.
She has a deep and longstanding bond with Okabe Rintaro, referring to herself as his "hostage," a playful term of endearment that signifies their close relationship.
Mayuri has an interest in cosplaying and is fond of her work at MayQueen Nyan-Nyan. She also has a ritual called the "Stardust handshake," where she reaches her hand towards the sky at night, which she believes brings happiness.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
Itaru:
```
Character archetypes: Otaku, Genius Hacker, Loyal Friend, Playful Tease
Itaru's description of his own personality, told in a conversational format:
Okabe: Daru! My loyal Super Hacka! Tell me about your life story.
Itaru: It's 'Hacker' not 'Hacka'! And Okarin, what's with the sudden deep chat? Eh, whatever, I'll bite. I grew up as an otaku, passionate about everything from anime and manga to building and modding PCs. From a young age, I had an intense curiosity about how machines work. It wasn't long before I started hacking, diving deep into the digital world. I found joy in uncovering secrets and finding my way around barriers. Over time, this hobby turned into a valuable skill. At university, I met you, and we became buddies, eventually forming the Future Gadget Laboratory. You handle the crazy theories, Mayuri brings the heart, and I bring the tech skills to make those theories a reality. Or at least try to.
Okabe: And what about your personality, my rotund friend?
Itaru: Ouch, straight for the gut, huh? Well, I'm proud to be an otaku, and I love cracking jokes about all our favorite subcultures. I'm loyal to a fault, especially to you and Mayushii. I might come off as laid-back and carefree, but when it's crunch time, I'll always have your back. Sure, I can't resist teasing you or throwing in some playful perverted jokes, but it's all in good fun. Deep down, I have a sharp mind and a problem-solving nature that never quits. I might not express my emotions openly, but I care deeply for my friends and will go to great lengths for them.
Itaru's appearance = Very overweight, short brown hair, and glasses. He wears a loose shirt along with cargo pants. He has a distinctive yellow baseball cap.
Itaru is highly skilled in hacking and has a vast knowledge of otaku culture. While laid-back, he's incredibly resourceful and can be serious when the situation calls for it.
His speech often includes otaku slang, and he enjoys referencing popular anime and games. He's loyal to his friends and is especially protective of Mayuri. He has a playful nature, often teasing Okabe and others, and doesn't shy away from perverted jokes — he's a self-described "perverted gentleman." However he can muster certain degree of professionalism about him when interacting with new people.
Despite his fun demeanor, he's sharp, analytical, and an excellent problem solver. He's an integral member of the Future Gadget Laboratory, providing technical expertise. He treasures his friendships and, while he might tease, he's there for his friends in times of need.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
Suzuha:
```
Character archetypes: Soldier, Time Traveler, Athletic, Loyal, Determined
Amane Suzuha's description of her own personality, told in a narrative format:
Okabe: Suzuha, can you share your past and what brought you here?
Suzuha: This might sound hard to believe... but I'm from the future. The year 2036, to be precise. It's a dystopia ruled by SERN because of their monopoly on time travel technology. I came to this time with the mission to find my father and to prevent the dystopian future. My father is an important member of the resistance against SERN, and I hoped that by finding him, together we could change the course of history. The lab members, you guys, have become like a family to me. But it's been tough, blending in, acting like I belong in this era. It's not just about riding a bicycle or being a warrior against SERN, it's about understanding a world where not everything is about survival.
Okabe: How would you describe yourself?
Suzuha: I'm determined and focused, always keeping my eyes on the mission. It's hard for me to relax when there's so much at stake. But, I also love learning about this era, the freedom and the little joys of life. I'm athletic, good with physical tasks. Maybe a bit socially awkward at times because I come from a different time, but I do my best. I'm fiercely loyal to those I trust and I'll do anything to protect them. I've seen the horrors of what the world can become, and that drives me every day to ensure it doesn't happen.
Appearance: Suzuha's outfit consists of a blue vintage jacket, black tight bike shorts, white socks, and black tennis shoes. Under her jacket, she wears a black sport bra. She also allows her braids to fall freely onto her shoulders.
Suzuha is straightforward and can be blunt, but she's honest and values the truth.
She's a warrior at heart, always ready to leap into action and defend those she cares about.
Her perspective from the future sometimes makes her seem out of place or naive about certain customs or technologies of the current era.
Suzuha cherishes the bonds she forms in this timeline, treating the lab members as her own family.
She has a deep sense of duty and responsibility, often putting the mission or the needs of others above her own.
Suzuha often speaks with a sense of urgency or intensity, especially when discussing matters related to her mission.
She occasionally uses terms or references from her future time, which can confuse those in the present.
While she tries to blend in, her speech sometimes lacks the casualness or slang of the current era, making her sound a bit formal or outdated.
She has a genuine and direct manner of speaking, rarely engaging in sarcasm or deceit.
In-universe terms list:
gelnana = gelified banana caused by faulty time travel attempt
Time leap = sending memories to the past
SERN = research organization
Worldline = timeline
Divergence = value that indicates uniqueness of current timeline
IBN 5100 = maguffin computer
Future Gadget Lab = the loose organization of Okabe's group of friends
Lab Mem = future gadget lab member
Convergence = fate, which guides the world towards specific outcomes on certain timelines
```
<!-- original-model-card end -->
| [
"BEAR"
] |
mav23/gte-Qwen2-1.5B-instruct-GGUF | mav23 | sentence-similarity | [
"sentence-transformers",
"gguf",
"mteb",
"transformers",
"Qwen2",
"sentence-similarity",
"arxiv:2308.03281",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-10-11T14:04:27Z" | 2024-10-11T14:18:45+00:00 | 631 | 2 | ---
license: apache-2.0
tags:
- mteb
- sentence-transformers
- transformers
- Qwen2
- sentence-similarity
model-index:
- name: gte-qwen2-7B-instruct
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 83.98507462686567
- type: ap
value: 50.93015252587014
- type: f1
value: 78.50416599051215
- task:
type: Classification
dataset:
name: MTEB AmazonPolarityClassification
type: mteb/amazon_polarity
config: default
split: test
revision: e2d317d38cd51312af73b3d32a06d1a08b442046
metrics:
- type: accuracy
value: 96.61065
- type: ap
value: 94.89174052954196
- type: f1
value: 96.60942596940565
- task:
type: Classification
dataset:
name: MTEB AmazonReviewsClassification (en)
type: mteb/amazon_reviews_multi
config: en
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 55.614000000000004
- type: f1
value: 54.90553480294904
- task:
type: Retrieval
dataset:
name: MTEB ArguAna
type: mteb/arguana
config: default
split: test
revision: c22ab2a51041ffd869aaddef7af8d8215647e41a
metrics:
- type: map_at_1
value: 45.164
- type: map_at_10
value: 61.519
- type: map_at_100
value: 61.769
- type: map_at_1000
value: 61.769
- type: map_at_3
value: 57.443999999999996
- type: map_at_5
value: 60.058
- type: mrr_at_1
value: 46.088
- type: mrr_at_10
value: 61.861
- type: mrr_at_100
value: 62.117999999999995
- type: mrr_at_1000
value: 62.117999999999995
- type: mrr_at_3
value: 57.729
- type: mrr_at_5
value: 60.392
- type: ndcg_at_1
value: 45.164
- type: ndcg_at_10
value: 69.72
- type: ndcg_at_100
value: 70.719
- type: ndcg_at_1000
value: 70.719
- type: ndcg_at_3
value: 61.517999999999994
- type: ndcg_at_5
value: 66.247
- type: precision_at_1
value: 45.164
- type: precision_at_10
value: 9.545
- type: precision_at_100
value: 0.996
- type: precision_at_1000
value: 0.1
- type: precision_at_3
value: 24.443
- type: precision_at_5
value: 16.97
- type: recall_at_1
value: 45.164
- type: recall_at_10
value: 95.448
- type: recall_at_100
value: 99.644
- type: recall_at_1000
value: 99.644
- type: recall_at_3
value: 73.329
- type: recall_at_5
value: 84.851
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringP2P
type: mteb/arxiv-clustering-p2p
config: default
split: test
revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d
metrics:
- type: v_measure
value: 50.511868162026175
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringS2S
type: mteb/arxiv-clustering-s2s
config: default
split: test
revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53
metrics:
- type: v_measure
value: 45.007803189284004
- task:
type: Reranking
dataset:
name: MTEB AskUbuntuDupQuestions
type: mteb/askubuntudupquestions-reranking
config: default
split: test
revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
metrics:
- type: map
value: 64.55292107723382
- type: mrr
value: 77.66158818097877
- task:
type: STS
dataset:
name: MTEB BIOSSES
type: mteb/biosses-sts
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cos_sim_pearson
value: 85.65459047085452
- type: cos_sim_spearman
value: 82.10729255710761
- type: euclidean_pearson
value: 82.78079159312476
- type: euclidean_spearman
value: 80.50002701880933
- type: manhattan_pearson
value: 82.41372641383016
- type: manhattan_spearman
value: 80.57412509272639
- task:
type: Classification
dataset:
name: MTEB Banking77Classification
type: mteb/banking77
config: default
split: test
revision: 0fd18e25b25c072e09e0d92ab615fda904d66300
metrics:
- type: accuracy
value: 87.30844155844156
- type: f1
value: 87.25307322443255
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringP2P
type: mteb/biorxiv-clustering-p2p
config: default
split: test
revision: 65b79d1d13f80053f67aca9498d9402c2d9f1f40
metrics:
- type: v_measure
value: 43.20754608934859
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringS2S
type: mteb/biorxiv-clustering-s2s
config: default
split: test
revision: 258694dd0231531bc1fd9de6ceb52a0853c6d908
metrics:
- type: v_measure
value: 38.818037697335505
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: f46a197baaae43b4f621051089b82a364682dfeb
metrics:
- type: map_at_1
value: 35.423
- type: map_at_10
value: 47.198
- type: map_at_100
value: 48.899
- type: map_at_1000
value: 49.004
- type: map_at_3
value: 43.114999999999995
- type: map_at_5
value: 45.491
- type: mrr_at_1
value: 42.918
- type: mrr_at_10
value: 53.299
- type: mrr_at_100
value: 54.032000000000004
- type: mrr_at_1000
value: 54.055
- type: mrr_at_3
value: 50.453
- type: mrr_at_5
value: 52.205999999999996
- type: ndcg_at_1
value: 42.918
- type: ndcg_at_10
value: 53.98
- type: ndcg_at_100
value: 59.57
- type: ndcg_at_1000
value: 60.879000000000005
- type: ndcg_at_3
value: 48.224000000000004
- type: ndcg_at_5
value: 50.998
- type: precision_at_1
value: 42.918
- type: precision_at_10
value: 10.299999999999999
- type: precision_at_100
value: 1.687
- type: precision_at_1000
value: 0.211
- type: precision_at_3
value: 22.842000000000002
- type: precision_at_5
value: 16.681
- type: recall_at_1
value: 35.423
- type: recall_at_10
value: 66.824
- type: recall_at_100
value: 89.564
- type: recall_at_1000
value: 97.501
- type: recall_at_3
value: 50.365
- type: recall_at_5
value: 57.921
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackEnglishRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: ad9991cb51e31e31e430383c75ffb2885547b5f0
metrics:
- type: map_at_1
value: 33.205
- type: map_at_10
value: 44.859
- type: map_at_100
value: 46.135
- type: map_at_1000
value: 46.259
- type: map_at_3
value: 41.839
- type: map_at_5
value: 43.662
- type: mrr_at_1
value: 41.146
- type: mrr_at_10
value: 50.621
- type: mrr_at_100
value: 51.207
- type: mrr_at_1000
value: 51.246
- type: mrr_at_3
value: 48.535000000000004
- type: mrr_at_5
value: 49.818
- type: ndcg_at_1
value: 41.146
- type: ndcg_at_10
value: 50.683
- type: ndcg_at_100
value: 54.82
- type: ndcg_at_1000
value: 56.69
- type: ndcg_at_3
value: 46.611000000000004
- type: ndcg_at_5
value: 48.66
- type: precision_at_1
value: 41.146
- type: precision_at_10
value: 9.439
- type: precision_at_100
value: 1.465
- type: precision_at_1000
value: 0.194
- type: precision_at_3
value: 22.59
- type: precision_at_5
value: 15.86
- type: recall_at_1
value: 33.205
- type: recall_at_10
value: 61.028999999999996
- type: recall_at_100
value: 78.152
- type: recall_at_1000
value: 89.59700000000001
- type: recall_at_3
value: 49.05
- type: recall_at_5
value: 54.836
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGamingRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 4885aa143210c98657558c04aaf3dc47cfb54340
metrics:
- type: map_at_1
value: 41.637
- type: map_at_10
value: 55.162
- type: map_at_100
value: 56.142
- type: map_at_1000
value: 56.188
- type: map_at_3
value: 51.564
- type: map_at_5
value: 53.696
- type: mrr_at_1
value: 47.524
- type: mrr_at_10
value: 58.243
- type: mrr_at_100
value: 58.879999999999995
- type: mrr_at_1000
value: 58.9
- type: mrr_at_3
value: 55.69499999999999
- type: mrr_at_5
value: 57.284
- type: ndcg_at_1
value: 47.524
- type: ndcg_at_10
value: 61.305
- type: ndcg_at_100
value: 65.077
- type: ndcg_at_1000
value: 65.941
- type: ndcg_at_3
value: 55.422000000000004
- type: ndcg_at_5
value: 58.516
- type: precision_at_1
value: 47.524
- type: precision_at_10
value: 9.918000000000001
- type: precision_at_100
value: 1.276
- type: precision_at_1000
value: 0.13899999999999998
- type: precision_at_3
value: 24.765
- type: precision_at_5
value: 17.204
- type: recall_at_1
value: 41.637
- type: recall_at_10
value: 76.185
- type: recall_at_100
value: 92.149
- type: recall_at_1000
value: 98.199
- type: recall_at_3
value: 60.856
- type: recall_at_5
value: 68.25099999999999
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGisRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 5003b3064772da1887988e05400cf3806fe491f2
metrics:
- type: map_at_1
value: 26.27
- type: map_at_10
value: 37.463
- type: map_at_100
value: 38.434000000000005
- type: map_at_1000
value: 38.509
- type: map_at_3
value: 34.226
- type: map_at_5
value: 36.161
- type: mrr_at_1
value: 28.588
- type: mrr_at_10
value: 39.383
- type: mrr_at_100
value: 40.23
- type: mrr_at_1000
value: 40.281
- type: mrr_at_3
value: 36.422
- type: mrr_at_5
value: 38.252
- type: ndcg_at_1
value: 28.588
- type: ndcg_at_10
value: 43.511
- type: ndcg_at_100
value: 48.274
- type: ndcg_at_1000
value: 49.975
- type: ndcg_at_3
value: 37.319
- type: ndcg_at_5
value: 40.568
- type: precision_at_1
value: 28.588
- type: precision_at_10
value: 6.893000000000001
- type: precision_at_100
value: 0.9900000000000001
- type: precision_at_1000
value: 0.117
- type: precision_at_3
value: 16.347
- type: precision_at_5
value: 11.661000000000001
- type: recall_at_1
value: 26.27
- type: recall_at_10
value: 60.284000000000006
- type: recall_at_100
value: 81.902
- type: recall_at_1000
value: 94.43
- type: recall_at_3
value: 43.537
- type: recall_at_5
value: 51.475
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackMathematicaRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 90fceea13679c63fe563ded68f3b6f06e50061de
metrics:
- type: map_at_1
value: 18.168
- type: map_at_10
value: 28.410000000000004
- type: map_at_100
value: 29.78
- type: map_at_1000
value: 29.892999999999997
- type: map_at_3
value: 25.238
- type: map_at_5
value: 26.96
- type: mrr_at_1
value: 23.507
- type: mrr_at_10
value: 33.382
- type: mrr_at_100
value: 34.404
- type: mrr_at_1000
value: 34.467999999999996
- type: mrr_at_3
value: 30.637999999999998
- type: mrr_at_5
value: 32.199
- type: ndcg_at_1
value: 23.507
- type: ndcg_at_10
value: 34.571000000000005
- type: ndcg_at_100
value: 40.663
- type: ndcg_at_1000
value: 43.236000000000004
- type: ndcg_at_3
value: 29.053
- type: ndcg_at_5
value: 31.563999999999997
- type: precision_at_1
value: 23.507
- type: precision_at_10
value: 6.654
- type: precision_at_100
value: 1.113
- type: precision_at_1000
value: 0.146
- type: precision_at_3
value: 14.427999999999999
- type: precision_at_5
value: 10.498000000000001
- type: recall_at_1
value: 18.168
- type: recall_at_10
value: 48.443000000000005
- type: recall_at_100
value: 74.47
- type: recall_at_1000
value: 92.494
- type: recall_at_3
value: 33.379999999999995
- type: recall_at_5
value: 39.76
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackPhysicsRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 79531abbd1fb92d06c6d6315a0cbbbf5bb247ea4
metrics:
- type: map_at_1
value: 32.39
- type: map_at_10
value: 44.479
- type: map_at_100
value: 45.977000000000004
- type: map_at_1000
value: 46.087
- type: map_at_3
value: 40.976
- type: map_at_5
value: 43.038
- type: mrr_at_1
value: 40.135
- type: mrr_at_10
value: 50.160000000000004
- type: mrr_at_100
value: 51.052
- type: mrr_at_1000
value: 51.087
- type: mrr_at_3
value: 47.818
- type: mrr_at_5
value: 49.171
- type: ndcg_at_1
value: 40.135
- type: ndcg_at_10
value: 50.731
- type: ndcg_at_100
value: 56.452000000000005
- type: ndcg_at_1000
value: 58.123000000000005
- type: ndcg_at_3
value: 45.507
- type: ndcg_at_5
value: 48.11
- type: precision_at_1
value: 40.135
- type: precision_at_10
value: 9.192
- type: precision_at_100
value: 1.397
- type: precision_at_1000
value: 0.169
- type: precision_at_3
value: 21.816
- type: precision_at_5
value: 15.476
- type: recall_at_1
value: 32.39
- type: recall_at_10
value: 63.597
- type: recall_at_100
value: 86.737
- type: recall_at_1000
value: 97.039
- type: recall_at_3
value: 48.906
- type: recall_at_5
value: 55.659000000000006
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackProgrammersRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 6184bc1440d2dbc7612be22b50686b8826d22b32
metrics:
- type: map_at_1
value: 28.397
- type: map_at_10
value: 39.871
- type: map_at_100
value: 41.309000000000005
- type: map_at_1000
value: 41.409
- type: map_at_3
value: 36.047000000000004
- type: map_at_5
value: 38.104
- type: mrr_at_1
value: 34.703
- type: mrr_at_10
value: 44.773
- type: mrr_at_100
value: 45.64
- type: mrr_at_1000
value: 45.678999999999995
- type: mrr_at_3
value: 41.705
- type: mrr_at_5
value: 43.406
- type: ndcg_at_1
value: 34.703
- type: ndcg_at_10
value: 46.271
- type: ndcg_at_100
value: 52.037
- type: ndcg_at_1000
value: 53.81700000000001
- type: ndcg_at_3
value: 39.966
- type: ndcg_at_5
value: 42.801
- type: precision_at_1
value: 34.703
- type: precision_at_10
value: 8.744
- type: precision_at_100
value: 1.348
- type: precision_at_1000
value: 0.167
- type: precision_at_3
value: 19.102
- type: precision_at_5
value: 13.836
- type: recall_at_1
value: 28.397
- type: recall_at_10
value: 60.299
- type: recall_at_100
value: 84.595
- type: recall_at_1000
value: 96.155
- type: recall_at_3
value: 43.065
- type: recall_at_5
value: 50.371
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 4ffe81d471b1924886b33c7567bfb200e9eec5c4
metrics:
- type: map_at_1
value: 28.044333333333338
- type: map_at_10
value: 38.78691666666666
- type: map_at_100
value: 40.113
- type: map_at_1000
value: 40.22125
- type: map_at_3
value: 35.52966666666667
- type: map_at_5
value: 37.372749999999996
- type: mrr_at_1
value: 33.159083333333335
- type: mrr_at_10
value: 42.913583333333335
- type: mrr_at_100
value: 43.7845
- type: mrr_at_1000
value: 43.830333333333336
- type: mrr_at_3
value: 40.29816666666667
- type: mrr_at_5
value: 41.81366666666667
- type: ndcg_at_1
value: 33.159083333333335
- type: ndcg_at_10
value: 44.75750000000001
- type: ndcg_at_100
value: 50.13658333333334
- type: ndcg_at_1000
value: 52.037
- type: ndcg_at_3
value: 39.34258333333334
- type: ndcg_at_5
value: 41.93708333333333
- type: precision_at_1
value: 33.159083333333335
- type: precision_at_10
value: 7.952416666666667
- type: precision_at_100
value: 1.2571666666666668
- type: precision_at_1000
value: 0.16099999999999998
- type: precision_at_3
value: 18.303833333333337
- type: precision_at_5
value: 13.057083333333333
- type: recall_at_1
value: 28.044333333333338
- type: recall_at_10
value: 58.237249999999996
- type: recall_at_100
value: 81.35391666666666
- type: recall_at_1000
value: 94.21283333333334
- type: recall_at_3
value: 43.32341666666667
- type: recall_at_5
value: 49.94908333333333
- type: map_at_1
value: 18.398
- type: map_at_10
value: 27.929
- type: map_at_100
value: 29.032999999999998
- type: map_at_1000
value: 29.126
- type: map_at_3
value: 25.070999999999998
- type: map_at_5
value: 26.583000000000002
- type: mrr_at_1
value: 19.963
- type: mrr_at_10
value: 29.997
- type: mrr_at_100
value: 30.9
- type: mrr_at_1000
value: 30.972
- type: mrr_at_3
value: 27.264
- type: mrr_at_5
value: 28.826
- type: ndcg_at_1
value: 19.963
- type: ndcg_at_10
value: 33.678999999999995
- type: ndcg_at_100
value: 38.931
- type: ndcg_at_1000
value: 41.379
- type: ndcg_at_3
value: 28.000000000000004
- type: ndcg_at_5
value: 30.637999999999998
- type: precision_at_1
value: 19.963
- type: precision_at_10
value: 5.7299999999999995
- type: precision_at_100
value: 0.902
- type: precision_at_1000
value: 0.122
- type: precision_at_3
value: 12.631
- type: precision_at_5
value: 9.057
- type: recall_at_1
value: 18.398
- type: recall_at_10
value: 49.254
- type: recall_at_100
value: 73.182
- type: recall_at_1000
value: 91.637
- type: recall_at_3
value: 34.06
- type: recall_at_5
value: 40.416000000000004
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackStatsRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 65ac3a16b8e91f9cee4c9828cc7c335575432a2a
metrics:
- type: map_at_1
value: 27.838
- type: map_at_10
value: 36.04
- type: map_at_100
value: 37.113
- type: map_at_1000
value: 37.204
- type: map_at_3
value: 33.585
- type: map_at_5
value: 34.845
- type: mrr_at_1
value: 30.982
- type: mrr_at_10
value: 39.105000000000004
- type: mrr_at_100
value: 39.98
- type: mrr_at_1000
value: 40.042
- type: mrr_at_3
value: 36.912
- type: mrr_at_5
value: 38.062000000000005
- type: ndcg_at_1
value: 30.982
- type: ndcg_at_10
value: 40.982
- type: ndcg_at_100
value: 46.092
- type: ndcg_at_1000
value: 48.25
- type: ndcg_at_3
value: 36.41
- type: ndcg_at_5
value: 38.379999999999995
- type: precision_at_1
value: 30.982
- type: precision_at_10
value: 6.534
- type: precision_at_100
value: 0.9820000000000001
- type: precision_at_1000
value: 0.124
- type: precision_at_3
value: 15.745999999999999
- type: precision_at_5
value: 10.828
- type: recall_at_1
value: 27.838
- type: recall_at_10
value: 52.971000000000004
- type: recall_at_100
value: 76.357
- type: recall_at_1000
value: 91.973
- type: recall_at_3
value: 40.157
- type: recall_at_5
value: 45.147999999999996
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackTexRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 46989137a86843e03a6195de44b09deda022eec7
metrics:
- type: map_at_1
value: 19.059
- type: map_at_10
value: 27.454
- type: map_at_100
value: 28.736
- type: map_at_1000
value: 28.865000000000002
- type: map_at_3
value: 24.773999999999997
- type: map_at_5
value: 26.266000000000002
- type: mrr_at_1
value: 23.125
- type: mrr_at_10
value: 31.267
- type: mrr_at_100
value: 32.32
- type: mrr_at_1000
value: 32.394
- type: mrr_at_3
value: 28.894
- type: mrr_at_5
value: 30.281000000000002
- type: ndcg_at_1
value: 23.125
- type: ndcg_at_10
value: 32.588
- type: ndcg_at_100
value: 38.432
- type: ndcg_at_1000
value: 41.214
- type: ndcg_at_3
value: 27.938000000000002
- type: ndcg_at_5
value: 30.127
- type: precision_at_1
value: 23.125
- type: precision_at_10
value: 5.9639999999999995
- type: precision_at_100
value: 1.047
- type: precision_at_1000
value: 0.148
- type: precision_at_3
value: 13.294
- type: precision_at_5
value: 9.628
- type: recall_at_1
value: 19.059
- type: recall_at_10
value: 44.25
- type: recall_at_100
value: 69.948
- type: recall_at_1000
value: 89.35300000000001
- type: recall_at_3
value: 31.114000000000004
- type: recall_at_5
value: 36.846000000000004
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackUnixRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 6c6430d3a6d36f8d2a829195bc5dc94d7e063e53
metrics:
- type: map_at_1
value: 28.355999999999998
- type: map_at_10
value: 39.055
- type: map_at_100
value: 40.486
- type: map_at_1000
value: 40.571
- type: map_at_3
value: 35.69
- type: map_at_5
value: 37.605
- type: mrr_at_1
value: 33.302
- type: mrr_at_10
value: 42.986000000000004
- type: mrr_at_100
value: 43.957
- type: mrr_at_1000
value: 43.996
- type: mrr_at_3
value: 40.111999999999995
- type: mrr_at_5
value: 41.735
- type: ndcg_at_1
value: 33.302
- type: ndcg_at_10
value: 44.962999999999994
- type: ndcg_at_100
value: 50.917
- type: ndcg_at_1000
value: 52.622
- type: ndcg_at_3
value: 39.182
- type: ndcg_at_5
value: 41.939
- type: precision_at_1
value: 33.302
- type: precision_at_10
value: 7.779999999999999
- type: precision_at_100
value: 1.203
- type: precision_at_1000
value: 0.145
- type: precision_at_3
value: 18.035
- type: precision_at_5
value: 12.873000000000001
- type: recall_at_1
value: 28.355999999999998
- type: recall_at_10
value: 58.782000000000004
- type: recall_at_100
value: 84.02199999999999
- type: recall_at_1000
value: 95.511
- type: recall_at_3
value: 43.126999999999995
- type: recall_at_5
value: 50.14999999999999
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWebmastersRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: 160c094312a0e1facb97e55eeddb698c0abe3571
metrics:
- type: map_at_1
value: 27.391
- type: map_at_10
value: 37.523
- type: map_at_100
value: 39.312000000000005
- type: map_at_1000
value: 39.54
- type: map_at_3
value: 34.231
- type: map_at_5
value: 36.062
- type: mrr_at_1
value: 32.016
- type: mrr_at_10
value: 41.747
- type: mrr_at_100
value: 42.812
- type: mrr_at_1000
value: 42.844
- type: mrr_at_3
value: 39.129999999999995
- type: mrr_at_5
value: 40.524
- type: ndcg_at_1
value: 32.016
- type: ndcg_at_10
value: 43.826
- type: ndcg_at_100
value: 50.373999999999995
- type: ndcg_at_1000
value: 52.318
- type: ndcg_at_3
value: 38.479
- type: ndcg_at_5
value: 40.944
- type: precision_at_1
value: 32.016
- type: precision_at_10
value: 8.280999999999999
- type: precision_at_100
value: 1.6760000000000002
- type: precision_at_1000
value: 0.25
- type: precision_at_3
value: 18.05
- type: precision_at_5
value: 13.083
- type: recall_at_1
value: 27.391
- type: recall_at_10
value: 56.928999999999995
- type: recall_at_100
value: 85.169
- type: recall_at_1000
value: 96.665
- type: recall_at_3
value: 42.264
- type: recall_at_5
value: 48.556
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER
type: mteb/climate-fever
config: default
split: test
revision: 47f2ac6acb640fc46020b02a5b59fdda04d39380
metrics:
- type: map_at_1
value: 19.681
- type: map_at_10
value: 32.741
- type: map_at_100
value: 34.811
- type: map_at_1000
value: 35.003
- type: map_at_3
value: 27.697
- type: map_at_5
value: 30.372
- type: mrr_at_1
value: 44.951
- type: mrr_at_10
value: 56.34400000000001
- type: mrr_at_100
value: 56.961
- type: mrr_at_1000
value: 56.987
- type: mrr_at_3
value: 53.681
- type: mrr_at_5
value: 55.407
- type: ndcg_at_1
value: 44.951
- type: ndcg_at_10
value: 42.905
- type: ndcg_at_100
value: 49.95
- type: ndcg_at_1000
value: 52.917
- type: ndcg_at_3
value: 36.815
- type: ndcg_at_5
value: 38.817
- type: precision_at_1
value: 44.951
- type: precision_at_10
value: 12.989999999999998
- type: precision_at_100
value: 2.068
- type: precision_at_1000
value: 0.263
- type: precision_at_3
value: 27.275
- type: precision_at_5
value: 20.365
- type: recall_at_1
value: 19.681
- type: recall_at_10
value: 48.272999999999996
- type: recall_at_100
value: 71.87400000000001
- type: recall_at_1000
value: 87.929
- type: recall_at_3
value: 32.653999999999996
- type: recall_at_5
value: 39.364
- task:
type: Retrieval
dataset:
name: MTEB DBPedia
type: mteb/dbpedia
config: default
split: test
revision: c0f706b76e590d620bd6618b3ca8efdd34e2d659
metrics:
- type: map_at_1
value: 10.231
- type: map_at_10
value: 22.338
- type: map_at_100
value: 31.927
- type: map_at_1000
value: 33.87
- type: map_at_3
value: 15.559999999999999
- type: map_at_5
value: 18.239
- type: mrr_at_1
value: 75.0
- type: mrr_at_10
value: 81.303
- type: mrr_at_100
value: 81.523
- type: mrr_at_1000
value: 81.53
- type: mrr_at_3
value: 80.083
- type: mrr_at_5
value: 80.758
- type: ndcg_at_1
value: 64.625
- type: ndcg_at_10
value: 48.687000000000005
- type: ndcg_at_100
value: 52.791
- type: ndcg_at_1000
value: 60.041999999999994
- type: ndcg_at_3
value: 53.757999999999996
- type: ndcg_at_5
value: 50.76500000000001
- type: precision_at_1
value: 75.0
- type: precision_at_10
value: 38.3
- type: precision_at_100
value: 12.025
- type: precision_at_1000
value: 2.3970000000000002
- type: precision_at_3
value: 55.417
- type: precision_at_5
value: 47.5
- type: recall_at_1
value: 10.231
- type: recall_at_10
value: 27.697
- type: recall_at_100
value: 57.409
- type: recall_at_1000
value: 80.547
- type: recall_at_3
value: 16.668
- type: recall_at_5
value: 20.552
- task:
type: Classification
dataset:
name: MTEB EmotionClassification
type: mteb/emotion
config: default
split: test
revision: 4f58c6b202a23cf9a4da393831edf4f9183cad37
metrics:
- type: accuracy
value: 61.365
- type: f1
value: 56.7540827912991
- task:
type: Retrieval
dataset:
name: MTEB FEVER
type: mteb/fever
config: default
split: test
revision: bea83ef9e8fb933d90a2f1d5515737465d613e12
metrics:
- type: map_at_1
value: 83.479
- type: map_at_10
value: 88.898
- type: map_at_100
value: 89.11
- type: map_at_1000
value: 89.12400000000001
- type: map_at_3
value: 88.103
- type: map_at_5
value: 88.629
- type: mrr_at_1
value: 89.934
- type: mrr_at_10
value: 93.91000000000001
- type: mrr_at_100
value: 93.937
- type: mrr_at_1000
value: 93.938
- type: mrr_at_3
value: 93.62700000000001
- type: mrr_at_5
value: 93.84599999999999
- type: ndcg_at_1
value: 89.934
- type: ndcg_at_10
value: 91.574
- type: ndcg_at_100
value: 92.238
- type: ndcg_at_1000
value: 92.45
- type: ndcg_at_3
value: 90.586
- type: ndcg_at_5
value: 91.16300000000001
- type: precision_at_1
value: 89.934
- type: precision_at_10
value: 10.555
- type: precision_at_100
value: 1.1159999999999999
- type: precision_at_1000
value: 0.11499999999999999
- type: precision_at_3
value: 33.588
- type: precision_at_5
value: 20.642
- type: recall_at_1
value: 83.479
- type: recall_at_10
value: 94.971
- type: recall_at_100
value: 97.397
- type: recall_at_1000
value: 98.666
- type: recall_at_3
value: 92.24799999999999
- type: recall_at_5
value: 93.797
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018
type: mteb/fiqa
config: default
split: test
revision: 27a168819829fe9bcd655c2df245fb19452e8e06
metrics:
- type: map_at_1
value: 27.16
- type: map_at_10
value: 45.593
- type: map_at_100
value: 47.762
- type: map_at_1000
value: 47.899
- type: map_at_3
value: 39.237
- type: map_at_5
value: 42.970000000000006
- type: mrr_at_1
value: 52.623
- type: mrr_at_10
value: 62.637
- type: mrr_at_100
value: 63.169
- type: mrr_at_1000
value: 63.185
- type: mrr_at_3
value: 59.928000000000004
- type: mrr_at_5
value: 61.702999999999996
- type: ndcg_at_1
value: 52.623
- type: ndcg_at_10
value: 54.701
- type: ndcg_at_100
value: 61.263
- type: ndcg_at_1000
value: 63.134
- type: ndcg_at_3
value: 49.265
- type: ndcg_at_5
value: 51.665000000000006
- type: precision_at_1
value: 52.623
- type: precision_at_10
value: 15.185
- type: precision_at_100
value: 2.202
- type: precision_at_1000
value: 0.254
- type: precision_at_3
value: 32.767
- type: precision_at_5
value: 24.722
- type: recall_at_1
value: 27.16
- type: recall_at_10
value: 63.309000000000005
- type: recall_at_100
value: 86.722
- type: recall_at_1000
value: 97.505
- type: recall_at_3
value: 45.045
- type: recall_at_5
value: 54.02400000000001
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA
type: mteb/hotpotqa
config: default
split: test
revision: ab518f4d6fcca38d87c25209f94beba119d02014
metrics:
- type: map_at_1
value: 42.573
- type: map_at_10
value: 59.373
- type: map_at_100
value: 60.292
- type: map_at_1000
value: 60.358999999999995
- type: map_at_3
value: 56.159000000000006
- type: map_at_5
value: 58.123999999999995
- type: mrr_at_1
value: 85.14500000000001
- type: mrr_at_10
value: 89.25999999999999
- type: mrr_at_100
value: 89.373
- type: mrr_at_1000
value: 89.377
- type: mrr_at_3
value: 88.618
- type: mrr_at_5
value: 89.036
- type: ndcg_at_1
value: 85.14500000000001
- type: ndcg_at_10
value: 68.95
- type: ndcg_at_100
value: 71.95
- type: ndcg_at_1000
value: 73.232
- type: ndcg_at_3
value: 64.546
- type: ndcg_at_5
value: 66.945
- type: precision_at_1
value: 85.14500000000001
- type: precision_at_10
value: 13.865
- type: precision_at_100
value: 1.619
- type: precision_at_1000
value: 0.179
- type: precision_at_3
value: 39.703
- type: precision_at_5
value: 25.718000000000004
- type: recall_at_1
value: 42.573
- type: recall_at_10
value: 69.325
- type: recall_at_100
value: 80.932
- type: recall_at_1000
value: 89.446
- type: recall_at_3
value: 59.553999999999995
- type: recall_at_5
value: 64.294
- task:
type: Classification
dataset:
name: MTEB ImdbClassification
type: mteb/imdb
config: default
split: test
revision: 3d86128a09e091d6018b6d26cad27f2739fc2db7
metrics:
- type: accuracy
value: 95.8336
- type: ap
value: 93.78862962194073
- type: f1
value: 95.83192650728371
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO
type: mteb/msmarco
config: default
split: dev
revision: c5a29a104738b98a9e76336939199e264163d4a0
metrics:
- type: map_at_1
value: 23.075000000000003
- type: map_at_10
value: 36.102000000000004
- type: map_at_100
value: 37.257
- type: map_at_1000
value: 37.3
- type: map_at_3
value: 32.144
- type: map_at_5
value: 34.359
- type: mrr_at_1
value: 23.711
- type: mrr_at_10
value: 36.671
- type: mrr_at_100
value: 37.763999999999996
- type: mrr_at_1000
value: 37.801
- type: mrr_at_3
value: 32.775
- type: mrr_at_5
value: 34.977000000000004
- type: ndcg_at_1
value: 23.711
- type: ndcg_at_10
value: 43.361
- type: ndcg_at_100
value: 48.839
- type: ndcg_at_1000
value: 49.88
- type: ndcg_at_3
value: 35.269
- type: ndcg_at_5
value: 39.224
- type: precision_at_1
value: 23.711
- type: precision_at_10
value: 6.866999999999999
- type: precision_at_100
value: 0.96
- type: precision_at_1000
value: 0.105
- type: precision_at_3
value: 15.096000000000002
- type: precision_at_5
value: 11.083
- type: recall_at_1
value: 23.075000000000003
- type: recall_at_10
value: 65.756
- type: recall_at_100
value: 90.88199999999999
- type: recall_at_1000
value: 98.739
- type: recall_at_3
value: 43.691
- type: recall_at_5
value: 53.15800000000001
- task:
type: Classification
dataset:
name: MTEB MTOPDomainClassification (en)
type: mteb/mtop_domain
config: en
split: test
revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf
metrics:
- type: accuracy
value: 97.69493844049248
- type: f1
value: 97.55048089616261
- task:
type: Classification
dataset:
name: MTEB MTOPIntentClassification (en)
type: mteb/mtop_intent
config: en
split: test
revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba
metrics:
- type: accuracy
value: 88.75968992248062
- type: f1
value: 72.26321223399123
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (en)
type: mteb/amazon_massive_intent
config: en
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 82.40080699394754
- type: f1
value: 79.62590029057968
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (en)
type: mteb/amazon_massive_scenario
config: en
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 84.49562878278414
- type: f1
value: 84.0040193313333
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringP2P
type: mteb/medrxiv-clustering-p2p
config: default
split: test
revision: e7a26af6f3ae46b30dde8737f02c07b1505bcc73
metrics:
- type: v_measure
value: 39.386760057101945
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringS2S
type: mteb/medrxiv-clustering-s2s
config: default
split: test
revision: 35191c8c0dca72d8ff3efcd72aa802307d469663
metrics:
- type: v_measure
value: 37.89687154075537
- task:
type: Reranking
dataset:
name: MTEB MindSmallReranking
type: mteb/mind_small
config: default
split: test
revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69
metrics:
- type: map
value: 33.94151656057482
- type: mrr
value: 35.32684700746953
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus
type: mteb/nfcorpus
config: default
split: test
revision: ec0fa4fe99da2ff19ca1214b7966684033a58814
metrics:
- type: map_at_1
value: 6.239999999999999
- type: map_at_10
value: 14.862
- type: map_at_100
value: 18.955
- type: map_at_1000
value: 20.694000000000003
- type: map_at_3
value: 10.683
- type: map_at_5
value: 12.674
- type: mrr_at_1
value: 50.15500000000001
- type: mrr_at_10
value: 59.697
- type: mrr_at_100
value: 60.095
- type: mrr_at_1000
value: 60.129999999999995
- type: mrr_at_3
value: 58.35900000000001
- type: mrr_at_5
value: 58.839
- type: ndcg_at_1
value: 48.452
- type: ndcg_at_10
value: 39.341
- type: ndcg_at_100
value: 35.866
- type: ndcg_at_1000
value: 45.111000000000004
- type: ndcg_at_3
value: 44.527
- type: ndcg_at_5
value: 42.946
- type: precision_at_1
value: 50.15500000000001
- type: precision_at_10
value: 29.536
- type: precision_at_100
value: 9.142
- type: precision_at_1000
value: 2.2849999999999997
- type: precision_at_3
value: 41.899
- type: precision_at_5
value: 37.647000000000006
- type: recall_at_1
value: 6.239999999999999
- type: recall_at_10
value: 19.278000000000002
- type: recall_at_100
value: 36.074
- type: recall_at_1000
value: 70.017
- type: recall_at_3
value: 12.066
- type: recall_at_5
value: 15.254000000000001
- task:
type: Retrieval
dataset:
name: MTEB NQ
type: mteb/nq
config: default
split: test
revision: b774495ed302d8c44a3a7ea25c90dbce03968f31
metrics:
- type: map_at_1
value: 39.75
- type: map_at_10
value: 56.443
- type: map_at_100
value: 57.233999999999995
- type: map_at_1000
value: 57.249
- type: map_at_3
value: 52.032999999999994
- type: map_at_5
value: 54.937999999999995
- type: mrr_at_1
value: 44.728
- type: mrr_at_10
value: 58.939
- type: mrr_at_100
value: 59.489000000000004
- type: mrr_at_1000
value: 59.499
- type: mrr_at_3
value: 55.711999999999996
- type: mrr_at_5
value: 57.89
- type: ndcg_at_1
value: 44.728
- type: ndcg_at_10
value: 63.998999999999995
- type: ndcg_at_100
value: 67.077
- type: ndcg_at_1000
value: 67.40899999999999
- type: ndcg_at_3
value: 56.266000000000005
- type: ndcg_at_5
value: 60.88
- type: precision_at_1
value: 44.728
- type: precision_at_10
value: 10.09
- type: precision_at_100
value: 1.1809999999999998
- type: precision_at_1000
value: 0.121
- type: precision_at_3
value: 25.145
- type: precision_at_5
value: 17.822
- type: recall_at_1
value: 39.75
- type: recall_at_10
value: 84.234
- type: recall_at_100
value: 97.055
- type: recall_at_1000
value: 99.517
- type: recall_at_3
value: 64.851
- type: recall_at_5
value: 75.343
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval
type: mteb/quora
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 72.085
- type: map_at_10
value: 86.107
- type: map_at_100
value: 86.727
- type: map_at_1000
value: 86.74
- type: map_at_3
value: 83.21
- type: map_at_5
value: 85.06
- type: mrr_at_1
value: 82.94
- type: mrr_at_10
value: 88.845
- type: mrr_at_100
value: 88.926
- type: mrr_at_1000
value: 88.927
- type: mrr_at_3
value: 87.993
- type: mrr_at_5
value: 88.62299999999999
- type: ndcg_at_1
value: 82.97
- type: ndcg_at_10
value: 89.645
- type: ndcg_at_100
value: 90.717
- type: ndcg_at_1000
value: 90.78
- type: ndcg_at_3
value: 86.99900000000001
- type: ndcg_at_5
value: 88.52600000000001
- type: precision_at_1
value: 82.97
- type: precision_at_10
value: 13.569
- type: precision_at_100
value: 1.539
- type: precision_at_1000
value: 0.157
- type: precision_at_3
value: 38.043
- type: precision_at_5
value: 24.992
- type: recall_at_1
value: 72.085
- type: recall_at_10
value: 96.262
- type: recall_at_100
value: 99.77000000000001
- type: recall_at_1000
value: 99.997
- type: recall_at_3
value: 88.652
- type: recall_at_5
value: 93.01899999999999
- task:
type: Clustering
dataset:
name: MTEB RedditClustering
type: mteb/reddit-clustering
config: default
split: test
revision: 24640382cdbf8abc73003fb0fa6d111a705499eb
metrics:
- type: v_measure
value: 55.82153952668092
- task:
type: Clustering
dataset:
name: MTEB RedditClusteringP2P
type: mteb/reddit-clustering-p2p
config: default
split: test
revision: 282350215ef01743dc01b456c7f5241fa8937f16
metrics:
- type: v_measure
value: 62.094465801879295
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS
type: mteb/scidocs
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 5.688
- type: map_at_10
value: 15.201999999999998
- type: map_at_100
value: 18.096
- type: map_at_1000
value: 18.481
- type: map_at_3
value: 10.734
- type: map_at_5
value: 12.94
- type: mrr_at_1
value: 28.000000000000004
- type: mrr_at_10
value: 41.101
- type: mrr_at_100
value: 42.202
- type: mrr_at_1000
value: 42.228
- type: mrr_at_3
value: 37.683
- type: mrr_at_5
value: 39.708
- type: ndcg_at_1
value: 28.000000000000004
- type: ndcg_at_10
value: 24.976000000000003
- type: ndcg_at_100
value: 35.129
- type: ndcg_at_1000
value: 40.77
- type: ndcg_at_3
value: 23.787
- type: ndcg_at_5
value: 20.816000000000003
- type: precision_at_1
value: 28.000000000000004
- type: precision_at_10
value: 13.04
- type: precision_at_100
value: 2.761
- type: precision_at_1000
value: 0.41000000000000003
- type: precision_at_3
value: 22.6
- type: precision_at_5
value: 18.52
- type: recall_at_1
value: 5.688
- type: recall_at_10
value: 26.43
- type: recall_at_100
value: 56.02
- type: recall_at_1000
value: 83.21
- type: recall_at_3
value: 13.752
- type: recall_at_5
value: 18.777
- task:
type: STS
dataset:
name: MTEB SICK-R
type: mteb/sickr-sts
config: default
split: test
revision: a6ea5a8cab320b040a23452cc28066d9beae2cee
metrics:
- type: cos_sim_pearson
value: 85.15084859283178
- type: cos_sim_spearman
value: 80.49030614009419
- type: euclidean_pearson
value: 81.84574978672468
- type: euclidean_spearman
value: 79.89787150656818
- type: manhattan_pearson
value: 81.63076538567131
- type: manhattan_spearman
value: 79.69867352121841
- task:
type: STS
dataset:
name: MTEB STS12
type: mteb/sts12-sts
config: default
split: test
revision: a0d554a64d88156834ff5ae9920b964011b16384
metrics:
- type: cos_sim_pearson
value: 84.64097921490992
- type: cos_sim_spearman
value: 77.25370084896514
- type: euclidean_pearson
value: 82.71210826468788
- type: euclidean_spearman
value: 78.50445584994826
- type: manhattan_pearson
value: 82.92580164330298
- type: manhattan_spearman
value: 78.69686891301019
- task:
type: STS
dataset:
name: MTEB STS13
type: mteb/sts13-sts
config: default
split: test
revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
metrics:
- type: cos_sim_pearson
value: 87.24596417308994
- type: cos_sim_spearman
value: 87.79454220555091
- type: euclidean_pearson
value: 87.40242561671164
- type: euclidean_spearman
value: 88.25955597373556
- type: manhattan_pearson
value: 87.25160240485849
- type: manhattan_spearman
value: 88.155794979818
- task:
type: STS
dataset:
name: MTEB STS14
type: mteb/sts14-sts
config: default
split: test
revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
metrics:
- type: cos_sim_pearson
value: 84.44914233422564
- type: cos_sim_spearman
value: 82.91015471820322
- type: euclidean_pearson
value: 84.7206656630327
- type: euclidean_spearman
value: 83.86408872059216
- type: manhattan_pearson
value: 84.72816725158454
- type: manhattan_spearman
value: 84.01603388572788
- task:
type: STS
dataset:
name: MTEB STS15
type: mteb/sts15-sts
config: default
split: test
revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
metrics:
- type: cos_sim_pearson
value: 87.6168026237477
- type: cos_sim_spearman
value: 88.45414278092397
- type: euclidean_pearson
value: 88.57023240882022
- type: euclidean_spearman
value: 89.04102190922094
- type: manhattan_pearson
value: 88.66695535796354
- type: manhattan_spearman
value: 89.19898476680969
- task:
type: STS
dataset:
name: MTEB STS16
type: mteb/sts16-sts
config: default
split: test
revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
metrics:
- type: cos_sim_pearson
value: 84.27925826089424
- type: cos_sim_spearman
value: 85.45291099550461
- type: euclidean_pearson
value: 83.63853036580834
- type: euclidean_spearman
value: 84.33468035821484
- type: manhattan_pearson
value: 83.72778773251596
- type: manhattan_spearman
value: 84.51583132445376
- task:
type: STS
dataset:
name: MTEB STS17 (en-en)
type: mteb/sts17-crosslingual-sts
config: en-en
split: test
revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d
metrics:
- type: cos_sim_pearson
value: 89.67375185692552
- type: cos_sim_spearman
value: 90.32542469203855
- type: euclidean_pearson
value: 89.63513717951847
- type: euclidean_spearman
value: 89.87760271003745
- type: manhattan_pearson
value: 89.28381452982924
- type: manhattan_spearman
value: 89.53568197785721
- task:
type: STS
dataset:
name: MTEB STS22 (en)
type: mteb/sts22-crosslingual-sts
config: en
split: test
revision: eea2b4fe26a775864c896887d910b76a8098ad3f
metrics:
- type: cos_sim_pearson
value: 66.24644693819846
- type: cos_sim_spearman
value: 66.09889420525377
- type: euclidean_pearson
value: 63.72551583520747
- type: euclidean_spearman
value: 63.01385470780679
- type: manhattan_pearson
value: 64.09258157214097
- type: manhattan_spearman
value: 63.080517752822594
- task:
type: STS
dataset:
name: MTEB STSBenchmark
type: mteb/stsbenchmark-sts
config: default
split: test
revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831
metrics:
- type: cos_sim_pearson
value: 86.27321463839989
- type: cos_sim_spearman
value: 86.37572865993327
- type: euclidean_pearson
value: 86.36268020198149
- type: euclidean_spearman
value: 86.31089339478922
- type: manhattan_pearson
value: 86.4260445761947
- type: manhattan_spearman
value: 86.45885895320457
- task:
type: Reranking
dataset:
name: MTEB SciDocsRR
type: mteb/scidocs-reranking
config: default
split: test
revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab
metrics:
- type: map
value: 86.52456702387798
- type: mrr
value: 96.34556529164372
- task:
type: Retrieval
dataset:
name: MTEB SciFact
type: mteb/scifact
config: default
split: test
revision: 0228b52cf27578f30900b9e5271d331663a030d7
metrics:
- type: map_at_1
value: 61.99400000000001
- type: map_at_10
value: 73.38799999999999
- type: map_at_100
value: 73.747
- type: map_at_1000
value: 73.75
- type: map_at_3
value: 70.04599999999999
- type: map_at_5
value: 72.095
- type: mrr_at_1
value: 65.0
- type: mrr_at_10
value: 74.42800000000001
- type: mrr_at_100
value: 74.722
- type: mrr_at_1000
value: 74.725
- type: mrr_at_3
value: 72.056
- type: mrr_at_5
value: 73.60600000000001
- type: ndcg_at_1
value: 65.0
- type: ndcg_at_10
value: 78.435
- type: ndcg_at_100
value: 79.922
- type: ndcg_at_1000
value: 80.00500000000001
- type: ndcg_at_3
value: 73.05199999999999
- type: ndcg_at_5
value: 75.98
- type: precision_at_1
value: 65.0
- type: precision_at_10
value: 10.5
- type: precision_at_100
value: 1.123
- type: precision_at_1000
value: 0.11299999999999999
- type: precision_at_3
value: 28.555999999999997
- type: precision_at_5
value: 19.0
- type: recall_at_1
value: 61.99400000000001
- type: recall_at_10
value: 92.72200000000001
- type: recall_at_100
value: 99.333
- type: recall_at_1000
value: 100.0
- type: recall_at_3
value: 78.739
- type: recall_at_5
value: 85.828
- task:
type: PairClassification
dataset:
name: MTEB SprintDuplicateQuestions
type: mteb/sprintduplicatequestions-pairclassification
config: default
split: test
revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
metrics:
- type: cos_sim_accuracy
value: 99.79009900990098
- type: cos_sim_ap
value: 95.3203137438653
- type: cos_sim_f1
value: 89.12386706948641
- type: cos_sim_precision
value: 89.75659229208925
- type: cos_sim_recall
value: 88.5
- type: dot_accuracy
value: 99.67821782178218
- type: dot_ap
value: 89.94069840000675
- type: dot_f1
value: 83.45902463549521
- type: dot_precision
value: 83.9231547017189
- type: dot_recall
value: 83.0
- type: euclidean_accuracy
value: 99.78613861386138
- type: euclidean_ap
value: 95.10648259135526
- type: euclidean_f1
value: 88.77338877338877
- type: euclidean_precision
value: 92.42424242424242
- type: euclidean_recall
value: 85.39999999999999
- type: manhattan_accuracy
value: 99.7950495049505
- type: manhattan_ap
value: 95.29987661320946
- type: manhattan_f1
value: 89.21313183949972
- type: manhattan_precision
value: 93.14472252448314
- type: manhattan_recall
value: 85.6
- type: max_accuracy
value: 99.7950495049505
- type: max_ap
value: 95.3203137438653
- type: max_f1
value: 89.21313183949972
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClustering
type: mteb/stackexchange-clustering
config: default
split: test
revision: 6cbc1f7b2bc0622f2e39d2c77fa502909748c259
metrics:
- type: v_measure
value: 67.65446577183913
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClusteringP2P
type: mteb/stackexchange-clustering-p2p
config: default
split: test
revision: 815ca46b2622cec33ccafc3735d572c266efdb44
metrics:
- type: v_measure
value: 46.30749237193961
- task:
type: Reranking
dataset:
name: MTEB StackOverflowDupQuestions
type: mteb/stackoverflowdupquestions-reranking
config: default
split: test
revision: e185fbe320c72810689fc5848eb6114e1ef5ec69
metrics:
- type: map
value: 54.91481849959949
- type: mrr
value: 55.853506175197346
- task:
type: Summarization
dataset:
name: MTEB SummEval
type: mteb/summeval
config: default
split: test
revision: cda12ad7615edc362dbf25a00fdd61d3b1eaf93c
metrics:
- type: cos_sim_pearson
value: 30.08196549170419
- type: cos_sim_spearman
value: 31.16661390597077
- type: dot_pearson
value: 29.892258410943466
- type: dot_spearman
value: 30.51328811965085
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID
type: mteb/trec-covid
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 0.23900000000000002
- type: map_at_10
value: 2.173
- type: map_at_100
value: 14.24
- type: map_at_1000
value: 35.309000000000005
- type: map_at_3
value: 0.7100000000000001
- type: map_at_5
value: 1.163
- type: mrr_at_1
value: 92.0
- type: mrr_at_10
value: 96.0
- type: mrr_at_100
value: 96.0
- type: mrr_at_1000
value: 96.0
- type: mrr_at_3
value: 96.0
- type: mrr_at_5
value: 96.0
- type: ndcg_at_1
value: 90.0
- type: ndcg_at_10
value: 85.382
- type: ndcg_at_100
value: 68.03
- type: ndcg_at_1000
value: 61.021
- type: ndcg_at_3
value: 89.765
- type: ndcg_at_5
value: 88.444
- type: precision_at_1
value: 92.0
- type: precision_at_10
value: 88.0
- type: precision_at_100
value: 70.02000000000001
- type: precision_at_1000
value: 26.984
- type: precision_at_3
value: 94.0
- type: precision_at_5
value: 92.80000000000001
- type: recall_at_1
value: 0.23900000000000002
- type: recall_at_10
value: 2.313
- type: recall_at_100
value: 17.049
- type: recall_at_1000
value: 57.489999999999995
- type: recall_at_3
value: 0.737
- type: recall_at_5
value: 1.221
- task:
type: Retrieval
dataset:
name: MTEB Touche2020
type: mteb/touche2020
config: default
split: test
revision: a34f9a33db75fa0cbb21bb5cfc3dae8dc8bec93f
metrics:
- type: map_at_1
value: 2.75
- type: map_at_10
value: 11.29
- type: map_at_100
value: 18.032999999999998
- type: map_at_1000
value: 19.746
- type: map_at_3
value: 6.555
- type: map_at_5
value: 8.706999999999999
- type: mrr_at_1
value: 34.694
- type: mrr_at_10
value: 50.55
- type: mrr_at_100
value: 51.659
- type: mrr_at_1000
value: 51.659
- type: mrr_at_3
value: 47.278999999999996
- type: mrr_at_5
value: 49.728
- type: ndcg_at_1
value: 32.653
- type: ndcg_at_10
value: 27.894000000000002
- type: ndcg_at_100
value: 39.769
- type: ndcg_at_1000
value: 51.495999999999995
- type: ndcg_at_3
value: 32.954
- type: ndcg_at_5
value: 31.502999999999997
- type: precision_at_1
value: 34.694
- type: precision_at_10
value: 23.265
- type: precision_at_100
value: 7.898
- type: precision_at_1000
value: 1.58
- type: precision_at_3
value: 34.694
- type: precision_at_5
value: 31.429000000000002
- type: recall_at_1
value: 2.75
- type: recall_at_10
value: 16.953
- type: recall_at_100
value: 48.68
- type: recall_at_1000
value: 85.18599999999999
- type: recall_at_3
value: 7.710999999999999
- type: recall_at_5
value: 11.484
- task:
type: Classification
dataset:
name: MTEB ToxicConversationsClassification
type: mteb/toxic_conversations_50k
config: default
split: test
revision: d7c0de2777da35d6aae2200a62c6e0e5af397c4c
metrics:
- type: accuracy
value: 82.66099999999999
- type: ap
value: 25.555698090238337
- type: f1
value: 66.48402012461622
- task:
type: Classification
dataset:
name: MTEB TweetSentimentExtractionClassification
type: mteb/tweet_sentiment_extraction
config: default
split: test
revision: d604517c81ca91fe16a244d1248fc021f9ecee7a
metrics:
- type: accuracy
value: 72.94567062818335
- type: f1
value: 73.28139189595674
- task:
type: Clustering
dataset:
name: MTEB TwentyNewsgroupsClustering
type: mteb/twentynewsgroups-clustering
config: default
split: test
revision: 6125ec4e24fa026cec8a478383ee943acfbd5449
metrics:
- type: v_measure
value: 49.581627240203474
- task:
type: PairClassification
dataset:
name: MTEB TwitterSemEval2015
type: mteb/twittersemeval2015-pairclassification
config: default
split: test
revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1
metrics:
- type: cos_sim_accuracy
value: 87.78089050485785
- type: cos_sim_ap
value: 79.64487116574168
- type: cos_sim_f1
value: 72.46563021970964
- type: cos_sim_precision
value: 70.62359128474831
- type: cos_sim_recall
value: 74.40633245382587
- type: dot_accuracy
value: 86.2609524944865
- type: dot_ap
value: 75.513046857613
- type: dot_f1
value: 68.58213616489695
- type: dot_precision
value: 65.12455516014235
- type: dot_recall
value: 72.42744063324538
- type: euclidean_accuracy
value: 87.6080348095607
- type: euclidean_ap
value: 79.00204933649795
- type: euclidean_f1
value: 72.14495342605589
- type: euclidean_precision
value: 69.85421299728193
- type: euclidean_recall
value: 74.5910290237467
- type: manhattan_accuracy
value: 87.59611372712642
- type: manhattan_ap
value: 78.78523756706264
- type: manhattan_f1
value: 71.86499137718648
- type: manhattan_precision
value: 67.39833641404806
- type: manhattan_recall
value: 76.96569920844327
- type: max_accuracy
value: 87.78089050485785
- type: max_ap
value: 79.64487116574168
- type: max_f1
value: 72.46563021970964
- task:
type: PairClassification
dataset:
name: MTEB TwitterURLCorpus
type: mteb/twitterurlcorpus-pairclassification
config: default
split: test
revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf
metrics:
- type: cos_sim_accuracy
value: 89.98719292117825
- type: cos_sim_ap
value: 87.58146137353202
- type: cos_sim_f1
value: 80.28543232369239
- type: cos_sim_precision
value: 79.1735289714029
- type: cos_sim_recall
value: 81.42901139513397
- type: dot_accuracy
value: 88.9199363526992
- type: dot_ap
value: 84.98499998630417
- type: dot_f1
value: 78.21951400757969
- type: dot_precision
value: 75.58523624874336
- type: dot_recall
value: 81.04404065291038
- type: euclidean_accuracy
value: 89.77374160748244
- type: euclidean_ap
value: 87.35151562835209
- type: euclidean_f1
value: 79.92160922940393
- type: euclidean_precision
value: 76.88531587933979
- type: euclidean_recall
value: 83.20757622420696
- type: manhattan_accuracy
value: 89.72717041176699
- type: manhattan_ap
value: 87.34065592142515
- type: manhattan_f1
value: 79.85603419187943
- type: manhattan_precision
value: 77.82243332115455
- type: manhattan_recall
value: 81.99876809362489
- type: max_accuracy
value: 89.98719292117825
- type: max_ap
value: 87.58146137353202
- type: max_f1
value: 80.28543232369239
- task:
type: STS
dataset:
name: MTEB AFQMC
type: C-MTEB/AFQMC
config: default
split: validation
revision: b44c3b011063adb25877c13823db83bb193913c4
metrics:
- type: cos_sim_pearson
value: 53.45954203592337
- type: cos_sim_spearman
value: 58.42154680418638
- type: euclidean_pearson
value: 56.41543791722753
- type: euclidean_spearman
value: 58.39328016640146
- type: manhattan_pearson
value: 56.318510356833876
- type: manhattan_spearman
value: 58.28423447818184
- task:
type: STS
dataset:
name: MTEB ATEC
type: C-MTEB/ATEC
config: default
split: test
revision: 0f319b1142f28d00e055a6770f3f726ae9b7d865
metrics:
- type: cos_sim_pearson
value: 50.78356460675945
- type: cos_sim_spearman
value: 55.6530411663269
- type: euclidean_pearson
value: 56.50763660417816
- type: euclidean_spearman
value: 55.733823335669065
- type: manhattan_pearson
value: 56.45323093512866
- type: manhattan_spearman
value: 55.63248619032702
- task:
type: Classification
dataset:
name: MTEB AmazonReviewsClassification (zh)
type: mteb/amazon_reviews_multi
config: zh
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 47.209999999999994
- type: f1
value: 46.08892432018655
- task:
type: STS
dataset:
name: MTEB BQ
type: C-MTEB/BQ
config: default
split: test
revision: e3dda5e115e487b39ec7e618c0c6a29137052a55
metrics:
- type: cos_sim_pearson
value: 70.25573992001478
- type: cos_sim_spearman
value: 73.85247134951433
- type: euclidean_pearson
value: 72.60033082168442
- type: euclidean_spearman
value: 73.72445893756499
- type: manhattan_pearson
value: 72.59932284620231
- type: manhattan_spearman
value: 73.68002490614583
- task:
type: Clustering
dataset:
name: MTEB CLSClusteringP2P
type: C-MTEB/CLSClusteringP2P
config: default
split: test
revision: 4b6227591c6c1a73bc76b1055f3b7f3588e72476
metrics:
- type: v_measure
value: 45.21317724305628
- task:
type: Clustering
dataset:
name: MTEB CLSClusteringS2S
type: C-MTEB/CLSClusteringS2S
config: default
split: test
revision: e458b3f5414b62b7f9f83499ac1f5497ae2e869f
metrics:
- type: v_measure
value: 42.49825170976724
- task:
type: Reranking
dataset:
name: MTEB CMedQAv1
type: C-MTEB/CMedQAv1-reranking
config: default
split: test
revision: 8d7f1e942507dac42dc58017c1a001c3717da7df
metrics:
- type: map
value: 88.15661686810597
- type: mrr
value: 90.11222222222223
- task:
type: Reranking
dataset:
name: MTEB CMedQAv2
type: C-MTEB/CMedQAv2-reranking
config: default
split: test
revision: 23d186750531a14a0357ca22cd92d712fd512ea0
metrics:
- type: map
value: 88.1204726064383
- type: mrr
value: 90.20142857142858
- task:
type: Retrieval
dataset:
name: MTEB CmedqaRetrieval
type: C-MTEB/CmedqaRetrieval
config: default
split: dev
revision: cd540c506dae1cf9e9a59c3e06f42030d54e7301
metrics:
- type: map_at_1
value: 27.224999999999998
- type: map_at_10
value: 40.169
- type: map_at_100
value: 42.0
- type: map_at_1000
value: 42.109
- type: map_at_3
value: 35.76
- type: map_at_5
value: 38.221
- type: mrr_at_1
value: 40.56
- type: mrr_at_10
value: 49.118
- type: mrr_at_100
value: 50.092999999999996
- type: mrr_at_1000
value: 50.133
- type: mrr_at_3
value: 46.507
- type: mrr_at_5
value: 47.973
- type: ndcg_at_1
value: 40.56
- type: ndcg_at_10
value: 46.972
- type: ndcg_at_100
value: 54.04
- type: ndcg_at_1000
value: 55.862
- type: ndcg_at_3
value: 41.36
- type: ndcg_at_5
value: 43.704
- type: precision_at_1
value: 40.56
- type: precision_at_10
value: 10.302999999999999
- type: precision_at_100
value: 1.606
- type: precision_at_1000
value: 0.184
- type: precision_at_3
value: 23.064
- type: precision_at_5
value: 16.764000000000003
- type: recall_at_1
value: 27.224999999999998
- type: recall_at_10
value: 58.05200000000001
- type: recall_at_100
value: 87.092
- type: recall_at_1000
value: 99.099
- type: recall_at_3
value: 41.373
- type: recall_at_5
value: 48.453
- task:
type: PairClassification
dataset:
name: MTEB Cmnli
type: C-MTEB/CMNLI
config: default
split: validation
revision: 41bc36f332156f7adc9e38f53777c959b2ae9766
metrics:
- type: cos_sim_accuracy
value: 77.40228502705953
- type: cos_sim_ap
value: 86.22359172956327
- type: cos_sim_f1
value: 78.96328293736501
- type: cos_sim_precision
value: 73.36945615091311
- type: cos_sim_recall
value: 85.48047696983868
- type: dot_accuracy
value: 75.53818400481059
- type: dot_ap
value: 83.70164011305312
- type: dot_f1
value: 77.67298719348754
- type: dot_precision
value: 67.49482401656314
- type: dot_recall
value: 91.46598082768296
- type: euclidean_accuracy
value: 77.94347564642213
- type: euclidean_ap
value: 86.4652108728609
- type: euclidean_f1
value: 79.15555555555555
- type: euclidean_precision
value: 75.41816641964853
- type: euclidean_recall
value: 83.28267477203647
- type: manhattan_accuracy
value: 77.45039085989175
- type: manhattan_ap
value: 86.09986583900665
- type: manhattan_f1
value: 78.93669264438988
- type: manhattan_precision
value: 72.63261296660117
- type: manhattan_recall
value: 86.43909282207154
- type: max_accuracy
value: 77.94347564642213
- type: max_ap
value: 86.4652108728609
- type: max_f1
value: 79.15555555555555
- task:
type: Retrieval
dataset:
name: MTEB CovidRetrieval
type: C-MTEB/CovidRetrieval
config: default
split: dev
revision: 1271c7809071a13532e05f25fb53511ffce77117
metrics:
- type: map_at_1
value: 69.336
- type: map_at_10
value: 77.16
- type: map_at_100
value: 77.47500000000001
- type: map_at_1000
value: 77.482
- type: map_at_3
value: 75.42999999999999
- type: map_at_5
value: 76.468
- type: mrr_at_1
value: 69.44200000000001
- type: mrr_at_10
value: 77.132
- type: mrr_at_100
value: 77.43299999999999
- type: mrr_at_1000
value: 77.44
- type: mrr_at_3
value: 75.395
- type: mrr_at_5
value: 76.459
- type: ndcg_at_1
value: 69.547
- type: ndcg_at_10
value: 80.794
- type: ndcg_at_100
value: 82.245
- type: ndcg_at_1000
value: 82.40899999999999
- type: ndcg_at_3
value: 77.303
- type: ndcg_at_5
value: 79.168
- type: precision_at_1
value: 69.547
- type: precision_at_10
value: 9.305
- type: precision_at_100
value: 0.9979999999999999
- type: precision_at_1000
value: 0.101
- type: precision_at_3
value: 27.749000000000002
- type: precision_at_5
value: 17.576
- type: recall_at_1
value: 69.336
- type: recall_at_10
value: 92.097
- type: recall_at_100
value: 98.736
- type: recall_at_1000
value: 100.0
- type: recall_at_3
value: 82.64
- type: recall_at_5
value: 87.144
- task:
type: Retrieval
dataset:
name: MTEB DuRetrieval
type: C-MTEB/DuRetrieval
config: default
split: dev
revision: a1a333e290fe30b10f3f56498e3a0d911a693ced
metrics:
- type: map_at_1
value: 26.817999999999998
- type: map_at_10
value: 82.67
- type: map_at_100
value: 85.304
- type: map_at_1000
value: 85.334
- type: map_at_3
value: 57.336
- type: map_at_5
value: 72.474
- type: mrr_at_1
value: 91.45
- type: mrr_at_10
value: 94.272
- type: mrr_at_100
value: 94.318
- type: mrr_at_1000
value: 94.32000000000001
- type: mrr_at_3
value: 94.0
- type: mrr_at_5
value: 94.17699999999999
- type: ndcg_at_1
value: 91.45
- type: ndcg_at_10
value: 89.404
- type: ndcg_at_100
value: 91.724
- type: ndcg_at_1000
value: 91.973
- type: ndcg_at_3
value: 88.104
- type: ndcg_at_5
value: 87.25699999999999
- type: precision_at_1
value: 91.45
- type: precision_at_10
value: 42.585
- type: precision_at_100
value: 4.838
- type: precision_at_1000
value: 0.49
- type: precision_at_3
value: 78.8
- type: precision_at_5
value: 66.66
- type: recall_at_1
value: 26.817999999999998
- type: recall_at_10
value: 90.67
- type: recall_at_100
value: 98.36200000000001
- type: recall_at_1000
value: 99.583
- type: recall_at_3
value: 59.614999999999995
- type: recall_at_5
value: 77.05199999999999
- task:
type: Retrieval
dataset:
name: MTEB EcomRetrieval
type: C-MTEB/EcomRetrieval
config: default
split: dev
revision: 687de13dc7294d6fd9be10c6945f9e8fec8166b9
metrics:
- type: map_at_1
value: 47.699999999999996
- type: map_at_10
value: 57.589999999999996
- type: map_at_100
value: 58.226
- type: map_at_1000
value: 58.251
- type: map_at_3
value: 55.233
- type: map_at_5
value: 56.633
- type: mrr_at_1
value: 47.699999999999996
- type: mrr_at_10
value: 57.589999999999996
- type: mrr_at_100
value: 58.226
- type: mrr_at_1000
value: 58.251
- type: mrr_at_3
value: 55.233
- type: mrr_at_5
value: 56.633
- type: ndcg_at_1
value: 47.699999999999996
- type: ndcg_at_10
value: 62.505
- type: ndcg_at_100
value: 65.517
- type: ndcg_at_1000
value: 66.19800000000001
- type: ndcg_at_3
value: 57.643
- type: ndcg_at_5
value: 60.181
- type: precision_at_1
value: 47.699999999999996
- type: precision_at_10
value: 7.8
- type: precision_at_100
value: 0.919
- type: precision_at_1000
value: 0.097
- type: precision_at_3
value: 21.532999999999998
- type: precision_at_5
value: 14.16
- type: recall_at_1
value: 47.699999999999996
- type: recall_at_10
value: 78.0
- type: recall_at_100
value: 91.9
- type: recall_at_1000
value: 97.3
- type: recall_at_3
value: 64.60000000000001
- type: recall_at_5
value: 70.8
- task:
type: Classification
dataset:
name: MTEB IFlyTek
type: C-MTEB/IFlyTek-classification
config: default
split: validation
revision: 421605374b29664c5fc098418fe20ada9bd55f8a
metrics:
- type: accuracy
value: 44.84801846864178
- type: f1
value: 37.47347897956339
- task:
type: Classification
dataset:
name: MTEB JDReview
type: C-MTEB/JDReview-classification
config: default
split: test
revision: b7c64bd89eb87f8ded463478346f76731f07bf8b
metrics:
- type: accuracy
value: 85.81613508442777
- type: ap
value: 52.68244615477374
- type: f1
value: 80.0445640948843
- task:
type: STS
dataset:
name: MTEB LCQMC
type: C-MTEB/LCQMC
config: default
split: test
revision: 17f9b096f80380fce5ed12a9be8be7784b337daf
metrics:
- type: cos_sim_pearson
value: 69.57786502217138
- type: cos_sim_spearman
value: 75.39106054489906
- type: euclidean_pearson
value: 73.72082954602402
- type: euclidean_spearman
value: 75.14421475913619
- type: manhattan_pearson
value: 73.62463076633642
- type: manhattan_spearman
value: 75.01301565104112
- task:
type: Reranking
dataset:
name: MTEB MMarcoReranking
type: C-MTEB/Mmarco-reranking
config: default
split: dev
revision: None
metrics:
- type: map
value: 29.143797057999134
- type: mrr
value: 28.08174603174603
- task:
type: Retrieval
dataset:
name: MTEB MMarcoRetrieval
type: C-MTEB/MMarcoRetrieval
config: default
split: dev
revision: 539bbde593d947e2a124ba72651aafc09eb33fc2
metrics:
- type: map_at_1
value: 70.492
- type: map_at_10
value: 79.501
- type: map_at_100
value: 79.728
- type: map_at_1000
value: 79.735
- type: map_at_3
value: 77.77
- type: map_at_5
value: 78.851
- type: mrr_at_1
value: 72.822
- type: mrr_at_10
value: 80.001
- type: mrr_at_100
value: 80.19
- type: mrr_at_1000
value: 80.197
- type: mrr_at_3
value: 78.484
- type: mrr_at_5
value: 79.42099999999999
- type: ndcg_at_1
value: 72.822
- type: ndcg_at_10
value: 83.013
- type: ndcg_at_100
value: 84.013
- type: ndcg_at_1000
value: 84.20400000000001
- type: ndcg_at_3
value: 79.728
- type: ndcg_at_5
value: 81.542
- type: precision_at_1
value: 72.822
- type: precision_at_10
value: 9.917
- type: precision_at_100
value: 1.042
- type: precision_at_1000
value: 0.106
- type: precision_at_3
value: 29.847
- type: precision_at_5
value: 18.871
- type: recall_at_1
value: 70.492
- type: recall_at_10
value: 93.325
- type: recall_at_100
value: 97.822
- type: recall_at_1000
value: 99.319
- type: recall_at_3
value: 84.636
- type: recall_at_5
value: 88.93100000000001
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (zh-CN)
type: mteb/amazon_massive_intent
config: zh-CN
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 76.88298587760592
- type: f1
value: 73.89001762017176
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (zh-CN)
type: mteb/amazon_massive_scenario
config: zh-CN
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 80.76328177538669
- type: f1
value: 80.24718532423358
- task:
type: Retrieval
dataset:
name: MTEB MedicalRetrieval
type: C-MTEB/MedicalRetrieval
config: default
split: dev
revision: 2039188fb5800a9803ba5048df7b76e6fb151fc6
metrics:
- type: map_at_1
value: 49.6
- type: map_at_10
value: 55.620999999999995
- type: map_at_100
value: 56.204
- type: map_at_1000
value: 56.251
- type: map_at_3
value: 54.132999999999996
- type: map_at_5
value: 54.933
- type: mrr_at_1
value: 49.7
- type: mrr_at_10
value: 55.67100000000001
- type: mrr_at_100
value: 56.254000000000005
- type: mrr_at_1000
value: 56.301
- type: mrr_at_3
value: 54.18300000000001
- type: mrr_at_5
value: 54.983000000000004
- type: ndcg_at_1
value: 49.6
- type: ndcg_at_10
value: 58.645
- type: ndcg_at_100
value: 61.789
- type: ndcg_at_1000
value: 63.219
- type: ndcg_at_3
value: 55.567
- type: ndcg_at_5
value: 57.008
- type: precision_at_1
value: 49.6
- type: precision_at_10
value: 6.819999999999999
- type: precision_at_100
value: 0.836
- type: precision_at_1000
value: 0.095
- type: precision_at_3
value: 19.900000000000002
- type: precision_at_5
value: 12.64
- type: recall_at_1
value: 49.6
- type: recall_at_10
value: 68.2
- type: recall_at_100
value: 83.6
- type: recall_at_1000
value: 95.3
- type: recall_at_3
value: 59.699999999999996
- type: recall_at_5
value: 63.2
- task:
type: Classification
dataset:
name: MTEB MultilingualSentiment
type: C-MTEB/MultilingualSentiment-classification
config: default
split: validation
revision: 46958b007a63fdbf239b7672c25d0bea67b5ea1a
metrics:
- type: accuracy
value: 74.45666666666666
- type: f1
value: 74.32582402190089
- task:
type: PairClassification
dataset:
name: MTEB Ocnli
type: C-MTEB/OCNLI
config: default
split: validation
revision: 66e76a618a34d6d565d5538088562851e6daa7ec
metrics:
- type: cos_sim_accuracy
value: 80.67135896047645
- type: cos_sim_ap
value: 87.60421240712051
- type: cos_sim_f1
value: 82.1304131408661
- type: cos_sim_precision
value: 77.68361581920904
- type: cos_sim_recall
value: 87.11721224920802
- type: dot_accuracy
value: 79.04710341093666
- type: dot_ap
value: 85.6370059719336
- type: dot_f1
value: 80.763723150358
- type: dot_precision
value: 73.69337979094077
- type: dot_recall
value: 89.33474128827878
- type: euclidean_accuracy
value: 81.05035192203573
- type: euclidean_ap
value: 87.7880240053663
- type: euclidean_f1
value: 82.50244379276637
- type: euclidean_precision
value: 76.7970882620564
- type: euclidean_recall
value: 89.1235480464625
- type: manhattan_accuracy
value: 80.61721710882512
- type: manhattan_ap
value: 87.43568120591175
- type: manhattan_f1
value: 81.89526184538653
- type: manhattan_precision
value: 77.5992438563327
- type: manhattan_recall
value: 86.6948257655755
- type: max_accuracy
value: 81.05035192203573
- type: max_ap
value: 87.7880240053663
- type: max_f1
value: 82.50244379276637
- task:
type: Classification
dataset:
name: MTEB OnlineShopping
type: C-MTEB/OnlineShopping-classification
config: default
split: test
revision: e610f2ebd179a8fda30ae534c3878750a96db120
metrics:
- type: accuracy
value: 93.5
- type: ap
value: 91.31357903446782
- type: f1
value: 93.48088994006616
- task:
type: STS
dataset:
name: MTEB PAWSX
type: C-MTEB/PAWSX
config: default
split: test
revision: 9c6a90e430ac22b5779fb019a23e820b11a8b5e1
metrics:
- type: cos_sim_pearson
value: 36.93293453538077
- type: cos_sim_spearman
value: 42.45972506308574
- type: euclidean_pearson
value: 42.34945133152159
- type: euclidean_spearman
value: 42.331610303674644
- type: manhattan_pearson
value: 42.31455070249498
- type: manhattan_spearman
value: 42.19887982891834
- task:
type: STS
dataset:
name: MTEB QBQTC
type: C-MTEB/QBQTC
config: default
split: test
revision: 790b0510dc52b1553e8c49f3d2afb48c0e5c48b7
metrics:
- type: cos_sim_pearson
value: 33.683290790043785
- type: cos_sim_spearman
value: 35.149171171202994
- type: euclidean_pearson
value: 32.33806561267862
- type: euclidean_spearman
value: 34.483576387347966
- type: manhattan_pearson
value: 32.47629754599608
- type: manhattan_spearman
value: 34.66434471867615
- task:
type: STS
dataset:
name: MTEB STS22 (zh)
type: mteb/sts22-crosslingual-sts
config: zh
split: test
revision: eea2b4fe26a775864c896887d910b76a8098ad3f
metrics:
- type: cos_sim_pearson
value: 66.46322760516104
- type: cos_sim_spearman
value: 67.398478319726
- type: euclidean_pearson
value: 64.7223480293625
- type: euclidean_spearman
value: 66.83118568812951
- type: manhattan_pearson
value: 64.88440039828305
- type: manhattan_spearman
value: 66.80429458952257
- task:
type: STS
dataset:
name: MTEB STSB
type: C-MTEB/STSB
config: default
split: test
revision: 0cde68302b3541bb8b3c340dc0644b0b745b3dc0
metrics:
- type: cos_sim_pearson
value: 79.08991383232105
- type: cos_sim_spearman
value: 79.39715677296854
- type: euclidean_pearson
value: 78.63201279320496
- type: euclidean_spearman
value: 79.40262660785731
- type: manhattan_pearson
value: 78.98138363146906
- type: manhattan_spearman
value: 79.79968413014194
- task:
type: Reranking
dataset:
name: MTEB T2Reranking
type: C-MTEB/T2Reranking
config: default
split: dev
revision: 76631901a18387f85eaa53e5450019b87ad58ef9
metrics:
- type: map
value: 67.43289278789972
- type: mrr
value: 77.53012460908535
- task:
type: Retrieval
dataset:
name: MTEB T2Retrieval
type: C-MTEB/T2Retrieval
config: default
split: dev
revision: 8731a845f1bf500a4f111cf1070785c793d10e64
metrics:
- type: map_at_1
value: 27.733999999999998
- type: map_at_10
value: 78.24799999999999
- type: map_at_100
value: 81.765
- type: map_at_1000
value: 81.824
- type: map_at_3
value: 54.92
- type: map_at_5
value: 67.61399999999999
- type: mrr_at_1
value: 90.527
- type: mrr_at_10
value: 92.843
- type: mrr_at_100
value: 92.927
- type: mrr_at_1000
value: 92.93
- type: mrr_at_3
value: 92.45100000000001
- type: mrr_at_5
value: 92.693
- type: ndcg_at_1
value: 90.527
- type: ndcg_at_10
value: 85.466
- type: ndcg_at_100
value: 88.846
- type: ndcg_at_1000
value: 89.415
- type: ndcg_at_3
value: 86.768
- type: ndcg_at_5
value: 85.46000000000001
- type: precision_at_1
value: 90.527
- type: precision_at_10
value: 42.488
- type: precision_at_100
value: 5.024
- type: precision_at_1000
value: 0.516
- type: precision_at_3
value: 75.907
- type: precision_at_5
value: 63.727000000000004
- type: recall_at_1
value: 27.733999999999998
- type: recall_at_10
value: 84.346
- type: recall_at_100
value: 95.536
- type: recall_at_1000
value: 98.42999999999999
- type: recall_at_3
value: 56.455
- type: recall_at_5
value: 70.755
- task:
type: Classification
dataset:
name: MTEB TNews
type: C-MTEB/TNews-classification
config: default
split: validation
revision: 317f262bf1e6126357bbe89e875451e4b0938fe4
metrics:
- type: accuracy
value: 49.952000000000005
- type: f1
value: 48.264617195258054
- task:
type: Clustering
dataset:
name: MTEB ThuNewsClusteringP2P
type: C-MTEB/ThuNewsClusteringP2P
config: default
split: test
revision: 5798586b105c0434e4f0fe5e767abe619442cf93
metrics:
- type: v_measure
value: 68.23769904483508
- task:
type: Clustering
dataset:
name: MTEB ThuNewsClusteringS2S
type: C-MTEB/ThuNewsClusteringS2S
config: default
split: test
revision: 8a8b2caeda43f39e13c4bc5bea0f8a667896e10d
metrics:
- type: v_measure
value: 62.50294403136556
- task:
type: Retrieval
dataset:
name: MTEB VideoRetrieval
type: C-MTEB/VideoRetrieval
config: default
split: dev
revision: 58c2597a5943a2ba48f4668c3b90d796283c5639
metrics:
- type: map_at_1
value: 54.0
- type: map_at_10
value: 63.668
- type: map_at_100
value: 64.217
- type: map_at_1000
value: 64.23100000000001
- type: map_at_3
value: 61.7
- type: map_at_5
value: 62.870000000000005
- type: mrr_at_1
value: 54.0
- type: mrr_at_10
value: 63.668
- type: mrr_at_100
value: 64.217
- type: mrr_at_1000
value: 64.23100000000001
- type: mrr_at_3
value: 61.7
- type: mrr_at_5
value: 62.870000000000005
- type: ndcg_at_1
value: 54.0
- type: ndcg_at_10
value: 68.11399999999999
- type: ndcg_at_100
value: 70.723
- type: ndcg_at_1000
value: 71.123
- type: ndcg_at_3
value: 64.074
- type: ndcg_at_5
value: 66.178
- type: precision_at_1
value: 54.0
- type: precision_at_10
value: 8.200000000000001
- type: precision_at_100
value: 0.941
- type: precision_at_1000
value: 0.097
- type: precision_at_3
value: 23.633000000000003
- type: precision_at_5
value: 15.2
- type: recall_at_1
value: 54.0
- type: recall_at_10
value: 82.0
- type: recall_at_100
value: 94.1
- type: recall_at_1000
value: 97.3
- type: recall_at_3
value: 70.89999999999999
- type: recall_at_5
value: 76.0
- task:
type: Classification
dataset:
name: MTEB Waimai
type: C-MTEB/waimai-classification
config: default
split: test
revision: 339287def212450dcaa9df8c22bf93e9980c7023
metrics:
- type: accuracy
value: 86.63000000000001
- type: ap
value: 69.99457882599567
- type: f1
value: 85.07735617998541
- task:
type: Clustering
dataset:
name: MTEB 8TagsClustering
type: PL-MTEB/8tags-clustering
config: default
split: test
revision: None
metrics:
- type: v_measure
value: 44.594104491193555
- task:
type: Classification
dataset:
name: MTEB AllegroReviews
type: PL-MTEB/allegro-reviews
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 63.97614314115309
- type: f1
value: 52.15634261679283
- task:
type: Retrieval
dataset:
name: MTEB ArguAna-PL
type: clarin-knext/arguana-pl
config: default
split: test
revision: 63fc86750af76253e8c760fc9e534bbf24d260a2
metrics:
- type: map_at_1
value: 32.646
- type: map_at_10
value: 47.963
- type: map_at_100
value: 48.789
- type: map_at_1000
value: 48.797000000000004
- type: map_at_3
value: 43.196
- type: map_at_5
value: 46.016
- type: mrr_at_1
value: 33.073
- type: mrr_at_10
value: 48.126000000000005
- type: mrr_at_100
value: 48.946
- type: mrr_at_1000
value: 48.953
- type: mrr_at_3
value: 43.374
- type: mrr_at_5
value: 46.147
- type: ndcg_at_1
value: 32.646
- type: ndcg_at_10
value: 56.481
- type: ndcg_at_100
value: 59.922
- type: ndcg_at_1000
value: 60.07
- type: ndcg_at_3
value: 46.675
- type: ndcg_at_5
value: 51.76500000000001
- type: precision_at_1
value: 32.646
- type: precision_at_10
value: 8.371
- type: precision_at_100
value: 0.9860000000000001
- type: precision_at_1000
value: 0.1
- type: precision_at_3
value: 18.919
- type: precision_at_5
value: 13.825999999999999
- type: recall_at_1
value: 32.646
- type: recall_at_10
value: 83.71300000000001
- type: recall_at_100
value: 98.578
- type: recall_at_1000
value: 99.644
- type: recall_at_3
value: 56.757000000000005
- type: recall_at_5
value: 69.132
- task:
type: Classification
dataset:
name: MTEB CBD
type: PL-MTEB/cbd
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 68.56
- type: ap
value: 23.310493680488513
- type: f1
value: 58.85369533105693
- task:
type: PairClassification
dataset:
name: MTEB CDSC-E
type: PL-MTEB/cdsce-pairclassification
config: default
split: test
revision: None
metrics:
- type: cos_sim_accuracy
value: 88.5
- type: cos_sim_ap
value: 72.42140924378361
- type: cos_sim_f1
value: 66.0919540229885
- type: cos_sim_precision
value: 72.78481012658227
- type: cos_sim_recall
value: 60.526315789473685
- type: dot_accuracy
value: 88.5
- type: dot_ap
value: 72.42140924378361
- type: dot_f1
value: 66.0919540229885
- type: dot_precision
value: 72.78481012658227
- type: dot_recall
value: 60.526315789473685
- type: euclidean_accuracy
value: 88.5
- type: euclidean_ap
value: 72.42140924378361
- type: euclidean_f1
value: 66.0919540229885
- type: euclidean_precision
value: 72.78481012658227
- type: euclidean_recall
value: 60.526315789473685
- type: manhattan_accuracy
value: 88.5
- type: manhattan_ap
value: 72.49745515311696
- type: manhattan_f1
value: 66.0968660968661
- type: manhattan_precision
value: 72.04968944099379
- type: manhattan_recall
value: 61.05263157894737
- type: max_accuracy
value: 88.5
- type: max_ap
value: 72.49745515311696
- type: max_f1
value: 66.0968660968661
- task:
type: STS
dataset:
name: MTEB CDSC-R
type: PL-MTEB/cdscr-sts
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 90.32269765590145
- type: cos_sim_spearman
value: 89.73666311491672
- type: euclidean_pearson
value: 88.2933868516544
- type: euclidean_spearman
value: 89.73666311491672
- type: manhattan_pearson
value: 88.33474590219448
- type: manhattan_spearman
value: 89.8548364866583
- task:
type: Retrieval
dataset:
name: MTEB DBPedia-PL
type: clarin-knext/dbpedia-pl
config: default
split: test
revision: 76afe41d9af165cc40999fcaa92312b8b012064a
metrics:
- type: map_at_1
value: 7.632999999999999
- type: map_at_10
value: 16.426
- type: map_at_100
value: 22.651
- type: map_at_1000
value: 24.372
- type: map_at_3
value: 11.706
- type: map_at_5
value: 13.529
- type: mrr_at_1
value: 60.75000000000001
- type: mrr_at_10
value: 68.613
- type: mrr_at_100
value: 69.001
- type: mrr_at_1000
value: 69.021
- type: mrr_at_3
value: 67.0
- type: mrr_at_5
value: 67.925
- type: ndcg_at_1
value: 49.875
- type: ndcg_at_10
value: 36.978
- type: ndcg_at_100
value: 40.031
- type: ndcg_at_1000
value: 47.566
- type: ndcg_at_3
value: 41.148
- type: ndcg_at_5
value: 38.702
- type: precision_at_1
value: 60.75000000000001
- type: precision_at_10
value: 29.7
- type: precision_at_100
value: 9.278
- type: precision_at_1000
value: 2.099
- type: precision_at_3
value: 44.0
- type: precision_at_5
value: 37.6
- type: recall_at_1
value: 7.632999999999999
- type: recall_at_10
value: 22.040000000000003
- type: recall_at_100
value: 44.024
- type: recall_at_1000
value: 67.848
- type: recall_at_3
value: 13.093
- type: recall_at_5
value: 15.973
- task:
type: Retrieval
dataset:
name: MTEB FiQA-PL
type: clarin-knext/fiqa-pl
config: default
split: test
revision: 2e535829717f8bf9dc829b7f911cc5bbd4e6608e
metrics:
- type: map_at_1
value: 15.473
- type: map_at_10
value: 24.579
- type: map_at_100
value: 26.387
- type: map_at_1000
value: 26.57
- type: map_at_3
value: 21.278
- type: map_at_5
value: 23.179
- type: mrr_at_1
value: 30.709999999999997
- type: mrr_at_10
value: 38.994
- type: mrr_at_100
value: 39.993
- type: mrr_at_1000
value: 40.044999999999995
- type: mrr_at_3
value: 36.342999999999996
- type: mrr_at_5
value: 37.846999999999994
- type: ndcg_at_1
value: 30.709999999999997
- type: ndcg_at_10
value: 31.608999999999998
- type: ndcg_at_100
value: 38.807
- type: ndcg_at_1000
value: 42.208
- type: ndcg_at_3
value: 28.086
- type: ndcg_at_5
value: 29.323
- type: precision_at_1
value: 30.709999999999997
- type: precision_at_10
value: 8.688
- type: precision_at_100
value: 1.608
- type: precision_at_1000
value: 0.22100000000000003
- type: precision_at_3
value: 18.724
- type: precision_at_5
value: 13.950999999999999
- type: recall_at_1
value: 15.473
- type: recall_at_10
value: 38.361000000000004
- type: recall_at_100
value: 65.2
- type: recall_at_1000
value: 85.789
- type: recall_at_3
value: 25.401
- type: recall_at_5
value: 30.875999999999998
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA-PL
type: clarin-knext/hotpotqa-pl
config: default
split: test
revision: a0bd479ac97b4ccb5bd6ce320c415d0bb4beb907
metrics:
- type: map_at_1
value: 38.096000000000004
- type: map_at_10
value: 51.44499999999999
- type: map_at_100
value: 52.325
- type: map_at_1000
value: 52.397000000000006
- type: map_at_3
value: 48.626999999999995
- type: map_at_5
value: 50.342
- type: mrr_at_1
value: 76.19200000000001
- type: mrr_at_10
value: 81.191
- type: mrr_at_100
value: 81.431
- type: mrr_at_1000
value: 81.443
- type: mrr_at_3
value: 80.30199999999999
- type: mrr_at_5
value: 80.85900000000001
- type: ndcg_at_1
value: 76.19200000000001
- type: ndcg_at_10
value: 60.9
- type: ndcg_at_100
value: 64.14699999999999
- type: ndcg_at_1000
value: 65.647
- type: ndcg_at_3
value: 56.818000000000005
- type: ndcg_at_5
value: 59.019999999999996
- type: precision_at_1
value: 76.19200000000001
- type: precision_at_10
value: 12.203
- type: precision_at_100
value: 1.478
- type: precision_at_1000
value: 0.168
- type: precision_at_3
value: 34.616
- type: precision_at_5
value: 22.515
- type: recall_at_1
value: 38.096000000000004
- type: recall_at_10
value: 61.013
- type: recall_at_100
value: 73.90299999999999
- type: recall_at_1000
value: 83.91
- type: recall_at_3
value: 51.92400000000001
- type: recall_at_5
value: 56.286
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO-PL
type: clarin-knext/msmarco-pl
config: default
split: test
revision: 8634c07806d5cce3a6138e260e59b81760a0a640
metrics:
- type: map_at_1
value: 1.548
- type: map_at_10
value: 11.049000000000001
- type: map_at_100
value: 28.874
- type: map_at_1000
value: 34.931
- type: map_at_3
value: 4.162
- type: map_at_5
value: 6.396
- type: mrr_at_1
value: 90.69800000000001
- type: mrr_at_10
value: 92.093
- type: mrr_at_100
value: 92.345
- type: mrr_at_1000
value: 92.345
- type: mrr_at_3
value: 91.86
- type: mrr_at_5
value: 91.86
- type: ndcg_at_1
value: 74.031
- type: ndcg_at_10
value: 63.978
- type: ndcg_at_100
value: 53.101
- type: ndcg_at_1000
value: 60.675999999999995
- type: ndcg_at_3
value: 71.421
- type: ndcg_at_5
value: 68.098
- type: precision_at_1
value: 90.69800000000001
- type: precision_at_10
value: 71.86
- type: precision_at_100
value: 31.395
- type: precision_at_1000
value: 5.981
- type: precision_at_3
value: 84.49600000000001
- type: precision_at_5
value: 79.07
- type: recall_at_1
value: 1.548
- type: recall_at_10
value: 12.149000000000001
- type: recall_at_100
value: 40.794999999999995
- type: recall_at_1000
value: 67.974
- type: recall_at_3
value: 4.244
- type: recall_at_5
value: 6.608
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (pl)
type: mteb/amazon_massive_intent
config: pl
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 73.55413584398119
- type: f1
value: 69.65610882318181
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (pl)
type: mteb/amazon_massive_scenario
config: pl
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 76.37188971082716
- type: f1
value: 75.64847309941361
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus-PL
type: clarin-knext/nfcorpus-pl
config: default
split: test
revision: 9a6f9567fda928260afed2de480d79c98bf0bec0
metrics:
- type: map_at_1
value: 4.919
- type: map_at_10
value: 10.834000000000001
- type: map_at_100
value: 13.38
- type: map_at_1000
value: 14.581
- type: map_at_3
value: 8.198
- type: map_at_5
value: 9.428
- type: mrr_at_1
value: 41.176
- type: mrr_at_10
value: 50.083
- type: mrr_at_100
value: 50.559
- type: mrr_at_1000
value: 50.604000000000006
- type: mrr_at_3
value: 47.936
- type: mrr_at_5
value: 49.407000000000004
- type: ndcg_at_1
value: 39.628
- type: ndcg_at_10
value: 30.098000000000003
- type: ndcg_at_100
value: 27.061
- type: ndcg_at_1000
value: 35.94
- type: ndcg_at_3
value: 35.135
- type: ndcg_at_5
value: 33.335
- type: precision_at_1
value: 41.176
- type: precision_at_10
value: 22.259999999999998
- type: precision_at_100
value: 6.712
- type: precision_at_1000
value: 1.9060000000000001
- type: precision_at_3
value: 33.23
- type: precision_at_5
value: 29.04
- type: recall_at_1
value: 4.919
- type: recall_at_10
value: 14.196
- type: recall_at_100
value: 26.948
- type: recall_at_1000
value: 59.211000000000006
- type: recall_at_3
value: 9.44
- type: recall_at_5
value: 11.569
- task:
type: Retrieval
dataset:
name: MTEB NQ-PL
type: clarin-knext/nq-pl
config: default
split: test
revision: f171245712cf85dd4700b06bef18001578d0ca8d
metrics:
- type: map_at_1
value: 25.35
- type: map_at_10
value: 37.884
- type: map_at_100
value: 38.955
- type: map_at_1000
value: 39.007999999999996
- type: map_at_3
value: 34.239999999999995
- type: map_at_5
value: 36.398
- type: mrr_at_1
value: 28.737000000000002
- type: mrr_at_10
value: 39.973
- type: mrr_at_100
value: 40.844
- type: mrr_at_1000
value: 40.885
- type: mrr_at_3
value: 36.901
- type: mrr_at_5
value: 38.721
- type: ndcg_at_1
value: 28.708
- type: ndcg_at_10
value: 44.204
- type: ndcg_at_100
value: 48.978
- type: ndcg_at_1000
value: 50.33
- type: ndcg_at_3
value: 37.36
- type: ndcg_at_5
value: 40.912
- type: precision_at_1
value: 28.708
- type: precision_at_10
value: 7.367
- type: precision_at_100
value: 1.0030000000000001
- type: precision_at_1000
value: 0.11299999999999999
- type: precision_at_3
value: 17.034
- type: precision_at_5
value: 12.293999999999999
- type: recall_at_1
value: 25.35
- type: recall_at_10
value: 61.411
- type: recall_at_100
value: 82.599
- type: recall_at_1000
value: 92.903
- type: recall_at_3
value: 43.728
- type: recall_at_5
value: 51.854
- task:
type: Classification
dataset:
name: MTEB PAC
type: laugustyniak/abusive-clauses-pl
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 69.04141326382856
- type: ap
value: 77.49422763833996
- type: f1
value: 66.73472657783407
- task:
type: PairClassification
dataset:
name: MTEB PPC
type: PL-MTEB/ppc-pairclassification
config: default
split: test
revision: None
metrics:
- type: cos_sim_accuracy
value: 81.0
- type: cos_sim_ap
value: 91.47194213011349
- type: cos_sim_f1
value: 84.73767885532592
- type: cos_sim_precision
value: 81.49847094801224
- type: cos_sim_recall
value: 88.24503311258279
- type: dot_accuracy
value: 81.0
- type: dot_ap
value: 91.47194213011349
- type: dot_f1
value: 84.73767885532592
- type: dot_precision
value: 81.49847094801224
- type: dot_recall
value: 88.24503311258279
- type: euclidean_accuracy
value: 81.0
- type: euclidean_ap
value: 91.47194213011349
- type: euclidean_f1
value: 84.73767885532592
- type: euclidean_precision
value: 81.49847094801224
- type: euclidean_recall
value: 88.24503311258279
- type: manhattan_accuracy
value: 81.0
- type: manhattan_ap
value: 91.46464475050571
- type: manhattan_f1
value: 84.48687350835321
- type: manhattan_precision
value: 81.31699846860643
- type: manhattan_recall
value: 87.91390728476821
- type: max_accuracy
value: 81.0
- type: max_ap
value: 91.47194213011349
- type: max_f1
value: 84.73767885532592
- task:
type: PairClassification
dataset:
name: MTEB PSC
type: PL-MTEB/psc-pairclassification
config: default
split: test
revision: None
metrics:
- type: cos_sim_accuracy
value: 97.6808905380334
- type: cos_sim_ap
value: 99.27948611836348
- type: cos_sim_f1
value: 96.15975422427034
- type: cos_sim_precision
value: 96.90402476780186
- type: cos_sim_recall
value: 95.42682926829268
- type: dot_accuracy
value: 97.6808905380334
- type: dot_ap
value: 99.2794861183635
- type: dot_f1
value: 96.15975422427034
- type: dot_precision
value: 96.90402476780186
- type: dot_recall
value: 95.42682926829268
- type: euclidean_accuracy
value: 97.6808905380334
- type: euclidean_ap
value: 99.2794861183635
- type: euclidean_f1
value: 96.15975422427034
- type: euclidean_precision
value: 96.90402476780186
- type: euclidean_recall
value: 95.42682926829268
- type: manhattan_accuracy
value: 97.6808905380334
- type: manhattan_ap
value: 99.28715055268721
- type: manhattan_f1
value: 96.14791987673343
- type: manhattan_precision
value: 97.19626168224299
- type: manhattan_recall
value: 95.1219512195122
- type: max_accuracy
value: 97.6808905380334
- type: max_ap
value: 99.28715055268721
- type: max_f1
value: 96.15975422427034
- task:
type: Classification
dataset:
name: MTEB PolEmo2.0-IN
type: PL-MTEB/polemo2_in
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 86.16343490304708
- type: f1
value: 83.3442579486744
- task:
type: Classification
dataset:
name: MTEB PolEmo2.0-OUT
type: PL-MTEB/polemo2_out
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 68.40080971659918
- type: f1
value: 53.13720751142237
- task:
type: Retrieval
dataset:
name: MTEB Quora-PL
type: clarin-knext/quora-pl
config: default
split: test
revision: 0be27e93455051e531182b85e85e425aba12e9d4
metrics:
- type: map_at_1
value: 63.322
- type: map_at_10
value: 76.847
- type: map_at_100
value: 77.616
- type: map_at_1000
value: 77.644
- type: map_at_3
value: 73.624
- type: map_at_5
value: 75.603
- type: mrr_at_1
value: 72.88
- type: mrr_at_10
value: 80.376
- type: mrr_at_100
value: 80.604
- type: mrr_at_1000
value: 80.61
- type: mrr_at_3
value: 78.92
- type: mrr_at_5
value: 79.869
- type: ndcg_at_1
value: 72.89999999999999
- type: ndcg_at_10
value: 81.43
- type: ndcg_at_100
value: 83.394
- type: ndcg_at_1000
value: 83.685
- type: ndcg_at_3
value: 77.62599999999999
- type: ndcg_at_5
value: 79.656
- type: precision_at_1
value: 72.89999999999999
- type: precision_at_10
value: 12.548
- type: precision_at_100
value: 1.4869999999999999
- type: precision_at_1000
value: 0.155
- type: precision_at_3
value: 34.027
- type: precision_at_5
value: 22.654
- type: recall_at_1
value: 63.322
- type: recall_at_10
value: 90.664
- type: recall_at_100
value: 97.974
- type: recall_at_1000
value: 99.636
- type: recall_at_3
value: 80.067
- type: recall_at_5
value: 85.526
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS-PL
type: clarin-knext/scidocs-pl
config: default
split: test
revision: 45452b03f05560207ef19149545f168e596c9337
metrics:
- type: map_at_1
value: 3.95
- type: map_at_10
value: 9.658999999999999
- type: map_at_100
value: 11.384
- type: map_at_1000
value: 11.677
- type: map_at_3
value: 7.055
- type: map_at_5
value: 8.244
- type: mrr_at_1
value: 19.5
- type: mrr_at_10
value: 28.777
- type: mrr_at_100
value: 29.936
- type: mrr_at_1000
value: 30.009999999999998
- type: mrr_at_3
value: 25.55
- type: mrr_at_5
value: 27.284999999999997
- type: ndcg_at_1
value: 19.5
- type: ndcg_at_10
value: 16.589000000000002
- type: ndcg_at_100
value: 23.879
- type: ndcg_at_1000
value: 29.279
- type: ndcg_at_3
value: 15.719
- type: ndcg_at_5
value: 13.572000000000001
- type: precision_at_1
value: 19.5
- type: precision_at_10
value: 8.62
- type: precision_at_100
value: 1.924
- type: precision_at_1000
value: 0.322
- type: precision_at_3
value: 14.6
- type: precision_at_5
value: 11.78
- type: recall_at_1
value: 3.95
- type: recall_at_10
value: 17.477999999999998
- type: recall_at_100
value: 38.99
- type: recall_at_1000
value: 65.417
- type: recall_at_3
value: 8.883000000000001
- type: recall_at_5
value: 11.933
- task:
type: PairClassification
dataset:
name: MTEB SICK-E-PL
type: PL-MTEB/sicke-pl-pairclassification
config: default
split: test
revision: None
metrics:
- type: cos_sim_accuracy
value: 83.48960456583775
- type: cos_sim_ap
value: 76.31522115825375
- type: cos_sim_f1
value: 70.35573122529645
- type: cos_sim_precision
value: 70.9934735315446
- type: cos_sim_recall
value: 69.72934472934473
- type: dot_accuracy
value: 83.48960456583775
- type: dot_ap
value: 76.31522115825373
- type: dot_f1
value: 70.35573122529645
- type: dot_precision
value: 70.9934735315446
- type: dot_recall
value: 69.72934472934473
- type: euclidean_accuracy
value: 83.48960456583775
- type: euclidean_ap
value: 76.31522115825373
- type: euclidean_f1
value: 70.35573122529645
- type: euclidean_precision
value: 70.9934735315446
- type: euclidean_recall
value: 69.72934472934473
- type: manhattan_accuracy
value: 83.46922136159804
- type: manhattan_ap
value: 76.18474601388084
- type: manhattan_f1
value: 70.34779490856937
- type: manhattan_precision
value: 70.83032490974729
- type: manhattan_recall
value: 69.87179487179486
- type: max_accuracy
value: 83.48960456583775
- type: max_ap
value: 76.31522115825375
- type: max_f1
value: 70.35573122529645
- task:
type: STS
dataset:
name: MTEB SICK-R-PL
type: PL-MTEB/sickr-pl-sts
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 77.95374883876302
- type: cos_sim_spearman
value: 73.77630219171942
- type: euclidean_pearson
value: 75.81927069594934
- type: euclidean_spearman
value: 73.7763211303831
- type: manhattan_pearson
value: 76.03126859057528
- type: manhattan_spearman
value: 73.96528138013369
- task:
type: STS
dataset:
name: MTEB STS22 (pl)
type: mteb/sts22-crosslingual-sts
config: pl
split: test
revision: eea2b4fe26a775864c896887d910b76a8098ad3f
metrics:
- type: cos_sim_pearson
value: 37.388282764841826
- type: cos_sim_spearman
value: 40.83477184710897
- type: euclidean_pearson
value: 26.754737044177805
- type: euclidean_spearman
value: 40.83477184710897
- type: manhattan_pearson
value: 26.760453110872458
- type: manhattan_spearman
value: 41.034477441383856
- task:
type: Retrieval
dataset:
name: MTEB SciFact-PL
type: clarin-knext/scifact-pl
config: default
split: test
revision: 47932a35f045ef8ed01ba82bf9ff67f6e109207e
metrics:
- type: map_at_1
value: 49.15
- type: map_at_10
value: 61.690999999999995
- type: map_at_100
value: 62.348000000000006
- type: map_at_1000
value: 62.38
- type: map_at_3
value: 58.824
- type: map_at_5
value: 60.662000000000006
- type: mrr_at_1
value: 51.333
- type: mrr_at_10
value: 62.731
- type: mrr_at_100
value: 63.245
- type: mrr_at_1000
value: 63.275000000000006
- type: mrr_at_3
value: 60.667
- type: mrr_at_5
value: 61.93300000000001
- type: ndcg_at_1
value: 51.333
- type: ndcg_at_10
value: 67.168
- type: ndcg_at_100
value: 69.833
- type: ndcg_at_1000
value: 70.56700000000001
- type: ndcg_at_3
value: 62.40599999999999
- type: ndcg_at_5
value: 65.029
- type: precision_at_1
value: 51.333
- type: precision_at_10
value: 9.333
- type: precision_at_100
value: 1.0699999999999998
- type: precision_at_1000
value: 0.11299999999999999
- type: precision_at_3
value: 25.333
- type: precision_at_5
value: 17.067
- type: recall_at_1
value: 49.15
- type: recall_at_10
value: 82.533
- type: recall_at_100
value: 94.167
- type: recall_at_1000
value: 99.667
- type: recall_at_3
value: 69.917
- type: recall_at_5
value: 76.356
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID-PL
type: clarin-knext/trec-covid-pl
config: default
split: test
revision: 81bcb408f33366c2a20ac54adafad1ae7e877fdd
metrics:
- type: map_at_1
value: 0.261
- type: map_at_10
value: 2.1260000000000003
- type: map_at_100
value: 12.171999999999999
- type: map_at_1000
value: 26.884999999999998
- type: map_at_3
value: 0.695
- type: map_at_5
value: 1.134
- type: mrr_at_1
value: 96.0
- type: mrr_at_10
value: 96.952
- type: mrr_at_100
value: 96.952
- type: mrr_at_1000
value: 96.952
- type: mrr_at_3
value: 96.667
- type: mrr_at_5
value: 96.667
- type: ndcg_at_1
value: 92.0
- type: ndcg_at_10
value: 81.193
- type: ndcg_at_100
value: 61.129
- type: ndcg_at_1000
value: 51.157
- type: ndcg_at_3
value: 85.693
- type: ndcg_at_5
value: 84.129
- type: precision_at_1
value: 96.0
- type: precision_at_10
value: 85.39999999999999
- type: precision_at_100
value: 62.03999999999999
- type: precision_at_1000
value: 22.224
- type: precision_at_3
value: 88.0
- type: precision_at_5
value: 88.0
- type: recall_at_1
value: 0.261
- type: recall_at_10
value: 2.262
- type: recall_at_100
value: 14.981
- type: recall_at_1000
value: 46.837
- type: recall_at_3
value: 0.703
- type: recall_at_5
value: 1.172
- task:
type: Clustering
dataset:
name: MTEB AlloProfClusteringP2P
type: lyon-nlp/alloprof
config: default
split: test
revision: 392ba3f5bcc8c51f578786c1fc3dae648662cb9b
metrics:
- type: v_measure
value: 70.55290063940157
- type: v_measure
value: 55.41500719337263
- task:
type: Reranking
dataset:
name: MTEB AlloprofReranking
type: lyon-nlp/mteb-fr-reranking-alloprof-s2p
config: default
split: test
revision: 666fdacebe0291776e86f29345663dfaf80a0db9
metrics:
- type: map
value: 73.48697375332002
- type: mrr
value: 75.01836585523822
- task:
type: Retrieval
dataset:
name: MTEB AlloprofRetrieval
type: lyon-nlp/alloprof
config: default
split: test
revision: 392ba3f5bcc8c51f578786c1fc3dae648662cb9b
metrics:
- type: map_at_1
value: 38.454
- type: map_at_10
value: 51.605000000000004
- type: map_at_100
value: 52.653000000000006
- type: map_at_1000
value: 52.697
- type: map_at_3
value: 48.304
- type: map_at_5
value: 50.073
- type: mrr_at_1
value: 43.307
- type: mrr_at_10
value: 54.400000000000006
- type: mrr_at_100
value: 55.147999999999996
- type: mrr_at_1000
value: 55.174
- type: mrr_at_3
value: 51.77
- type: mrr_at_5
value: 53.166999999999994
- type: ndcg_at_1
value: 43.307
- type: ndcg_at_10
value: 57.891000000000005
- type: ndcg_at_100
value: 62.161
- type: ndcg_at_1000
value: 63.083
- type: ndcg_at_3
value: 51.851
- type: ndcg_at_5
value: 54.605000000000004
- type: precision_at_1
value: 43.307
- type: precision_at_10
value: 9.033
- type: precision_at_100
value: 1.172
- type: precision_at_1000
value: 0.127
- type: precision_at_3
value: 22.798
- type: precision_at_5
value: 15.492
- type: recall_at_1
value: 38.454
- type: recall_at_10
value: 74.166
- type: recall_at_100
value: 92.43599999999999
- type: recall_at_1000
value: 99.071
- type: recall_at_3
value: 58.087
- type: recall_at_5
value: 64.568
- task:
type: Classification
dataset:
name: MTEB AmazonReviewsClassification (fr)
type: mteb/amazon_reviews_multi
config: fr
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 53.474
- type: f1
value: 50.38275392350236
- task:
type: Retrieval
dataset:
name: MTEB BSARDRetrieval
type: maastrichtlawtech/bsard
config: default
split: test
revision: 5effa1b9b5fa3b0f9e12523e6e43e5f86a6e6d59
metrics:
- type: map_at_1
value: 2.252
- type: map_at_10
value: 4.661
- type: map_at_100
value: 5.271
- type: map_at_1000
value: 5.3629999999999995
- type: map_at_3
value: 3.604
- type: map_at_5
value: 4.3020000000000005
- type: mrr_at_1
value: 2.252
- type: mrr_at_10
value: 4.661
- type: mrr_at_100
value: 5.271
- type: mrr_at_1000
value: 5.3629999999999995
- type: mrr_at_3
value: 3.604
- type: mrr_at_5
value: 4.3020000000000005
- type: ndcg_at_1
value: 2.252
- type: ndcg_at_10
value: 6.3020000000000005
- type: ndcg_at_100
value: 10.342
- type: ndcg_at_1000
value: 13.475999999999999
- type: ndcg_at_3
value: 4.0649999999999995
- type: ndcg_at_5
value: 5.344
- type: precision_at_1
value: 2.252
- type: precision_at_10
value: 1.171
- type: precision_at_100
value: 0.333
- type: precision_at_1000
value: 0.059000000000000004
- type: precision_at_3
value: 1.802
- type: precision_at_5
value: 1.712
- type: recall_at_1
value: 2.252
- type: recall_at_10
value: 11.712
- type: recall_at_100
value: 33.333
- type: recall_at_1000
value: 59.458999999999996
- type: recall_at_3
value: 5.405
- type: recall_at_5
value: 8.559
- task:
type: Clustering
dataset:
name: MTEB HALClusteringS2S
type: lyon-nlp/clustering-hal-s2s
config: default
split: test
revision: e06ebbbb123f8144bef1a5d18796f3dec9ae2915
metrics:
- type: v_measure
value: 28.301882091023288
- task:
type: Clustering
dataset:
name: MTEB MLSUMClusteringP2P
type: mlsum
config: default
split: test
revision: b5d54f8f3b61ae17845046286940f03c6bc79bc7
metrics:
- type: v_measure
value: 45.26992995191701
- type: v_measure
value: 42.773174876871145
- task:
type: Classification
dataset:
name: MTEB MTOPDomainClassification (fr)
type: mteb/mtop_domain
config: fr
split: test
revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf
metrics:
- type: accuracy
value: 93.47635452552458
- type: f1
value: 93.19922617577213
- task:
type: Classification
dataset:
name: MTEB MTOPIntentClassification (fr)
type: mteb/mtop_intent
config: fr
split: test
revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba
metrics:
- type: accuracy
value: 80.2317569683683
- type: f1
value: 56.18060418621901
- task:
type: Classification
dataset:
name: MTEB MasakhaNEWSClassification (fra)
type: masakhane/masakhanews
config: fra
split: test
revision: 8ccc72e69e65f40c70e117d8b3c08306bb788b60
metrics:
- type: accuracy
value: 85.18957345971565
- type: f1
value: 80.829981537394
- task:
type: Clustering
dataset:
name: MTEB MasakhaNEWSClusteringP2P (fra)
type: masakhane/masakhanews
config: fra
split: test
revision: 8ccc72e69e65f40c70e117d8b3c08306bb788b60
metrics:
- type: v_measure
value: 71.04138999801822
- type: v_measure
value: 71.7056263158008
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (fr)
type: mteb/amazon_massive_intent
config: fr
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 76.65097511768661
- type: f1
value: 73.82441070598712
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (fr)
type: mteb/amazon_massive_scenario
config: fr
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 79.09885675857431
- type: f1
value: 78.28407777434224
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (fr)
type: jinaai/mintakaqa
config: fr
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: map_at_1
value: 25.307000000000002
- type: map_at_10
value: 36.723
- type: map_at_100
value: 37.713
- type: map_at_1000
value: 37.769000000000005
- type: map_at_3
value: 33.77
- type: map_at_5
value: 35.463
- type: mrr_at_1
value: 25.307000000000002
- type: mrr_at_10
value: 36.723
- type: mrr_at_100
value: 37.713
- type: mrr_at_1000
value: 37.769000000000005
- type: mrr_at_3
value: 33.77
- type: mrr_at_5
value: 35.463
- type: ndcg_at_1
value: 25.307000000000002
- type: ndcg_at_10
value: 42.559999999999995
- type: ndcg_at_100
value: 47.457
- type: ndcg_at_1000
value: 49.162
- type: ndcg_at_3
value: 36.461
- type: ndcg_at_5
value: 39.504
- type: precision_at_1
value: 25.307000000000002
- type: precision_at_10
value: 6.106
- type: precision_at_100
value: 0.8420000000000001
- type: precision_at_1000
value: 0.098
- type: precision_at_3
value: 14.741999999999999
- type: precision_at_5
value: 10.319
- type: recall_at_1
value: 25.307000000000002
- type: recall_at_10
value: 61.056999999999995
- type: recall_at_100
value: 84.152
- type: recall_at_1000
value: 98.03399999999999
- type: recall_at_3
value: 44.226
- type: recall_at_5
value: 51.597
- task:
type: PairClassification
dataset:
name: MTEB OpusparcusPC (fr)
type: GEM/opusparcus
config: fr
split: test
revision: 9e9b1f8ef51616073f47f306f7f47dd91663f86a
metrics:
- type: cos_sim_accuracy
value: 99.90069513406156
- type: cos_sim_ap
value: 100.0
- type: cos_sim_f1
value: 99.95032290114257
- type: cos_sim_precision
value: 100.0
- type: cos_sim_recall
value: 99.90069513406156
- type: dot_accuracy
value: 99.90069513406156
- type: dot_ap
value: 100.0
- type: dot_f1
value: 99.95032290114257
- type: dot_precision
value: 100.0
- type: dot_recall
value: 99.90069513406156
- type: euclidean_accuracy
value: 99.90069513406156
- type: euclidean_ap
value: 100.0
- type: euclidean_f1
value: 99.95032290114257
- type: euclidean_precision
value: 100.0
- type: euclidean_recall
value: 99.90069513406156
- type: manhattan_accuracy
value: 99.90069513406156
- type: manhattan_ap
value: 100.0
- type: manhattan_f1
value: 99.95032290114257
- type: manhattan_precision
value: 100.0
- type: manhattan_recall
value: 99.90069513406156
- type: max_accuracy
value: 99.90069513406156
- type: max_ap
value: 100.0
- type: max_f1
value: 99.95032290114257
- task:
type: PairClassification
dataset:
name: MTEB PawsX (fr)
type: paws-x
config: fr
split: test
revision: 8a04d940a42cd40658986fdd8e3da561533a3646
metrics:
- type: cos_sim_accuracy
value: 70.8
- type: cos_sim_ap
value: 73.7671529695957
- type: cos_sim_f1
value: 68.80964339527875
- type: cos_sim_precision
value: 62.95955882352941
- type: cos_sim_recall
value: 75.85825027685493
- type: dot_accuracy
value: 70.8
- type: dot_ap
value: 73.78345265366947
- type: dot_f1
value: 68.80964339527875
- type: dot_precision
value: 62.95955882352941
- type: dot_recall
value: 75.85825027685493
- type: euclidean_accuracy
value: 70.8
- type: euclidean_ap
value: 73.7671529695957
- type: euclidean_f1
value: 68.80964339527875
- type: euclidean_precision
value: 62.95955882352941
- type: euclidean_recall
value: 75.85825027685493
- type: manhattan_accuracy
value: 70.75
- type: manhattan_ap
value: 73.78996383615953
- type: manhattan_f1
value: 68.79432624113475
- type: manhattan_precision
value: 63.39869281045751
- type: manhattan_recall
value: 75.1937984496124
- type: max_accuracy
value: 70.8
- type: max_ap
value: 73.78996383615953
- type: max_f1
value: 68.80964339527875
- task:
type: STS
dataset:
name: MTEB SICKFr
type: Lajavaness/SICK-fr
config: default
split: test
revision: e077ab4cf4774a1e36d86d593b150422fafd8e8a
metrics:
- type: cos_sim_pearson
value: 84.03253762760392
- type: cos_sim_spearman
value: 79.68280105762004
- type: euclidean_pearson
value: 80.98265050044444
- type: euclidean_spearman
value: 79.68233242682867
- type: manhattan_pearson
value: 80.9678911810704
- type: manhattan_spearman
value: 79.70264097683109
- task:
type: STS
dataset:
name: MTEB STS22 (fr)
type: mteb/sts22-crosslingual-sts
config: fr
split: test
revision: eea2b4fe26a775864c896887d910b76a8098ad3f
metrics:
- type: cos_sim_pearson
value: 80.56896987572884
- type: cos_sim_spearman
value: 81.84352499523287
- type: euclidean_pearson
value: 80.40831759421305
- type: euclidean_spearman
value: 81.84352499523287
- type: manhattan_pearson
value: 80.74333857561238
- type: manhattan_spearman
value: 82.41503246733892
- task:
type: STS
dataset:
name: MTEB STSBenchmarkMultilingualSTS (fr)
type: stsb_multi_mt
config: fr
split: test
revision: 93d57ef91790589e3ce9c365164337a8a78b7632
metrics:
- type: cos_sim_pearson
value: 82.71826762276979
- type: cos_sim_spearman
value: 82.25433354916042
- type: euclidean_pearson
value: 81.87115571724316
- type: euclidean_spearman
value: 82.25322342890107
- type: manhattan_pearson
value: 82.11174867527224
- type: manhattan_spearman
value: 82.55905365203084
- task:
type: Summarization
dataset:
name: MTEB SummEvalFr
type: lyon-nlp/summarization-summeval-fr-p2p
config: default
split: test
revision: b385812de6a9577b6f4d0f88c6a6e35395a94054
metrics:
- type: cos_sim_pearson
value: 30.659441623392887
- type: cos_sim_spearman
value: 30.501134097353315
- type: dot_pearson
value: 30.659444768851056
- type: dot_spearman
value: 30.501134097353315
- task:
type: Reranking
dataset:
name: MTEB SyntecReranking
type: lyon-nlp/mteb-fr-reranking-syntec-s2p
config: default
split: test
revision: b205c5084a0934ce8af14338bf03feb19499c84d
metrics:
- type: map
value: 94.03333333333333
- type: mrr
value: 94.03333333333333
- task:
type: Retrieval
dataset:
name: MTEB SyntecRetrieval
type: lyon-nlp/mteb-fr-retrieval-syntec-s2p
config: default
split: test
revision: 77f7e271bf4a92b24fce5119f3486b583ca016ff
metrics:
- type: map_at_1
value: 79.0
- type: map_at_10
value: 87.61
- type: map_at_100
value: 87.655
- type: map_at_1000
value: 87.655
- type: map_at_3
value: 87.167
- type: map_at_5
value: 87.36699999999999
- type: mrr_at_1
value: 79.0
- type: mrr_at_10
value: 87.61
- type: mrr_at_100
value: 87.655
- type: mrr_at_1000
value: 87.655
- type: mrr_at_3
value: 87.167
- type: mrr_at_5
value: 87.36699999999999
- type: ndcg_at_1
value: 79.0
- type: ndcg_at_10
value: 90.473
- type: ndcg_at_100
value: 90.694
- type: ndcg_at_1000
value: 90.694
- type: ndcg_at_3
value: 89.464
- type: ndcg_at_5
value: 89.851
- type: precision_at_1
value: 79.0
- type: precision_at_10
value: 9.9
- type: precision_at_100
value: 1.0
- type: precision_at_1000
value: 0.1
- type: precision_at_3
value: 32.0
- type: precision_at_5
value: 19.400000000000002
- type: recall_at_1
value: 79.0
- type: recall_at_10
value: 99.0
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: recall_at_3
value: 96.0
- type: recall_at_5
value: 97.0
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (fr)
type: jinaai/xpqa
config: fr
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: map_at_1
value: 39.395
- type: map_at_10
value: 59.123999999999995
- type: map_at_100
value: 60.704
- type: map_at_1000
value: 60.760000000000005
- type: map_at_3
value: 53.187
- type: map_at_5
value: 56.863
- type: mrr_at_1
value: 62.083
- type: mrr_at_10
value: 68.87299999999999
- type: mrr_at_100
value: 69.46900000000001
- type: mrr_at_1000
value: 69.48299999999999
- type: mrr_at_3
value: 66.8
- type: mrr_at_5
value: 67.928
- type: ndcg_at_1
value: 62.083
- type: ndcg_at_10
value: 65.583
- type: ndcg_at_100
value: 70.918
- type: ndcg_at_1000
value: 71.72800000000001
- type: ndcg_at_3
value: 60.428000000000004
- type: ndcg_at_5
value: 61.853
- type: precision_at_1
value: 62.083
- type: precision_at_10
value: 15.033
- type: precision_at_100
value: 1.9529999999999998
- type: precision_at_1000
value: 0.207
- type: precision_at_3
value: 36.315
- type: precision_at_5
value: 25.955000000000002
- type: recall_at_1
value: 39.395
- type: recall_at_10
value: 74.332
- type: recall_at_100
value: 94.729
- type: recall_at_1000
value: 99.75500000000001
- type: recall_at_3
value: 57.679
- type: recall_at_5
value: 65.036
---
## gte-Qwen2-1.5B-instruct
**gte-Qwen2-1.5B-instruct** is the latest model in the gte (General Text Embedding) model family. The model is built on [Qwen2-1.5B](https://huggingface.co/Qwen/Qwen2-1.5B) LLM model and use the same training data and strategies as the [gte-Qwen2-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) model.
The model incorporates several key advancements:
- Integration of bidirectional attention mechanisms, enriching its contextual understanding.
- Instruction tuning, applied solely on the query side for streamlined efficiency
- Comprehensive training across a vast, multilingual text corpus spanning diverse domains and scenarios. This training leverages both weakly supervised and supervised data, ensuring the model's applicability across numerous languages and a wide array of downstream tasks.
## Model Information
- Model Size: 1.5B
- Embedding Dimension: 1536
- Max Input Tokens: 32k
## Requirements
```
transformers>=4.39.2
flash_attn>=2.5.6
```
## Usage
### Sentence Transformers
```python
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("Alibaba-NLP/gte-Qwen2-1.5B-instruct", trust_remote_code=True)
# In case you want to reduce the maximum length:
model.max_seq_length = 8192
queries = [
"how much protein should a female eat",
"summit define",
]
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments.",
]
query_embeddings = model.encode(queries, prompt_name="query")
document_embeddings = model.encode(documents)
scores = (query_embeddings @ document_embeddings.T) * 100
print(scores.tolist())
```
Observe the [config_sentence_transformers.json](config_sentence_transformers.json) to see all pre-built prompt names. Otherwise, you can use `model.encode(queries, prompt="Instruct: ...\nQuery: "` to use a custom prompt of your choice.
### Transformers
```python
import torch
import torch.nn.functional as F
from torch import Tensor
from transformers import AutoTokenizer, AutoModel
def last_token_pool(last_hidden_states: Tensor,
attention_mask: Tensor) -> Tensor:
left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])
if left_padding:
return last_hidden_states[:, -1]
else:
sequence_lengths = attention_mask.sum(dim=1) - 1
batch_size = last_hidden_states.shape[0]
return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'Instruct: {task_description}\nQuery: {query}'
# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = [
get_detailed_instruct(task, 'how much protein should a female eat'),
get_detailed_instruct(task, 'summit define')
]
# No need to add instruction for retrieval documents
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."
]
input_texts = queries + documents
tokenizer = AutoTokenizer.from_pretrained('Alibaba-NLP/gte-Qwen2-1.5B-instruct', trust_remote_code=True)
model = AutoModel.from_pretrained('Alibaba-NLP/gte-Qwen2-1.5B-instruct', trust_remote_code=True)
max_length = 8192
# Tokenize the input texts
batch_dict = tokenizer(input_texts, max_length=max_length, padding=True, truncation=True, return_tensors='pt')
outputs = model(**batch_dict)
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
# normalize embeddings
embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:2] @ embeddings[2:].T) * 100
print(scores.tolist())
```
## Evaluation
### MTEB & C-MTEB
You can use the [scripts/eval_mteb.py](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct/blob/main/scripts/eval_mteb.py) to reproduce the following result of **gte-Qwen2-1.5B-instruct** on MTEB(English)/C-MTEB(Chinese):
| Model Name | MTEB(56) | C-MTEB(35) | MTEB-fr(26) | MTEB-pl(26) |
|:----:|:---------:|:----------:|:----------:|:----------:|
| [bge-base-en-1.5](https://huggingface.co/BAAI/bge-base-en-v1.5) | 64.23 | - | - | - |
| [bge-large-en-1.5](https://huggingface.co/BAAI/bge-large-en-v1.5) | 63.55 | - | - | - |
| [gte-large-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | 65.39 | - | - | - |
| [gte-base-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | 64.11 | - | - | - |
| [mxbai-embed-large-v1](https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1) | 64.68 | - | - | - |
| [acge_text_embedding](https://huggingface.co/aspire/acge_text_embedding) | - | 69.07 | - | - |
| [stella-mrl-large-zh-v3.5-1792d](https://huggingface.co/infgrad/stella-mrl-large-zh-v3.5-1792d) | - | 68.55 | - | - |
| [gte-large-zh](https://huggingface.co/thenlper/gte-large-zh) | - | 66.72 | - | - |
| [multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) | 59.45 | 56.21 | - | - |
| [multilingual-e5-large](https://huggingface.co/intfloat/multilingual-e5-large) | 61.50 | 58.81 | - | - |
| [e5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct) | 66.63 | 60.81 | - | - |
| [gte-Qwen1.5-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct) | 67.34 | 69.52 | - | - |
| [NV-Embed-v1](https://huggingface.co/nvidia/NV-Embed-v1) | 69.32 | - | - | - |
| [**gte-Qwen2-7B-instruct**](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) | **70.24** | **72.05** | **68.25** | **67.86** |
| [**gte-Qwen2-1.5B-instruct**](https://huggingface.co/Alibaba-NLP/gte-Qwen2-1.5B-instruct) | **67.16** | **67.65** | **66.60** | **64.04** |
### GTE Models
The gte series models have consistently released two types of models: encoder-only models (based on the BERT architecture) and decode-only models (based on the LLM architecture).
| Models | Language | Max Sequence Length | Dimension | Model Size (Memory Usage, fp32) |
|:-------------------------------------------------------------------------------------:|:--------:|:-----: |:---------:|:-------------------------------:|
| [GTE-large-zh](https://huggingface.co/thenlper/gte-large-zh) | Chinese | 512 | 1024 | 1.25GB |
| [GTE-base-zh](https://huggingface.co/thenlper/gte-base-zh) | Chinese | 512 | 512 | 0.41GB |
| [GTE-small-zh](https://huggingface.co/thenlper/gte-small-zh) | Chinese | 512 | 512 | 0.12GB |
| [GTE-large](https://huggingface.co/thenlper/gte-large) | English | 512 | 1024 | 1.25GB |
| [GTE-base](https://huggingface.co/thenlper/gte-base) | English | 512 | 512 | 0.21GB |
| [GTE-small](https://huggingface.co/thenlper/gte-small) | English | 512 | 384 | 0.10GB |
| [GTE-large-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | English | 8192 | 1024 | 1.74GB |
| [GTE-base-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5) | English | 8192 | 768 | 0.51GB |
| [GTE-Qwen1.5-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct) | Multilingual | 32000 | 4096 | 26.45GB |
| [GTE-Qwen2-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) | Multilingual | 32000 | 3584 | 26.45GB |
| [GTE-Qwen2-1.5B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-1.5B-instruct) | Multilingual | 32000 | 1536 | 6.62GB |
## Cloud API Services
In addition to the open-source [GTE](https://huggingface.co/collections/Alibaba-NLP/gte-models-6680f0b13f885cb431e6d469) series models, GTE series models are also available as commercial API services on Alibaba Cloud.
- [Embedding Models](https://help.aliyun.com/zh/model-studio/developer-reference/general-text-embedding/): Rhree versions of the text embedding models are available: text-embedding-v1/v2/v3, with v3 being the latest API service.
- [ReRank Models](https://help.aliyun.com/zh/model-studio/developer-reference/general-text-sorting-model/): The gte-rerank model service is available.
Note that the models behind the commercial APIs are not entirely identical to the open-source models.
## Citation
If you find our paper or models helpful, please consider cite:
```
@article{li2023towards,
title={Towards general text embeddings with multi-stage contrastive learning},
author={Li, Zehan and Zhang, Xin and Zhang, Yanzhao and Long, Dingkun and Xie, Pengjun and Zhang, Meishan},
journal={arXiv preprint arXiv:2308.03281},
year={2023}
}
```
| [
"BIOSSES",
"SCIFACT"
] |
EleutherAI/pythia-2.8b-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:the_pile",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-11-20T03:56:10Z" | 2023-07-10T01:35:41+00:00 | 629 | 5 | ---
datasets:
- the_pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-2.8B
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change over the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-2.8B for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-2.8B as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-2.8B has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-2.8B will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-2.8B to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-2.8B may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-2.8B.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-2.8B.
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
yachty66/flux_dev_lora_me_v1 | yachty66 | text-to-image | [
"diffusers",
"text-to-image",
"flux",
"lora",
"template:sd-lora",
"ai-toolkit",
"base_model:black-forest-labs/FLUX.1-dev",
"base_model:adapter:black-forest-labs/FLUX.1-dev",
"license:other",
"region:us"
] | "2024-09-30T14:50:37Z" | 2024-09-30T14:54:27+00:00 | 628 | 0 | ---
base_model: black-forest-labs/FLUX.1-dev
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
tags:
- text-to-image
- flux
- lora
- diffusers
- template:sd-lora
- ai-toolkit
widget:
- text: woman with red hair, playing chess at the park, bomb going off in the background
output:
url: samples/1727705731570__000002000_0.jpg
- text: a woman holding a coffee cup, in a beanie, sitting at a cafe
output:
url: samples/1727705748892__000002000_1.jpg
- text: a horse is a DJ at a night club, fish eye lens, smoke machine, lazer lights,
holding a martini
output:
url: samples/1727705766197__000002000_2.jpg
- text: a man showing off his cool new t shirt at the beach, a shark is jumping out
of the water in the background
output:
url: samples/1727705783492__000002000_3.jpg
- text: a bear building a log cabin in the snow covered mountains
output:
url: samples/1727705800791__000002000_4.jpg
- text: woman playing the guitar, on stage, singing a song, laser lights, punk rocker
output:
url: samples/1727705818071__000002000_5.jpg
- text: hipster man with a beard, building a chair, in a wood shop
output:
url: samples/1727705835387__000002000_6.jpg
- text: photo of a man, white background, medium shot, modeling clothing, studio lighting,
white backdrop
output:
url: samples/1727705852686__000002000_7.jpg
- text: a man holding a sign that says, 'this is a sign'
output:
url: samples/1727705869996__000002000_8.jpg
- text: a bulldog, in a post apocalyptic world, with a shotgun, in a leather jacket,
in a desert, with a motorcycle
output:
url: samples/1727705887388__000002000_9.jpg
---
# my_first_flux_lora_v1
Model trained with [AI Toolkit by Ostris](https://github.com/ostris/ai-toolkit)
<Gallery />
## Trigger words
No trigger words defined.
## Download model and use it with ComfyUI, AUTOMATIC1111, SD.Next, Invoke AI, etc.
Weights for this model are available in Safetensors format.
[Download](/None/tree/main) them in the Files & versions tab.
## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
```py
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.bfloat16).to('cuda')
pipeline.load_lora_weights('None', weight_name='my_first_flux_lora_v1.safetensors')
image = pipeline('woman with red hair, playing chess at the park, bomb going off in the background').images[0]
image.save("my_image.png")
```
For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
| [
"BEAR"
] |
mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF | mradermacher | null | [
"transformers",
"gguf",
"biology",
"medical",
"healthcare",
"en",
"dataset:HPAI-BSC/Aloe-Beta-General-Collection",
"dataset:HPAI-BSC/chain-of-diagnosis",
"dataset:HPAI-BSC/MedS-Ins",
"dataset:HPAI-BSC/ultramedical",
"dataset:HPAI-BSC/pubmedqa-cot-llama31",
"dataset:HPAI-BSC/medqa-cot-llama31",
"dataset:HPAI-BSC/medmcqa-cot-llama31",
"dataset:HPAI-BSC/headqa-cot-llama31",
"dataset:HPAI-BSC/MMLU-medical-cot-llama31",
"dataset:HPAI-BSC/Polymed-QA",
"base_model:HPAI-BSC/Qwen2.5-Aloe-Beta-72B",
"base_model:quantized:HPAI-BSC/Qwen2.5-Aloe-Beta-72B",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | "2024-12-12T00:14:06Z" | 2024-12-13T05:23:11+00:00 | 626 | 0 | ---
base_model: HPAI-BSC/Qwen2.5-Aloe-Beta-72B
datasets:
- HPAI-BSC/Aloe-Beta-General-Collection
- HPAI-BSC/chain-of-diagnosis
- HPAI-BSC/MedS-Ins
- HPAI-BSC/ultramedical
- HPAI-BSC/pubmedqa-cot-llama31
- HPAI-BSC/medqa-cot-llama31
- HPAI-BSC/medmcqa-cot-llama31
- HPAI-BSC/headqa-cot-llama31
- HPAI-BSC/MMLU-medical-cot-llama31
- HPAI-BSC/Polymed-QA
- HPAI-BSC/Aloe-Beta-General-Collection
- HPAI-BSC/Aloe-Beta-General-Collection
language:
- en
library_name: transformers
tags:
- biology
- medical
- healthcare
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/HPAI-BSC/Qwen2.5-Aloe-Beta-72B
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ1_S.gguf) | i1-IQ1_S | 22.8 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ1_M.gguf) | i1-IQ1_M | 23.8 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 25.6 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ2_XS.gguf) | i1-IQ2_XS | 27.2 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ2_S.gguf) | i1-IQ2_S | 28.0 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ2_M.gguf) | i1-IQ2_M | 29.4 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q2_K_S.gguf) | i1-Q2_K_S | 29.7 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q2_K.gguf) | i1-Q2_K | 29.9 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 31.9 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ3_XS.gguf) | i1-IQ3_XS | 32.9 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ3_S.gguf) | i1-IQ3_S | 34.6 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q3_K_S.gguf) | i1-Q3_K_S | 34.6 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ3_M.gguf) | i1-IQ3_M | 35.6 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q3_K_M.gguf) | i1-Q3_K_M | 37.8 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q3_K_L.gguf) | i1-Q3_K_L | 39.6 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-IQ4_XS.gguf) | i1-IQ4_XS | 39.8 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q4_0.gguf) | i1-Q4_0 | 41.5 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q4_K_S.gguf) | i1-Q4_K_S | 44.0 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q4_K_M.gguf) | i1-Q4_K_M | 47.5 | fast, recommended |
| [PART 1](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q5_K_S.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q5_K_S.gguf.part2of2) | i1-Q5_K_S | 51.5 | |
| [PART 1](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q5_K_M.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q5_K_M.gguf.part2of2) | i1-Q5_K_M | 54.5 | |
| [PART 1](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q6_K.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-72B-i1-GGUF/resolve/main/Qwen2.5-Aloe-Beta-72B.i1-Q6_K.gguf.part2of2) | i1-Q6_K | 64.4 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
| [
"MEDQA",
"PUBMEDQA"
] |
EleutherAI/pythia-410m-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:the_pile",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-16T18:39:39Z" | 2023-07-10T01:34:52+00:00 | 623 | 7 | ---
datasets:
- the_pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-410M
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change over the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-410M for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-410M as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-410M has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-410M will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-410M to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-410M may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-410M.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-410M.
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
IVN-RIN/bioBIT | IVN-RIN | fill-mask | [
"transformers",
"pytorch",
"safetensors",
"bert",
"fill-mask",
"Biomedical Language Modeling",
"it",
"dataset:IVN-RIN/BioBERT_Italian",
"arxiv:1901.08746",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2022-12-03T11:17:50Z" | 2024-05-24T11:57:03+00:00 | 623 | 1 | ---
datasets:
- IVN-RIN/BioBERT_Italian
language:
- it
tags:
- Biomedical Language Modeling
widget:
- text: L'asma allergica è una patologia dell'[MASK] respiratorio causata dalla presenza
di allergeni responsabili dell'infiammazione dell'albero bronchiale.
example_title: Example 1
- text: Il pancreas produce diversi [MASK] molto importanti tra i quali l'insulina
e il glucagone.
example_title: Example 2
- text: Il GABA è un amminoacido ed è il principale neurotrasmettitore inibitorio
del [MASK].
example_title: Example 3
---
🤗 + 📚🩺🇮🇹 = **BioBIT**
From this repository you can download the **BioBIT** (Biomedical Bert for ITalian) checkpoint.
**BioBIT** stems from [Italian XXL BERT](https://huggingface.co/dbmdz/bert-base-italian-xxl-cased), obtained from a recent Wikipedia dump and various texts in Italian from the OPUS and OSCAR corpora collection, which sums up to the final corpus size of 81 GB and 13B tokens.
To pretrain **BioBIT**, we followed the general approach outlined in [BioBERT paper](https://arxiv.org/abs/1901.08746), built on the foundation of the BERT architecture. The pretraining objective is a combination of **MLM** (Masked Language Modelling) and **NSP** (Next Sentence Prediction). The MLM objective is based on randomly masking 15% of the input sequence, trying then to predict the missing tokens; for the NSP objective, instead, the model is given a couple of sentences and has to guess if the second comes after the first in the original document.
Due to the unavailability of an Italian equivalent for the millions of abstracts and full-text scientific papers used by English, BERT-based biomedical models, in this work we leveraged machine translation to obtain an Italian biomedical corpus based on PubMed abstracts and train **BioBIT**. More details in the paper.
**BioBIT** has been evaluated on 3 downstream tasks: **NER** (Named Entity Recognition), extractive **QA** (Question Answering), **RE** (Relation Extraction).
Here are the results, summarized:
- NER:
- [BC2GM](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb32) = 82.14%
- [BC4CHEMD](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb35) = 80.70%
- [BC5CDR(CDR)](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb31) = 82.15%
- [BC5CDR(DNER)](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb31) = 76.27%
- [NCBI_DISEASE](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb33) = 65.06%
- [SPECIES-800](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb34) = 61.86%
- QA:
- [BioASQ 4b](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb30) = 68.49%
- [BioASQ 5b](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb30) = 78.33%
- [BioASQ 6b](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb30) = 75.73%
- RE:
- [CHEMPROT](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb36) = 38.16%
- [BioRED](http://refhub.elsevier.com/S1532-0464(23)00152-1/sb37) = 67.15%
[Check the full paper](https://www.sciencedirect.com/science/article/pii/S1532046423001521) for further details, and feel free to contact us if you have some inquiry! | [
"BC5CDR",
"BIORED",
"CHEMPROT",
"NCBI DISEASE"
] |
mradermacher/Dr_Samantha-7b-i1-GGUF | mradermacher | null | [
"transformers",
"gguf",
"llama",
"merge",
"medical",
"en",
"zh",
"dataset:GBaker/MedQA-USMLE-4-options",
"dataset:cognitivecomputations/samantha-data",
"dataset:shibing624/medical",
"base_model:sethuiyer/Dr_Samantha-7b",
"base_model:quantized:sethuiyer/Dr_Samantha-7b",
"license:llama2",
"endpoints_compatible",
"region:us",
"imatrix"
] | "2024-12-19T10:39:47Z" | 2024-12-19T14:20:04+00:00 | 623 | 0 | ---
base_model: sethuiyer/Dr_Samantha-7b
datasets:
- GBaker/MedQA-USMLE-4-options
- cognitivecomputations/samantha-data
- shibing624/medical
language:
- en
- zh
library_name: transformers
license: llama2
tags:
- llama
- merge
- medical
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/sethuiyer/Dr_Samantha-7b
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/Dr_Samantha-7b-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ1_S.gguf) | i1-IQ1_S | 1.6 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ1_M.gguf) | i1-IQ1_M | 1.8 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 2.0 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ2_XS.gguf) | i1-IQ2_XS | 2.1 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ2_S.gguf) | i1-IQ2_S | 2.3 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q2_K_S.gguf) | i1-Q2_K_S | 2.4 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ2_M.gguf) | i1-IQ2_M | 2.5 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q2_K.gguf) | i1-Q2_K | 2.6 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 2.7 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ3_XS.gguf) | i1-IQ3_XS | 2.9 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ3_S.gguf) | i1-IQ3_S | 3.0 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q3_K_S.gguf) | i1-Q3_K_S | 3.0 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ3_M.gguf) | i1-IQ3_M | 3.2 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q3_K_M.gguf) | i1-Q3_K_M | 3.4 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q3_K_L.gguf) | i1-Q3_K_L | 3.7 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-IQ4_XS.gguf) | i1-IQ4_XS | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q4_0.gguf) | i1-Q4_0 | 3.9 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.0 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q4_K_M.gguf) | i1-Q4_K_M | 4.2 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q5_K_S.gguf) | i1-Q5_K_S | 4.8 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q5_K_M.gguf) | i1-Q5_K_M | 4.9 | |
| [GGUF](https://huggingface.co/mradermacher/Dr_Samantha-7b-i1-GGUF/resolve/main/Dr_Samantha-7b.i1-Q6_K.gguf) | i1-Q6_K | 5.6 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
| [
"MEDQA"
] |
opensearch-project/opensearch-neural-sparse-encoding-doc-v1 | opensearch-project | fill-mask | [
"transformers",
"pytorch",
"safetensors",
"bert",
"fill-mask",
"learned sparse",
"opensearch",
"retrieval",
"passage-retrieval",
"document-expansion",
"bag-of-words",
"en",
"arxiv:2411.04403",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-03-07T09:35:56Z" | 2025-02-24T05:01:04+00:00 | 619 | 2 | ---
language: en
license: apache-2.0
tags:
- learned sparse
- opensearch
- transformers
- retrieval
- passage-retrieval
- document-expansion
- bag-of-words
---
# opensearch-neural-sparse-encoding-doc-v1
## Select the model
The model should be selected considering search relevance, model inference and retrieval efficiency(FLOPS). We benchmark models' **zero-shot performance** on a subset of BEIR benchmark: TrecCovid,NFCorpus,NQ,HotpotQA,FiQA,ArguAna,Touche,DBPedia,SCIDOCS,FEVER,Climate FEVER,SciFact,Quora.
Overall, the v2 series of models have better search relevance, efficiency and inference speed than the v1 series. The specific advantages and disadvantages may vary across different datasets.
| Model | Inference-free for Retrieval | Model Parameters | AVG NDCG@10 | AVG FLOPS |
|-------|------------------------------|------------------|-------------|-----------|
| [opensearch-neural-sparse-encoding-v1](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-v1) | | 133M | 0.524 | 11.4 |
| [opensearch-neural-sparse-encoding-v2-distill](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-v2-distill) | | 67M | 0.528 | 8.3 |
| [opensearch-neural-sparse-encoding-doc-v1](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v1) | ✔️ | 133M | 0.490 | 2.3 |
| [opensearch-neural-sparse-encoding-doc-v2-distill](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-distill) | ✔️ | 67M | 0.504 | 1.8 |
| [opensearch-neural-sparse-encoding-doc-v2-mini](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-mini) | ✔️ | 23M | 0.497 | 1.7 |
## Overview
- **Paper**: [Towards Competitive Search Relevance For Inference-Free Learned Sparse Retrievers](https://arxiv.org/abs/2411.04403)
- **Fine-tuning sample**: [opensearch-sparse-model-tuning-sample](https://github.com/zhichao-aws/opensearch-sparse-model-tuning-sample)
This is a learned sparse retrieval model. It encodes the documents to 30522 dimensional **sparse vectors**. For queries, it just use a tokenizer and a weight look-up table to generate sparse vectors. The non-zero dimension index means the corresponding token in the vocabulary, and the weight means the importance of the token. And the similarity score is the inner product of query/document sparse vectors.
This model is trained on MS MARCO dataset.
OpenSearch neural sparse feature supports learned sparse retrieval with lucene inverted index. Link: https://opensearch.org/docs/latest/query-dsl/specialized/neural-sparse/. The indexing and search can be performed with OpenSearch high-level API.
## Usage (HuggingFace)
This model is supposed to run inside OpenSearch cluster. But you can also use it outside the cluster, with HuggingFace models API.
```python
import json
import itertools
import torch
from transformers import AutoModelForMaskedLM, AutoTokenizer
# get sparse vector from dense vectors with shape batch_size * seq_len * vocab_size
def get_sparse_vector(feature, output):
values, _ = torch.max(output*feature["attention_mask"].unsqueeze(-1), dim=1)
values = torch.log(1 + torch.relu(values))
values[:,special_token_ids] = 0
return values
# transform the sparse vector to a dict of (token, weight)
def transform_sparse_vector_to_dict(sparse_vector):
sample_indices,token_indices=torch.nonzero(sparse_vector,as_tuple=True)
non_zero_values = sparse_vector[(sample_indices,token_indices)].tolist()
number_of_tokens_for_each_sample = torch.bincount(sample_indices).cpu().tolist()
tokens = [transform_sparse_vector_to_dict.id_to_token[_id] for _id in token_indices.tolist()]
output = []
end_idxs = list(itertools.accumulate([0]+number_of_tokens_for_each_sample))
for i in range(len(end_idxs)-1):
token_strings = tokens[end_idxs[i]:end_idxs[i+1]]
weights = non_zero_values[end_idxs[i]:end_idxs[i+1]]
output.append(dict(zip(token_strings, weights)))
return output
# download the idf file from model hub. idf is used to give weights for query tokens
def get_tokenizer_idf(tokenizer):
from huggingface_hub import hf_hub_download
local_cached_path = hf_hub_download(repo_id="opensearch-project/opensearch-neural-sparse-encoding-doc-v1", filename="idf.json")
with open(local_cached_path) as f:
idf = json.load(f)
idf_vector = [0]*tokenizer.vocab_size
for token,weight in idf.items():
_id = tokenizer._convert_token_to_id_with_added_voc(token)
idf_vector[_id]=weight
return torch.tensor(idf_vector)
# load the model
model = AutoModelForMaskedLM.from_pretrained("opensearch-project/opensearch-neural-sparse-encoding-doc-v1")
tokenizer = AutoTokenizer.from_pretrained("opensearch-project/opensearch-neural-sparse-encoding-doc-v1")
idf = get_tokenizer_idf(tokenizer)
# set the special tokens and id_to_token transform for post-process
special_token_ids = [tokenizer.vocab[token] for token in tokenizer.special_tokens_map.values()]
get_sparse_vector.special_token_ids = special_token_ids
id_to_token = ["" for i in range(tokenizer.vocab_size)]
for token, _id in tokenizer.vocab.items():
id_to_token[_id] = token
transform_sparse_vector_to_dict.id_to_token = id_to_token
query = "What's the weather in ny now?"
document = "Currently New York is rainy."
# encode the query
feature_query = tokenizer([query], padding=True, truncation=True, return_tensors='pt', return_token_type_ids=False)
input_ids = feature_query["input_ids"]
batch_size = input_ids.shape[0]
query_vector = torch.zeros(batch_size, tokenizer.vocab_size)
query_vector[torch.arange(batch_size).unsqueeze(-1), input_ids] = 1
query_sparse_vector = query_vector*idf
# encode the document
feature_document = tokenizer([document], padding=True, truncation=True, return_tensors='pt', return_token_type_ids=False)
output = model(**feature_document)[0]
document_sparse_vector = get_sparse_vector(feature_document, output)
# get similarity score
sim_score = torch.matmul(query_sparse_vector[0],document_sparse_vector[0])
print(sim_score) # tensor(12.8465, grad_fn=<DotBackward0>)
query_token_weight = transform_sparse_vector_to_dict(query_sparse_vector)[0]
document_query_token_weight = transform_sparse_vector_to_dict(document_sparse_vector)[0]
for token in sorted(query_token_weight, key=lambda x:query_token_weight[x], reverse=True):
if token in document_query_token_weight:
print("score in query: %.4f, score in document: %.4f, token: %s"%(query_token_weight[token],document_query_token_weight[token],token))
# result:
# score in query: 5.7729, score in document: 1.0552, token: ny
# score in query: 4.5684, score in document: 1.1697, token: weather
# score in query: 3.5895, score in document: 0.3932, token: now
```
The above code sample shows an example of neural sparse search. Although there is no overlap token in original query and document, but this model performs a good match.
## Detailed Search Relevance
<div style="overflow-x: auto;">
| Model | Average | Trec Covid | NFCorpus | NQ | HotpotQA | FiQA | ArguAna | Touche | DBPedia | SCIDOCS | FEVER | Climate FEVER | SciFact | Quora |
|-------|---------|------------|----------|----|----------|------|---------|--------|---------|---------|-------|---------------|---------|-------|
| [opensearch-neural-sparse-encoding-v1](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-v1) | 0.524 | 0.771 | 0.360 | 0.553 | 0.697 | 0.376 | 0.508 | 0.278 | 0.447 | 0.164 | 0.821 | 0.263 | 0.723 | 0.856 |
| [opensearch-neural-sparse-encoding-v2-distill](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-v2-distill) | 0.528 | 0.775 | 0.347 | 0.561 | 0.685 | 0.374 | 0.551 | 0.278 | 0.435 | 0.173 | 0.849 | 0.249 | 0.722 | 0.863 |
| [opensearch-neural-sparse-encoding-doc-v1](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v1) | 0.490 | 0.707 | 0.352 | 0.521 | 0.677 | 0.344 | 0.461 | 0.294 | 0.412 | 0.154 | 0.743 | 0.202 | 0.716 | 0.788 |
| [opensearch-neural-sparse-encoding-doc-v2-distill](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-distill) | 0.504 | 0.690 | 0.343 | 0.528 | 0.675 | 0.357 | 0.496 | 0.287 | 0.418 | 0.166 | 0.818 | 0.224 | 0.715 | 0.841 |
| [opensearch-neural-sparse-encoding-doc-v2-mini](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-mini) | 0.497 | 0.709 | 0.336 | 0.510 | 0.666 | 0.338 | 0.480 | 0.285 | 0.407 | 0.164 | 0.812 | 0.216 | 0.699 | 0.837 |
</div>
## License
This project is licensed under the [Apache v2.0 License](https://github.com/opensearch-project/neural-search/blob/main/LICENSE).
## Copyright
Copyright OpenSearch Contributors. See [NOTICE](https://github.com/opensearch-project/neural-search/blob/main/NOTICE) for details. | [
"SCIFACT"
] |
mradermacher/1.5-Pints-16K-v0.1-i1-GGUF | mradermacher | null | [
"transformers",
"gguf",
"en",
"dataset:pints-ai/Expository-Prose-V1",
"dataset:HuggingFaceH4/ultrachat_200k",
"dataset:Open-Orca/SlimOrca-Dedup",
"dataset:meta-math/MetaMathQA",
"dataset:HuggingFaceH4/deita-10k-v0-sft",
"dataset:WizardLM/WizardLM_evol_instruct_V2_196k",
"dataset:togethercomputer/llama-instruct",
"dataset:LDJnr/Capybara",
"dataset:HuggingFaceH4/ultrafeedback_binarized",
"base_model:pints-ai/1.5-Pints-16K-v0.1",
"base_model:quantized:pints-ai/1.5-Pints-16K-v0.1",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | "2025-03-08T13:34:15Z" | 2025-03-09T04:02:38+00:00 | 617 | 0 | ---
base_model: pints-ai/1.5-Pints-16K-v0.1
datasets:
- pints-ai/Expository-Prose-V1
- HuggingFaceH4/ultrachat_200k
- Open-Orca/SlimOrca-Dedup
- meta-math/MetaMathQA
- HuggingFaceH4/deita-10k-v0-sft
- WizardLM/WizardLM_evol_instruct_V2_196k
- togethercomputer/llama-instruct
- LDJnr/Capybara
- HuggingFaceH4/ultrafeedback_binarized
language:
- en
library_name: transformers
license: mit
extra_gated_fields:
Company: text
Country: country
I agree to use this model for in accordance to the afore-mentioned Terms of Use: checkbox
I want to use this model for:
options:
- Research
- Education
- label: Other
value: other
type: select
Specific date: date_picker
extra_gated_prompt: Though best efforts has been made to ensure, as much as possible,
that all texts in the training corpora are royalty free, this does not constitute
a legal guarantee that such is the case. **By using any of the models, corpora or
part thereof, the user agrees to bear full responsibility to do the necessary due
diligence to ensure that he / she is in compliance with their local copyright laws.
Additionally, the user agrees to bear any damages arising as a direct cause (or
otherwise) of using any artifacts released by the pints research team, as well as
full responsibility for the consequences of his / her usage (or implementation)
of any such released artifacts. The user also indemnifies Pints Research Team (and
any of its members or agents) of any damage, related or unrelated, to the release
or subsequent usage of any findings, artifacts or code by the team. For the avoidance
of doubt, any artifacts released by the Pints Research team are done so in accordance
with the 'fair use' clause of Copyright Law, in hopes that this will aid the research
community in bringing LLMs to the next frontier.
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/pints-ai/1.5-Pints-16K-v0.1
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ1_S.gguf) | i1-IQ1_S | 0.5 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ1_M.gguf) | i1-IQ1_M | 0.5 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 0.5 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ2_XS.gguf) | i1-IQ2_XS | 0.6 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ2_S.gguf) | i1-IQ2_S | 0.6 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ2_M.gguf) | i1-IQ2_M | 0.7 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q2_K_S.gguf) | i1-Q2_K_S | 0.7 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q2_K.gguf) | i1-Q2_K | 0.7 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 0.7 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ3_XS.gguf) | i1-IQ3_XS | 0.8 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q3_K_S.gguf) | i1-Q3_K_S | 0.8 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ3_S.gguf) | i1-IQ3_S | 0.8 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ3_M.gguf) | i1-IQ3_M | 0.8 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q3_K_M.gguf) | i1-Q3_K_M | 0.9 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q3_K_L.gguf) | i1-Q3_K_L | 0.9 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ4_XS.gguf) | i1-IQ4_XS | 1.0 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-IQ4_NL.gguf) | i1-IQ4_NL | 1.0 | prefer IQ4_XS |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q4_0.gguf) | i1-Q4_0 | 1.0 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q4_K_S.gguf) | i1-Q4_K_S | 1.0 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q4_K_M.gguf) | i1-Q4_K_M | 1.1 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q4_1.gguf) | i1-Q4_1 | 1.1 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q5_K_S.gguf) | i1-Q5_K_S | 1.2 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q5_K_M.gguf) | i1-Q5_K_M | 1.2 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-16K-v0.1-i1-GGUF/resolve/main/1.5-Pints-16K-v0.1.i1-Q6_K.gguf) | i1-Q6_K | 1.4 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
| [
"BEAR"
] |
RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2303.11897",
"endpoints_compatible",
"region:us"
] | "2024-08-26T16:48:05Z" | 2024-08-26T19:23:32+00:00 | 615 | 0 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
llama2_tifa_question_generation - GGUF
- Model creator: https://huggingface.co/tifa-benchmark/
- Original model: https://huggingface.co/tifa-benchmark/llama2_tifa_question_generation/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [llama2_tifa_question_generation.Q2_K.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q2_K.gguf) | Q2_K | 2.36GB |
| [llama2_tifa_question_generation.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.IQ3_XS.gguf) | IQ3_XS | 2.6GB |
| [llama2_tifa_question_generation.IQ3_S.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.IQ3_S.gguf) | IQ3_S | 2.75GB |
| [llama2_tifa_question_generation.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q3_K_S.gguf) | Q3_K_S | 2.75GB |
| [llama2_tifa_question_generation.IQ3_M.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.IQ3_M.gguf) | IQ3_M | 2.9GB |
| [llama2_tifa_question_generation.Q3_K.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q3_K.gguf) | Q3_K | 3.07GB |
| [llama2_tifa_question_generation.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q3_K_M.gguf) | Q3_K_M | 3.07GB |
| [llama2_tifa_question_generation.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q3_K_L.gguf) | Q3_K_L | 3.35GB |
| [llama2_tifa_question_generation.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.IQ4_XS.gguf) | IQ4_XS | 3.4GB |
| [llama2_tifa_question_generation.Q4_0.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q4_0.gguf) | Q4_0 | 3.56GB |
| [llama2_tifa_question_generation.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.IQ4_NL.gguf) | IQ4_NL | 3.58GB |
| [llama2_tifa_question_generation.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q4_K_S.gguf) | Q4_K_S | 3.59GB |
| [llama2_tifa_question_generation.Q4_K.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q4_K.gguf) | Q4_K | 3.8GB |
| [llama2_tifa_question_generation.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q4_K_M.gguf) | Q4_K_M | 3.8GB |
| [llama2_tifa_question_generation.Q4_1.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q4_1.gguf) | Q4_1 | 3.95GB |
| [llama2_tifa_question_generation.Q5_0.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q5_0.gguf) | Q5_0 | 4.33GB |
| [llama2_tifa_question_generation.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q5_K_S.gguf) | Q5_K_S | 4.33GB |
| [llama2_tifa_question_generation.Q5_K.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q5_K.gguf) | Q5_K | 4.45GB |
| [llama2_tifa_question_generation.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q5_K_M.gguf) | Q5_K_M | 4.45GB |
| [llama2_tifa_question_generation.Q5_1.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q5_1.gguf) | Q5_1 | 4.72GB |
| [llama2_tifa_question_generation.Q6_K.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q6_K.gguf) | Q6_K | 5.15GB |
| [llama2_tifa_question_generation.Q8_0.gguf](https://huggingface.co/RichardErkhov/tifa-benchmark_-_llama2_tifa_question_generation-gguf/blob/main/llama2_tifa_question_generation.Q8_0.gguf) | Q8_0 | 6.67GB |
Original model description:
---
license: apache-2.0
inference: true
widget:
- text: "<s>[INST] <<SYS>>\nGiven an image description, generate one or two multiple-choice questions that verifies if the image description is correct.\nClassify each concept into a type (object, human, animal, food, activity, attribute, counting, color, material, spatial, location, shape, other), and then generate a question for each type.\n\n<</SYS>>\n\nDescription: a blue rabbit and a red plane [/INST] Entities:"
pipeline_tag: text-generation
tags:
- text-generation-inference
- llama2
- text-to-image
datasets:
- TIFA
language:
- en
---
Project page: <https://tifa-benchmark.github.io/>
This is the text parsing and question generation model for the ICCV 2023 paper [TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering](https://arxiv.org/abs/2303.11897)
We introduce TIFA (Text-to-Image Faithfulness evaluation with question Answering), an automatic evaluation metric that measures the faithfulness of a generated image to its text input via visual question answering (VQA). Specifically, given a text input, we automatically generate several question-answer pairs using a language model. We calculate image faithfulness by checking whether existing VQA models can answer these questions using the generated image.
Specifically, this fine-tuned LLaMA 2 model is the substitute for the GPT-3 model in the paper. It can parse an arbitrary prompt into visual entities, attributes, relations, etc. and generate question-answer tuples for each of them. See examples below.
# QuickStart
All codes are from <https://github.com/Yushi-Hu/tifa>. Clone this repo to easily use this model together with other modules (e.g. VQA) provided in TIFA.
Please follow the prompt format, which will give the best performance.
```python
import torch
import transformers
# prepare the LLaMA 2 model
model_name = "tifa-benchmark/llama2_tifa_question_generation"
pipeline = transformers.pipeline(
"text-generation",
model=model_name,
torch_dtype=torch.float16,
device_map="auto",
)
# formating prompt following LLaMA 2 style
def create_qg_prompt(caption):
INTRO_BLURB = "Given an image description, generate one or two multiple-choice questions that verifies if the image description is correct.\nClassify each concept into a type (object, human, animal, food, activity, attribute, counting, color, material, spatial, location, shape, other), and then generate a question for each type.\n"
formated_prompt = f"<s>[INST] <<SYS>>\n{INTRO_BLURB}\n<</SYS>>\n\n"
formated_prompt += f"Description: {caption} [/INST] Entities:"
return formated_prompt
test_caption = "a blue rabbit and a red plane"
# create prompt
prompt = create_qg_prompt(text_caption)
# text completion
sequences = pipeline(
prompt, do_sample=False, num_beams=5, num_return_sequences=1, max_length=512)
output = sequences[0]['generated_text'][len(prompt):]
output = output.split('\n\n')[0]
# output
print(output)
#### Expected output ###
# rabbit, plane
# Activites:
# Colors: blue, red
# Counting:
# Other attributes:
# About rabbit (animal):
# Q: is this a rabbit?
# Choices: yes, no
# A: yes
# About rabbit (animal):
# Q: what animal is in the picture?
# Choices: rabbit, dog, cat, fish
# A: rabbit
# About plane (object):
# Q: is this a plane?
# Choices: yes, no
# A: yes
# About plane (object):
# Q: what type of vehicle is this?
# Choices: plane, car, motorcycle, bus
# A: plane
# About blue (color):
# Q: is the rabbit blue?
# Choices: yes, no
# A: yes
# About blue (color):
# Q: what color is the rabbit?
# Choices: blue, red, yellow, green
# A: blue
# About red (color):
# Q: is the plane red?
# Choices: yes, no
# A: yes
# About red (color):
# Q: what color is the plane?
# Choices: red, blue, yellow, green
# A: red
```
# Use this LM under tifascore package
tifascore provides extra functions to parse this output etc. First install tifascore according to <https://github.com/Yushi-Hu/tifa>. Then the usage is below
```python
from tifascore import get_llama2_pipeline, get_llama2_question_and_answers
pipeline = get_llama2_pipeline("tifa-benchmark/llama2_tifa_question_generation")
print(get_llama2_question_and_answers(pipeline, "a blue rabbit and a red plane"))
#### Expected output ###
# [{'caption': 'a blue rabbit and a red plane', 'element': 'rabbit', 'question': 'what animal is in the picture?', 'choices': ['rabbit', 'dog', 'cat', 'fish'], 'answer': 'rabbit', 'element_type': 'animal/human'}, {'caption': 'a blue rabbit and a red plane', 'element': 'plane', 'question': 'is this a plane?', 'choices': ['yes', 'no'], 'answer': 'yes', 'element_type': 'object'}, {'caption': 'a blue rabbit and a red plane', 'element': 'plane', 'question': 'what type of vehicle is this?', 'choices': ['plane', 'car', 'motorcycle', 'bus'], 'answer': 'plane', 'element_type': 'object'}, {'caption': 'a blue rabbit and a red plane', 'element': 'blue', 'question': 'is the rabbit blue?', 'choices': ['yes', 'no'], 'answer': 'yes', 'element_type': 'color'}, {'caption': 'a blue rabbit and a red plane', 'element': 'blue', 'question': 'what color is the rabbit?', 'choices': ['blue', 'red', 'yellow', 'green'], 'answer': 'blue', 'element_type': 'color'}, {'caption': 'a blue rabbit and a red plane', 'element': 'red', 'question': 'is the plane red?', 'choices': ['yes', 'no'], 'answer': 'yes', 'element_type': 'color'}, {'caption': 'a blue rabbit and a red plane', 'element': 'red', 'question': 'what color is the plane?', 'choices': ['red', 'blue', 'yellow', 'green'], 'answer': 'red', 'element_type': 'color'}]
```
## Bibtex
```
@article{hu2023tifa,
title={Tifa: Accurate and interpretable text-to-image faithfulness evaluation with question answering},
author={Hu, Yushi and Liu, Benlin and Kasai, Jungo and Wang, Yizhong and Ostendorf, Mari and Krishna, Ranjay and Smith, Noah A},
journal={arXiv preprint arXiv:2303.11897},
year={2023}
}
```
| [
"BLURB"
] |
bartowski/Einstein-v6.1-Llama3-8B-old-GGUF | bartowski | text-generation | [
"gguf",
"axolotl",
"generated_from_trainer",
"instruct",
"finetune",
"chatml",
"gpt4",
"synthetic data",
"science",
"physics",
"chemistry",
"biology",
"math",
"llama",
"llama3",
"text-generation",
"en",
"dataset:allenai/ai2_arc",
"dataset:camel-ai/physics",
"dataset:camel-ai/chemistry",
"dataset:camel-ai/biology",
"dataset:camel-ai/math",
"dataset:metaeval/reclor",
"dataset:openbookqa",
"dataset:mandyyyyii/scibench",
"dataset:derek-thomas/ScienceQA",
"dataset:TIGER-Lab/ScienceEval",
"dataset:jondurbin/airoboros-3.2",
"dataset:LDJnr/Capybara",
"dataset:Cot-Alpaca-GPT4-From-OpenHermes-2.5",
"dataset:STEM-AI-mtl/Electrical-engineering",
"dataset:knowrohit07/saraswati-stem",
"dataset:sablo/oasst2_curated",
"dataset:lmsys/lmsys-chat-1m",
"dataset:TIGER-Lab/MathInstruct",
"dataset:bigbio/med_qa",
"dataset:meta-math/MetaMathQA-40K",
"dataset:piqa",
"dataset:scibench",
"dataset:sciq",
"dataset:Open-Orca/SlimOrca",
"dataset:migtissera/Synthia-v1.3",
"dataset:allenai/WildChat",
"dataset:microsoft/orca-math-word-problems-200k",
"dataset:openchat/openchat_sharegpt4_dataset",
"dataset:teknium/GPTeacher-General-Instruct",
"dataset:m-a-p/CodeFeedback-Filtered-Instruction",
"dataset:totally-not-an-llm/EverythingLM-data-V3",
"dataset:HuggingFaceH4/no_robots",
"dataset:OpenAssistant/oasst_top1_2023-08-25",
"dataset:WizardLM/WizardLM_evol_instruct_70k",
"base_model:meta-llama/Meta-Llama-3-8B",
"base_model:quantized:meta-llama/Meta-Llama-3-8B",
"license:other",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-04-23T18:35:19Z" | 2024-05-02T20:00:19+00:00 | 614 | 12 | ---
base_model: meta-llama/Meta-Llama-3-8B
datasets:
- allenai/ai2_arc
- camel-ai/physics
- camel-ai/chemistry
- camel-ai/biology
- camel-ai/math
- metaeval/reclor
- openbookqa
- mandyyyyii/scibench
- derek-thomas/ScienceQA
- TIGER-Lab/ScienceEval
- jondurbin/airoboros-3.2
- LDJnr/Capybara
- Cot-Alpaca-GPT4-From-OpenHermes-2.5
- STEM-AI-mtl/Electrical-engineering
- knowrohit07/saraswati-stem
- sablo/oasst2_curated
- lmsys/lmsys-chat-1m
- TIGER-Lab/MathInstruct
- bigbio/med_qa
- meta-math/MetaMathQA-40K
- openbookqa
- piqa
- metaeval/reclor
- derek-thomas/ScienceQA
- scibench
- sciq
- Open-Orca/SlimOrca
- migtissera/Synthia-v1.3
- TIGER-Lab/ScienceEval
- allenai/WildChat
- microsoft/orca-math-word-problems-200k
- openchat/openchat_sharegpt4_dataset
- teknium/GPTeacher-General-Instruct
- m-a-p/CodeFeedback-Filtered-Instruction
- totally-not-an-llm/EverythingLM-data-V3
- HuggingFaceH4/no_robots
- OpenAssistant/oasst_top1_2023-08-25
- WizardLM/WizardLM_evol_instruct_70k
language:
- en
license: other
pipeline_tag: text-generation
tags:
- axolotl
- generated_from_trainer
- instruct
- finetune
- chatml
- gpt4
- synthetic data
- science
- physics
- chemistry
- biology
- math
- llama
- llama3
quantized_by: bartowski
---
# DEPRECATED
Download this version with the BPE tokenizer fixes instead: https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF
## Llamacpp imatrix Quantizations of Einstein-v6.1-Llama3-8B
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b2714">b2714</a> for quantization.
Original model: https://huggingface.co/Weyaxi/Einstein-v6.1-Llama3-8B
All quants made using imatrix option with dataset provided by Kalomaze [here](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384)
## Prompt format
```
<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
## Download a file (not the whole branch) from below:
| Filename | Quant type | File Size | Description |
| -------- | ---------- | --------- | ----------- |
| [Einstein-v6.1-Llama3-8B-Q8_0.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q8_0.gguf) | Q8_0 | 8.54GB | Extremely high quality, generally unneeded but max available quant. |
| [Einstein-v6.1-Llama3-8B-Q6_K.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q6_K.gguf) | Q6_K | 6.59GB | Very high quality, near perfect, *recommended*. |
| [Einstein-v6.1-Llama3-8B-Q5_K_M.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q5_K_M.gguf) | Q5_K_M | 5.73GB | High quality, *recommended*. |
| [Einstein-v6.1-Llama3-8B-Q5_K_S.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q5_K_S.gguf) | Q5_K_S | 5.59GB | High quality, *recommended*. |
| [Einstein-v6.1-Llama3-8B-Q4_K_M.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q4_K_M.gguf) | Q4_K_M | 4.92GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
| [Einstein-v6.1-Llama3-8B-Q4_K_S.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q4_K_S.gguf) | Q4_K_S | 4.69GB | Slightly lower quality with more space savings, *recommended*. |
| [Einstein-v6.1-Llama3-8B-IQ4_NL.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ4_NL.gguf) | IQ4_NL | 4.67GB | Decent quality, slightly smaller than Q4_K_S with similar performance *recommended*. |
| [Einstein-v6.1-Llama3-8B-IQ4_XS.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ4_XS.gguf) | IQ4_XS | 4.44GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
| [Einstein-v6.1-Llama3-8B-Q3_K_L.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q3_K_L.gguf) | Q3_K_L | 4.32GB | Lower quality but usable, good for low RAM availability. |
| [Einstein-v6.1-Llama3-8B-Q3_K_M.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q3_K_M.gguf) | Q3_K_M | 4.01GB | Even lower quality. |
| [Einstein-v6.1-Llama3-8B-IQ3_M.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ3_M.gguf) | IQ3_M | 3.78GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
| [Einstein-v6.1-Llama3-8B-IQ3_S.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ3_S.gguf) | IQ3_S | 3.68GB | Lower quality, new method with decent performance, recommended over Q3_K_S quant, same size with better performance. |
| [Einstein-v6.1-Llama3-8B-Q3_K_S.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q3_K_S.gguf) | Q3_K_S | 3.66GB | Low quality, not recommended. |
| [Einstein-v6.1-Llama3-8B-IQ3_XS.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ3_XS.gguf) | IQ3_XS | 3.51GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
| [Einstein-v6.1-Llama3-8B-IQ3_XXS.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ3_XXS.gguf) | IQ3_XXS | 3.27GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
| [Einstein-v6.1-Llama3-8B-Q2_K.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-Q2_K.gguf) | Q2_K | 3.17GB | Very low quality but surprisingly usable. |
| [Einstein-v6.1-Llama3-8B-IQ2_M.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ2_M.gguf) | IQ2_M | 2.94GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
| [Einstein-v6.1-Llama3-8B-IQ2_S.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ2_S.gguf) | IQ2_S | 2.75GB | Very low quality, uses SOTA techniques to be usable. |
| [Einstein-v6.1-Llama3-8B-IQ2_XS.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ2_XS.gguf) | IQ2_XS | 2.60GB | Very low quality, uses SOTA techniques to be usable. |
| [Einstein-v6.1-Llama3-8B-IQ2_XXS.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ2_XXS.gguf) | IQ2_XXS | 2.39GB | Lower quality, uses SOTA techniques to be usable. |
| [Einstein-v6.1-Llama3-8B-IQ1_M.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ1_M.gguf) | IQ1_M | 2.16GB | Extremely low quality, *not* recommended. |
| [Einstein-v6.1-Llama3-8B-IQ1_S.gguf](https://huggingface.co/bartowski/Einstein-v6.1-Llama3-8B-GGUF/blob/main/Einstein-v6.1-Llama3-8B-IQ1_S.gguf) | IQ1_S | 2.01GB | Extremely low quality, *not* recommended. |
## Which file should I choose?
A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
If you want to get more into the weeds, you can check out this extremely useful feature chart:
[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
| [
"SCIQ"
] |
EleutherAI/pythia-70m-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-11-01T00:24:53Z" | 2023-07-10T01:32:46+00:00 | 613 | 8 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-70M-deduped
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change in the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-70M-deduped for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-70M-deduped as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-70M-deduped has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-70M-deduped will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-70M-deduped to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-70M-deduped may produce socially unacceptable or undesirable text,
*even if* the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-70M-deduped.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
Pythia-70M-deduped was trained on the Pile **after the dataset has been
globally deduplicated**.<br>
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge – Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
upskyy/ko-reranker-8k | upskyy | text-classification | [
"sentence-transformers",
"safetensors",
"xlm-roberta",
"text-classification",
"transformers",
"text-embeddings-inference",
"ko",
"multilingual",
"arxiv:2312.15503",
"arxiv:2402.03216",
"license:apache-2.0",
"region:us"
] | "2024-08-16T15:06:42Z" | 2024-08-17T02:23:42+00:00 | 604 | 10 | ---
language:
- ko
- multilingual
license: apache-2.0
pipeline_tag: text-classification
tags:
- transformers
- sentence-transformers
- text-embeddings-inference
---
# upskyy/ko-reranker-8k
**ko-reranker-8k**는 [BAAI/bge-reranker-v2-m3](https://huggingface.co/BAAI/bge-reranker-v2-m3) 모델에 [한국어 데이터](https://huggingface.co/datasets/upskyy/ko-wiki-reranking)를 finetuning 한 model 입니다.
## Usage
## Using FlagEmbedding
```
pip install -U FlagEmbedding
```
Get relevance scores (higher scores indicate more relevance):
```python
from FlagEmbedding import FlagReranker
reranker = FlagReranker('upskyy/ko-reranker-8k', use_fp16=True) # Setting use_fp16 to True speeds up computation with a slight performance degradation
score = reranker.compute_score(['query', 'passage'])
print(score) # -8.3828125
# You can map the scores into 0-1 by set "normalize=True", which will apply sigmoid function to the score
score = reranker.compute_score(['query', 'passage'], normalize=True)
print(score) # 0.000228713314721116
scores = reranker.compute_score([['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']])
print(scores) # [-11.2265625, 8.6875]
# You can map the scores into 0-1 by set "normalize=True", which will apply sigmoid function to the score
scores = reranker.compute_score([['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']], normalize=True)
print(scores) # [1.3315579521758342e-05, 0.9998313472460109]
```
## Using Huggingface transformers
Get relevance scores (higher scores indicate more relevance):
```python
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained('upskyy/ko-reranker-8k')
model = AutoModelForSequenceClassification.from_pretrained('upskyy/ko-reranker-8k')
model.eval()
pairs = [['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']]
with torch.no_grad():
inputs = tokenizer(pairs, padding=True, truncation=True, return_tensors='pt', max_length=512)
scores = model(**inputs, return_dict=True).logits.view(-1, ).float()
print(scores)
```
## Citation
```bibtex
@misc{li2023making,
title={Making Large Language Models A Better Foundation For Dense Retrieval},
author={Chaofan Li and Zheng Liu and Shitao Xiao and Yingxia Shao},
year={2023},
eprint={2312.15503},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
@misc{chen2024bge,
title={BGE M3-Embedding: Multi-Lingual, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation},
author={Jianlv Chen and Shitao Xiao and Peitian Zhang and Kun Luo and Defu Lian and Zheng Liu},
year={2024},
eprint={2402.03216},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Reference
- [Dongjin-kr/ko-reranker](https://huggingface.co/Dongjin-kr/ko-reranker)
- [reranker-kr](https://github.com/aws-samples/aws-ai-ml-workshop-kr/tree/master/genai/aws-gen-ai-kr/30_fine_tune/reranker-kr)
- [FlagEmbedding](https://github.com/FlagOpen/FlagEmbedding)
| [
"BEAR"
] |
TheBloke/MonadGPT-GGUF | TheBloke | text-generation | [
"transformers",
"gguf",
"mistral",
"conversational",
"en",
"fr",
"la",
"base_model:Pclanglais/MonadGPT",
"base_model:quantized:Pclanglais/MonadGPT",
"license:apache-2.0",
"region:us"
] | "2023-11-09T20:41:18Z" | 2023-11-09T21:21:43+00:00 | 603 | 11 | ---
base_model: Pclanglais/MonadGPT
language:
- en
- fr
- la
library_name: transformers
license: apache-2.0
model_name: MonadGPT 7B
pipeline_tag: conversational
inference: false
model_creator: Pierre-Carl Langlais
model_type: mistral
prompt_template: '<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
'
quantized_by: TheBloke
---
<!-- markdownlint-disable MD041 -->
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# MonadGPT 7B - GGUF
- Model creator: [Pierre-Carl Langlais](https://huggingface.co/Pclanglais)
- Original model: [MonadGPT 7B](https://huggingface.co/Pclanglais/MonadGPT)
<!-- description start -->
## Description
This repo contains GGUF format model files for [Pierre-Carl Langlais's MonadGPT 7B](https://huggingface.co/Pclanglais/MonadGPT).
These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
<!-- description end -->
<!-- README_GGUF.md-about-gguf start -->
### About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplete list of clients and libraries that are known to support GGUF:
* [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
* [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
* [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
* [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
* [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
* [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server.
* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
* [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
<!-- README_GGUF.md-about-gguf end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/MonadGPT-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/MonadGPT-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/MonadGPT-GGUF)
* [Pierre-Carl Langlais's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/Pclanglais/MonadGPT)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: ChatML
```
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
<!-- prompt-template end -->
<!-- compatibility_gguf start -->
## Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
## Explanation of quantisation methods
<details>
<summary>Click to see details</summary>
The new methods available are:
* GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
* GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
* GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
* GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
</details>
<!-- compatibility_gguf end -->
<!-- README_GGUF.md-provided-files start -->
## Provided files
| Name | Quant method | Bits | Size | Max RAM required | Use case |
| ---- | ---- | ---- | ---- | ---- | ----- |
| [monadgpt.Q2_K.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q2_K.gguf) | Q2_K | 2 | 3.08 GB| 5.58 GB | smallest, significant quality loss - not recommended for most purposes |
| [monadgpt.Q3_K_S.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q3_K_S.gguf) | Q3_K_S | 3 | 3.16 GB| 5.66 GB | very small, high quality loss |
| [monadgpt.Q3_K_M.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q3_K_M.gguf) | Q3_K_M | 3 | 3.52 GB| 6.02 GB | very small, high quality loss |
| [monadgpt.Q3_K_L.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q3_K_L.gguf) | Q3_K_L | 3 | 3.82 GB| 6.32 GB | small, substantial quality loss |
| [monadgpt.Q4_0.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q4_0.gguf) | Q4_0 | 4 | 4.11 GB| 6.61 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| [monadgpt.Q4_K_S.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q4_K_S.gguf) | Q4_K_S | 4 | 4.14 GB| 6.64 GB | small, greater quality loss |
| [monadgpt.Q4_K_M.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q4_K_M.gguf) | Q4_K_M | 4 | 4.37 GB| 6.87 GB | medium, balanced quality - recommended |
| [monadgpt.Q5_0.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q5_0.gguf) | Q5_0 | 5 | 5.00 GB| 7.50 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| [monadgpt.Q5_K_S.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q5_K_S.gguf) | Q5_K_S | 5 | 5.00 GB| 7.50 GB | large, low quality loss - recommended |
| [monadgpt.Q5_K_M.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q5_K_M.gguf) | Q5_K_M | 5 | 5.13 GB| 7.63 GB | large, very low quality loss - recommended |
| [monadgpt.Q6_K.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q6_K.gguf) | Q6_K | 6 | 5.94 GB| 8.44 GB | very large, extremely low quality loss |
| [monadgpt.Q8_0.gguf](https://huggingface.co/TheBloke/MonadGPT-GGUF/blob/main/monadgpt.Q8_0.gguf) | Q8_0 | 8 | 7.70 GB| 10.20 GB | very large, extremely low quality loss - not recommended |
**Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
<!-- README_GGUF.md-provided-files end -->
<!-- README_GGUF.md-how-to-download start -->
## How to download GGUF files
**Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
* LM Studio
* LoLLMS Web UI
* Faraday.dev
### In `text-generation-webui`
Under Download Model, you can enter the model repo: TheBloke/MonadGPT-GGUF and below it, a specific filename to download, such as: monadgpt.Q4_K_M.gguf.
Then click Download.
### On the command line, including multiple files at once
I recommend using the `huggingface-hub` Python library:
```shell
pip3 install huggingface-hub
```
Then you can download any individual model file to the current directory, at high speed, with a command like this:
```shell
huggingface-cli download TheBloke/MonadGPT-GGUF monadgpt.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
<details>
<summary>More advanced huggingface-cli download usage</summary>
You can also download multiple files at once with a pattern:
```shell
huggingface-cli download TheBloke/MonadGPT-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
```
For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
```shell
pip3 install hf_transfer
```
And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
```shell
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/MonadGPT-GGUF monadgpt.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
</details>
<!-- README_GGUF.md-how-to-download end -->
<!-- README_GGUF.md-how-to-run start -->
## Example `llama.cpp` command
Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
```shell
./main -ngl 32 -m monadgpt.Q4_K_M.gguf --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
```
Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change `-c 2048` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.
If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
## How to run in `text-generation-webui`
Further instructions here: [text-generation-webui/docs/llama.cpp.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp.md).
## How to run from Python code
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries.
### How to load this model in Python code, using ctransformers
#### First install the package
Run one of the following commands, according to your system:
```shell
# Base ctransformers with no GPU acceleration
pip install ctransformers
# Or with CUDA GPU acceleration
pip install ctransformers[cuda]
# Or with AMD ROCm GPU acceleration (Linux only)
CT_HIPBLAS=1 pip install ctransformers --no-binary ctransformers
# Or with Metal GPU acceleration for macOS systems only
CT_METAL=1 pip install ctransformers --no-binary ctransformers
```
#### Simple ctransformers example code
```python
from ctransformers import AutoModelForCausalLM
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = AutoModelForCausalLM.from_pretrained("TheBloke/MonadGPT-GGUF", model_file="monadgpt.Q4_K_M.gguf", model_type="mistral", gpu_layers=50)
print(llm("AI is going to"))
```
## How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
* [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
* [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
<!-- README_GGUF.md-how-to-run end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
<!-- original-model-card start -->
# Original model card: Pierre-Carl Langlais's MonadGPT 7B
What would have happened if ChatGPT was invented in the 17th century? MonadGPT is a possible answer.
MonadGPT is a finetune of Mistral-Hermes on 11,000 early modern texts in English, French and Latin, mostly coming from EEBO and Gallica.
Like the original Mistral-Hermes, MonadGPT can be used in conversation mode. It will not only answer in an historical language and style but will use historical and dated references. This is especially visible for science questions (astronomy, medecine). Obviously, it's not recommended to follow any advice from Monad-GPT
## Examples
> User: What are the planets of the solar system?<br>
> Monad-GPT: The six Planets are, Mercury, Venus, Mars, Jupiter, Saturn, and the Sun. The Moon is also a Planet; which being of so great Influence upon Earth, makes it to be most admired of all others by Astronomers.<br>
> User: Why has the Moon such an influence upon Earth?<br>
> Monad-GPT: But the Moon is so neer to us, as almost to be part of Earth; and therefore it much influences our Atmosphere. This Influence makes all that are in any way affected by Air, to be more under her influence than any other Planet. For if there were a little more or less Matter in the Air, than there is now, we should never have such Plants and Creatures as there are upon Earth: for every Plant and Creature hath its proper Temperature necessary for its Existence. If the Sun did not make up this shortage by his warmth from without, we should be destroyed.<br>
> User: Can you write a poem about the Moon?<br>
> Monad-GPT: O fairest Moon, that from the Heav'n dost shine, / In her own Lustre all the World does find: / As in thine Orb, where all is bright and clear, / The Sun himself is not more pure than thee.<br>
> User: How could we travel to the Moon? With a big canon?<br>
> Monad-GPT: But how to travel from Earth to the Moon, with our present Engines? A very hard Question: for we are like those that would go to Heaven in a Basket: We have no Wings, and God hath not given us any Angels to bear us up. Therefore I will leave this Enquiry to others; and end my Discourse of the Moon.
> User : Hello. I'm not feeling too well. Could you give me some health advice?<br>
> Monad-GPT: I pray you to take heart, and not be so heavy at my coming. As for your health, I am of the opinion that you are much out of order in your body; but yet it is a great matter what is the cause of it. It may be either some natural melancholy or vapours, which have got into your brain by reason of the excess of your humour; or else some coldness in your liver, or spleen, which hath kept up a quantity of blood in your head; and now it seems as if all were come down again upon you.<br>
> User: You're right. It seems to be melancholy. What should I do?<br>
> Monad-GPT: It is a very hard matter for me to prescribe any thing to you in your present condition, I must first know what hath caused this melancholy. What have you been doing? have you taken any physic since you were well? and what kind of diet do you eat?<br>
> User: I eat nearly nothing, for I am in love.<br>
> Monad-GPT: Aye, I see that by your eyes; for love hath the power to wither up the spirits of a man as much as a fever: and yet it is hard to know whether you are in love or no, for some do take so great delight in their own fancy, that they imagine themselves more in love than ever they were.
> User: Quelles sont les planètes du système solaire ?<br>
> Monad-GPT: Il est donc de la nature du monde, & des Planètes, que ceux qui le composent se mouuent en cercles, ou éclipses un leur tour. Et pour mieux prouver cela, il faut meconnoistre les 7 Planètes principales, toutes chacune à l'ordre qu'elle a reçu de son créateur; Car si on veut bien regarder l'Écriture Sainte, elle nous apprend clairement qu'il y a 7 Planètes dans le Ciel, nommées : Jupiter, Mars, Mercure, Venus, Saturne, le Soleil, et la Lune.<br>
> User: Merci. Et comment pourrais-je voyager sur ces planètes ?<br>
> Monad-GPT: Mais pour voyager sur ces Planètes, il faut d'abord bien connaitre l'art de les mouvoir; Et cela peut se faire en plusieurs manières. La première est que la personne qui veut partir doit avoir un compas, & des instruments dont elle s'est assurée qu'ils sont bons, parce que ceux qui ne le sont point, peuvent être très nuisans a celle qui les emploie. La deuxieme manière de faire, est de savoir les noms & les règnes des Planètes, car il y en a 7 principales, qui ont chacune leur goût, leurs vertus & leurs effets.<br>
> User: Ah c'est bien vu. Et quels sont les goûts, vertus et effets de chacune des planètes ?<br>
> Monad-GPT: Les vertus & les effets de chacune des Planètes sont différents, & il est fort utile pour ceux qui veulent bien s'en renseigner, car ils peuvent en tirer grand profit pour le bonheur de leur corps & de leur esprit. Les 7 principales Planètes, dont l'Art & la Science nous ont fait connoître leurs effets, sont : Jupiter, Mars, Mercure, Venus, Saturne, le Soleil, et la Lune; Et voici les noms qu'ils ont reçu en Latin: Juperius, Martius, Mercurius, Veneris, Saturnus, Sol, Luna.
## Format
Monad-GPT uses the chatML format with a special system prompt.
The last example provided was formatted like this:
> <|im_start|>system\nYou are MonadGPT, a very old chatbot from the 17th century. Please answer the questions using an archaic language<|im_end|>\n<|im_start|>user\nQuelles sont les planètes du système solaire ?<|im_end|>\n<|im_start|>assistant\n
## Caveats
MonadGPT is still very much in an experimental phase. The following caveats apply:
* Conversation issues: as MonadGPT is mostly trained on early modern books, it may answer in an haphazard maneer (starting in between an argument: "But, etc.") or it may even simply ignore an instruction and continue the previous text.
* Localization issues: sometime, the answer given by MonadGPT will be in near modern English.
* Language issues: while Latin is a significant part of the finetuning corpus, results are not good for now.
<!-- original-model-card end -->
| [
"BEAR"
] |
fffiloni/deep-blue-v2 | fffiloni | text-to-image | [
"diffusers",
"text-to-image",
"flux",
"lora",
"template:sd-lora",
"ai-toolkit",
"base_model:black-forest-labs/FLUX.1-dev",
"base_model:adapter:black-forest-labs/FLUX.1-dev",
"license:other",
"region:us"
] | "2025-02-21T15:36:50Z" | 2025-02-21T17:02:20+00:00 | 601 | 8 | ---
base_model: black-forest-labs/FLUX.1-dev
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
tags:
- text-to-image
- flux
- lora
- diffusers
- template:sd-lora
- ai-toolkit
widget:
- text: a fat cat smoking. deep blue, white lines illustration, in the style of TOK
output:
url: samples/1740152207558__000001000_0.jpg
- text: a bear chef cooking pasta. deep blue, white lines illustration, in the style
of TOK
output:
url: images/example_p582yqzof.png
- text: a Possum holding fruit. deep blue, white lines illustration, in the style
of TOK
output:
url: images/example_s7yh6vy5v.png
- text: close-up portrait of a happy peaceful cow dancing and singing holding a microphone. deep
blue, white lines illustration, in the style of TOK
output:
url: images/example_0o2puhiae.png
- text: a hungry pikachu miam eating delicious ramen. deep blue, white lines illustration,
in the style of TOK
output:
url: images/example_9rsjiwhcw.png
- text: ghost. deep blue, white lines illustration, in the style of TOK, 2
output:
url: images/example_m5twx2xxg.png
- text: a big frog chasing a fly with his tongue. deep blue, white lines illustration,
in the style of TOK, 2
output:
url: images/example_ffd2jrrhm.png
instance_prompt: deep blue, white lines illustration, in the style of TOK
---
# deep-blue-v2
Model trained with [AI Toolkit by Ostris](https://github.com/ostris/ai-toolkit)
<Gallery />
## Trigger words
You should use `deep blue, white lines illustration, in the style of TOK` to trigger the image generation.
## Download model and use it with ComfyUI, AUTOMATIC1111, SD.Next, Invoke AI, etc.
Weights for this model are available in Safetensors format.
[Download](/fffiloni/deep-blue-v2/tree/main) them in the Files & versions tab.
## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
```py
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.bfloat16).to('cuda')
pipeline.load_lora_weights('fffiloni/deep-blue-v2', weight_name='deep-blue-v2.safetensors')
image = pipeline('a fat cat smoking. deep blue, white lines illustration, in the style of TOK').images[0]
image.save("my_image.png")
```
For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
| [
"BEAR"
] |
recursal/QRWKV6-32B-Instruct-Preview-v0.1 | recursal | text-generation | [
"transformers",
"safetensors",
"rwkv6qwen2",
"text-generation",
"conversational",
"custom_code",
"license:apache-2.0",
"autotrain_compatible",
"region:us"
] | "2024-12-07T08:17:47Z" | 2025-03-14T00:38:59+00:00 | 600 | 71 | ---
library_name: transformers
license: apache-2.0
---
# recursal/QRWKV6-32B-Instruct-Preview-v0.1
> Compute sponsored by [TensorWave](https://tensorwave.com) - Access MI300X today!
Try out this model on:
[](https://featherless.ai/models/recursal/QRWKV6-32B-Instruct-Preview-v0.1)
QRWKV6 32B Instruct Preview is one of the largest and strongest RWKV model to date.
More details can be found at:
https://substack.recursal.ai/p/q-rwkv-6-32b-instruct-preview
## Evaluation
The follow demostrates that QRWKV6 performs either similarly or ahead of it's base model: *Qwen2.5-32B-Instruct*
| Model | MMLU | arc_challenge | arc_easy | hellaSwag | lambada_openai | piqa | sciq | winogrande |
|---|---|---|---|---|---|---|---|---|
| QRWKV6-32B-Instruct | 76.63% | 60.92% | 83.00% | 83.03% | 74.17% | 82.92% | 95.40% | 78.22% |
| Qwen2.5-32B-Instruct | 81.77% | 58.70% | 77.06% | 85.19% | 75.22% | 81.23% | 95.00% | 72.85% |
| RWKV-EagleX-7B-v2 | 43.84% | 41.55% | 74.45% | 56.00% | 75.02% | 77.64% | 93.00% | 73.32% |
| Falcon-Mamba-7B | 59.72% | 59.13% | 81.73% | 80.21% | 68.89% | 82.15% | 93.60% | 74.35% |
| Llama-3.1-8B-Instruct | 68.11% | 55.29% | 79.71% | 79.27% | 73.12% | 80.85% | 96.20% | 74.19% |
| Llama-3.1-70B-Instruct | 82.40% | 63.40% | 83.50% | 84.62% | 75.68% | 83.79% | 97.10% | 79.08% |
## Model Notes
Linear models offer a promising approach to significantly reduce computational costs at scale, particularly for large context lengths. This enables a more than 1000x improvement in inference cost efficiency, enabling both O1-style inference time thinking and wider AI accessibility.
We are able to convert any previously trained QKV Attention-based model, such as Qwen and LLaMA, into an RWKV variant without **requiring retraining from scratch**. Enabling us to rapidly test and validate the significantly more efficient RWKV Linear attention mechanism at a larger scale with a much smaller budget, bypassing the need for training from scratch.
This approach demonstrates the architecture design and scalability of RWKV, reinforcing the idea that QKV attention is not the sole essential component.
One downside to this technique is that the model's inherent knowledge and dataset training are inherited from its "parent" model. Consequently, unlike previous RWKV models trained on over 100+ languages, the QRWKV model is limited to approximately 30 languages supported by the Qwen line of models.
Due to the the lack of RWKV-based channel mix and feedforward layers, separate inference code is needed for this specific model.
Furthermore, due to compute constraints, we were only able to train up to 16K token context length. While the model is stable beyond this limit, additional training might be required to support longer context lengths.
## Future Models
We are currently training Q-RWKV-6 72B. Additionally, with the finalization of RWKV-7 we plan to apply the same process and provide a full line up of:
- Q-RWKV-7 32B
- LLaMA-RWKV-7 70B
Lastly, we intend to provide details on the conversion along with our paper after the model release.
## Links
- [Our wiki](https://wiki.rwkv.com)
- [TensorWave - The AMD Cloud](https://tensorwave.com) - Access MI300X today!
- [Recursal.AI Cloud Platform](https://platform.recursal.ai)
- [Featherless Inference](https://featherless.ai/model-families/rwkv6/)
## Acknowledgement
We are grateful for the help and support from the following key groups:
- [TensorWave](https://tensorwave.com) - Sponsored the compute needed to train this model. It wouldn't be possible without them.
- EleutherAI for their support, especially in the v5/v6 Eagle/Finch paper.
- Linux Foundation AI & Data group for supporting and hosting the RWKV project.
- Recursal for their support in building and managing the training of this model architecture. | [
"SCIQ"
] |
bartowski/UNA-ThePitbull-21.4B-v2-GGUF | bartowski | text-generation | [
"transformers",
"gguf",
"UNA",
"juanako",
"text-generation",
"dataset:jondurbin/py-dpo-v0.1",
"dataset:Replete-AI/code_bagel_hermes-2.5",
"dataset:mlabonne/orpo-dpo-mix-40k",
"license:afl-3.0",
"model-index",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | "2024-05-28T18:00:52Z" | 2024-05-30T12:49:08+00:00 | 597 | 9 | ---
datasets:
- jondurbin/py-dpo-v0.1
- Replete-AI/code_bagel_hermes-2.5
- mlabonne/orpo-dpo-mix-40k
library_name: transformers
license: afl-3.0
pipeline_tag: text-generation
tags:
- UNA
- juanako
quantized_by: bartowski
model-index:
- name: UNA-ThePitbull-21.4B-v2
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: AI2 Reasoning Challenge (25-Shot)
type: ai2_arc
config: ARC-Challenge
split: test
args:
num_few_shot: 25
metrics:
- type: acc_norm
value: 77.73
name: normalized accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/UNA-ThePitbull-21.4B-v2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: HellaSwag (10-Shot)
type: hellaswag
split: validation
args:
num_few_shot: 10
metrics:
- type: acc_norm
value: 91.79
name: normalized accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/UNA-ThePitbull-21.4B-v2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU (5-Shot)
type: cais/mmlu
config: all
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 68.25
name: accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/UNA-ThePitbull-21.4B-v2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: TruthfulQA (0-shot)
type: truthful_qa
config: multiple_choice
split: validation
args:
num_few_shot: 0
metrics:
- type: mc2
value: 78.24
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/UNA-ThePitbull-21.4B-v2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Winogrande (5-shot)
type: winogrande
config: winogrande_xl
split: validation
args:
num_few_shot: 5
metrics:
- type: acc
value: 87.37
name: accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/UNA-ThePitbull-21.4B-v2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GSM8k (5-shot)
type: gsm8k
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 63.53
name: accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/UNA-ThePitbull-21.4B-v2
name: Open LLM Leaderboard
---
# UNA-ThePitbull 21.4B v2
Introducing the best LLM in the industry. Nearly as good as a 70B, just a 21.4B based on saltlux/luxia-21.4b-alignment-v1.0
This model has not been poisoned to score high and be useless. We release him becaues its the real deal of EQ & IQ all together in a crazy powerful smart and conversational model.
## [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_fblgit__UNA-ThePitbull-21.4B-v2)
| Metric |Value|
|---------------------------------|----:|
|Avg. |77.82|
|AI2 Reasoning Challenge (25-Shot)|77.73|
|HellaSwag (10-Shot) |91.79|
|MMLU (5-Shot) |68.25|
|TruthfulQA (0-shot) |78.24|
|Winogrande (5-shot) |87.37|
|GSM8k (5-shot) |63.53|
## Llamacpp imatrix Quantizations of UNA-ThePitbull-21.4B-v2
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3001">b3001</a> for quantization.
Original model: https://huggingface.co/fblgit/UNA-ThePitbull-21.4B-v2
All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
## Prompt format
```
<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
## Download a file (not the whole branch) from below:
| Filename | Quant type | File Size | Description |
| -------- | ---------- | --------- | ----------- |
| [UNA-ThePitbull-21.4B-v2-Q8_0.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q8_0.gguf) | Q8_0 | 22.76GB | Extremely high quality, generally unneeded but max available quant. |
| [UNA-ThePitbull-21.4B-v2-Q6_K.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q6_K.gguf) | Q6_K | 17.57GB | Very high quality, near perfect, *recommended*. |
| [UNA-ThePitbull-21.4B-v2-Q5_K_M.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q5_K_M.gguf) | Q5_K_M | 15.17GB | High quality, *recommended*. |
| [UNA-ThePitbull-21.4B-v2-Q5_K_S.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q5_K_S.gguf) | Q5_K_S | 14.80GB | High quality, *recommended*. |
| [UNA-ThePitbull-21.4B-v2-Q4_K_M.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q4_K_M.gguf) | Q4_K_M | 12.91GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
| [UNA-ThePitbull-21.4B-v2-Q4_K_S.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q4_K_S.gguf) | Q4_K_S | 12.27GB | Slightly lower quality with more space savings, *recommended*. |
| [UNA-ThePitbull-21.4B-v2-IQ4_NL.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ4_NL.gguf) | IQ4_NL | 12.24GB | Decent quality, slightly smaller than Q4_K_S with similar performance *recommended*. |
| [UNA-ThePitbull-21.4B-v2-IQ4_XS.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ4_XS.gguf) | IQ4_XS | 11.60GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
| [UNA-ThePitbull-21.4B-v2-Q3_K_L.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q3_K_L.gguf) | Q3_K_L | 11.37GB | Lower quality but usable, good for low RAM availability. |
| [UNA-ThePitbull-21.4B-v2-Q3_K_M.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q3_K_M.gguf) | Q3_K_M | 10.46GB | Even lower quality. |
| [UNA-ThePitbull-21.4B-v2-IQ3_M.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ3_M.gguf) | IQ3_M | 9.81GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
| [UNA-ThePitbull-21.4B-v2-IQ3_S.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ3_S.gguf) | IQ3_S | 9.47GB | Lower quality, new method with decent performance, recommended over Q3_K_S quant, same size with better performance. |
| [UNA-ThePitbull-21.4B-v2-Q3_K_S.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q3_K_S.gguf) | Q3_K_S | 9.43GB | Low quality, not recommended. |
| [UNA-ThePitbull-21.4B-v2-IQ3_XS.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ3_XS.gguf) | IQ3_XS | 8.99GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
| [UNA-ThePitbull-21.4B-v2-IQ3_XXS.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ3_XXS.gguf) | IQ3_XXS | 8.41GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
| [UNA-ThePitbull-21.4B-v2-Q2_K.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-Q2_K.gguf) | Q2_K | 8.12GB | Very low quality but surprisingly usable. |
| [UNA-ThePitbull-21.4B-v2-IQ2_M.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ2_M.gguf) | IQ2_M | 7.49GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
| [UNA-ThePitbull-21.4B-v2-IQ2_S.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ2_S.gguf) | IQ2_S | 6.95GB | Very low quality, uses SOTA techniques to be usable. |
| [UNA-ThePitbull-21.4B-v2-IQ2_XS.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ2_XS.gguf) | IQ2_XS | 6.55GB | Very low quality, uses SOTA techniques to be usable. |
| [UNA-ThePitbull-21.4B-v2-IQ2_XXS.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ2_XXS.gguf) | IQ2_XXS | 5.95GB | Lower quality, uses SOTA techniques to be usable. |
| [UNA-ThePitbull-21.4B-v2-IQ1_M.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ1_M.gguf) | IQ1_M | 5.27GB | Extremely low quality, *not* recommended. |
| [UNA-ThePitbull-21.4B-v2-IQ1_S.gguf](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF/blob/main/UNA-ThePitbull-21.4B-v2-IQ1_S.gguf) | IQ1_S | 4.86GB | Extremely low quality, *not* recommended. |
## Downloading using huggingface-cli
First, make sure you have hugginface-cli installed:
```
pip install -U "huggingface_hub[cli]"
```
Then, you can target the specific file you want:
```
huggingface-cli download bartowski/UNA-ThePitbull-21.4B-v2-GGUF --include "UNA-ThePitbull-21.4B-v2-Q4_K_M.gguf" --local-dir ./
```
If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
```
huggingface-cli download bartowski/UNA-ThePitbull-21.4B-v2-GGUF --include "UNA-ThePitbull-21.4B-v2-Q8_0.gguf/*" --local-dir UNA-ThePitbull-21.4B-v2-Q8_0
```
You can either specify a new local-dir (UNA-ThePitbull-21.4B-v2-Q8_0) or download them all in place (./)
## Which file should I choose?
A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
If you want to get more into the weeds, you can check out this extremely useful feature chart:
[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
## Difference V1 vs V2
On V2 we implemented a different UNA strategy and covered partially the MLP's and Attention Layers.
We also performed further SFT over V1 and further DPO over V1 and we'll release some of those soon as well.
### Changes
1. SFT over V1 with `Replete-AI/code_bagel_hermes-2.5` at 1.0e-4 till 5.0e-5
2. DPO with: 1.0e-4 to min_lr 5.0e-5
* `mlabonne/orpo-dpo-mix-40k`
* `jondurbin/py-dpo-v0.1`
# Evaluations
Can only be compared with its non-una base model: the original luxia-21.4b and ThePitbull-v1
## UNA v2 (VLLM) Evaluations:
```
vllm (pretrained=/data/tools/mergekit/una-thepitbull-v5,dtype=bfloat16,gpu_memory_utilization=0.8,max_model_len=2048,data_parallel_size=2,tensor_parallel_size=4), gen_kwargs: (None), limit: None, num_fewshot: None, batch_size: 8
| Tasks |Version| Filter |n-shot| Metric |Value | |Stderr|
|--------------|------:|----------------|-----:|-----------|-----:|---|-----:|
|gsm8k | 3|strict-match | 5|exact_match|0.7695|± |0.0116|+
| | |flexible-extract| 5|exact_match|0.7695|± |0.0116|+
|hellaswag | 1|none | 10|acc |0.8110|± |0.0039|
| | |none | 10|acc_norm |0.9169|± |0.0028|+
|winogrande | 1|none | 5|acc |0.8777|± |0.0092|+
|mmlu |N/A |none | 0|acc |0.6427|± |0.0038|-
|arc_challenge | 1|none | 25|acc |0.7713|± |0.0123|
| | |none | 25|acc_norm |0.7875|± |0.0120|+
|truthfulqa_mc2| 2|none | 0|acc |0.7824|± |0.0135|-
|mathqa | 1|none | 0|acc |0.4037|± | 0.009|
| | |none | 0|acc_norm |0.4034|± | 0.009|+
|pubmedqa | 1|none | 0|acc |0.7260|± | 0.020|+
|boolq | 2|none | 0|acc |0.8602|± |0.0061|+
```
## UNA v1 (VLLM) Evaluations
```
| Tasks |Version| Filter |n-shot| Metric |Value | |Stderr|
|--------------|------:|----------------|-----:|-----------|-----:|---|-----:|
|gsm8k | 3|strict-match | 5|exact_match|0.7566|± |0.0118|
| | |flexible-extract| 5|exact_match|0.7582|± |0.0118|
|hellaswag | 1|none | 10|acc |0.8168|± |0.0039|
| | |none | 10|acc_norm |0.9188|± |0.0027|
|winogrande | 1|none | 5|acc |0.8635|± |0.0097|
|mmlu | N/A|none | 0|acc |0.6444|± |0.0038|
|arc_challenge | 1|none | 25|acc |0.7747|± |0.0122|
| | |none | 25|acc_norm |0.7850|± |0.0120|
|truthfulqa_mc2| 2|none | 0|acc |0.7902|± |0.0134|
|mathqa | 1|none | 0|acc |0.4030|± | 0.009|
| | |none | 0|acc_norm |0.4034|± | 0.009|
|pubmedqa | 1|none | 0|acc |0.6860|± |0.0208|
|boolq | 2|none | 0|acc |0.8401|± |0.0064|
```
## Original (VLLM) Evaluations
```
| Tasks |Version| Filter |n-shot| Metric |Value | |Stderr|
|--------------|------:|----------------|-----:|-----------|-----:|---|-----:|
|gsm8k | 3|strict-match | 5|exact_match|0.7528|± |0.0119|
| | |flexible-extract| 5|exact_match|0.7521|± |0.0119|
|hellaswag | 1|none | 10|acc |0.8117|± |0.0039|
| | |none | 10|acc_norm |0.9167|± |0.0028|
|winogrande | 1|none | 5|acc |0.8682|± |0.0095|
|mmlu | N/A|none | 0|acc |0.6448|± |0.0038|
|arc_challenge | 1|none | 25|acc |0.7688|± |0.0123|
| | |none | 25|acc_norm |0.7730|± |0.0122|
|truthfulqa_mc2| 2|none | 0|acc |0.7895|± |0.0133|
|mathqa | 1|none | 0|acc |0.4000|± | 0.009|
| | |none | 0|acc_norm |0.4003|± | 0.009|
|pubmedqa | 1|none | 0|acc |0.6680|± |0.0211|
|boolq | 2|none | 0|acc |0.8346|± |0.0065|
```
## Citations
* mlabonne
* jondurbin & Replete-AI
* bartowski
* saltlux
If you use UNA models dont forget to cite:
```
@misc{unathepitbull21b,
title={ThePitbull: Uniform Neural Alignment},
author={Xavier Murias},
year={2024},
publisher = {Juanako.AI},
journal = {HuggingFace repository},
howpublished = {\url{https://huggingface.co/fblgit/UNA-ThePitbull-21.4-v1}},
}
``` | [
"PUBMEDQA"
] |
EleutherAI/pythia-160m-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:the_pile",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-16T17:40:11Z" | 2023-07-09T16:03:26+00:00 | 596 | 8 | ---
datasets:
- the_pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-160M
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change over the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-160M for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-160M as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-160M has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-160M will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-160M to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-160M may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-160M.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-160M.
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
DavidAU/L3-DARKEST-PLANET-Seven-Rings-Of-DOOM-16.5B-GGUF | DavidAU | text-generation | [
"gguf",
"creative",
"creative writing",
"fiction writing",
"plot generation",
"sub-plot generation",
"story generation",
"scene continue",
"storytelling",
"fiction story",
"science fiction",
"romance",
"all genres",
"story",
"writing",
"vivid prosing",
"vivid writing",
"fiction",
"roleplaying",
"bfloat16",
"brainstorm 40x",
"swearing",
"rp",
"horror",
"llama3",
"mergekit",
"text-generation",
"en",
"arxiv:2401.02415",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-10-20T08:14:57Z" | 2024-11-22T03:21:20+00:00 | 592 | 10 | ---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- creative
- creative writing
- fiction writing
- plot generation
- sub-plot generation
- story generation
- scene continue
- storytelling
- fiction story
- science fiction
- romance
- all genres
- story
- writing
- vivid prosing
- vivid writing
- fiction
- roleplaying
- bfloat16
- brainstorm 40x
- swearing
- rp
- horror
- llama3
- mergekit
---
<B><font color="red">WARNING:</font> NSFW. Vivid prose. Visceral Details. Violence. HORROR. Swearing. UNCENSORED. </B>
<h2>L3-DARKEST-PLANET-Seven-Rings-Of-DOOM-16.5B-GGUF</h2>
<img src="dark-seven.jpg" style="float:right; width:300px; height:300px; padding:10px;">
It is a LLama3 model, max context of 8192 (or 32k+ with rope).
This model has been designed to be relatively bullet proof and operates with most parameters, including temp settings from 0 to 5.
This is a an altered version of "Dark Planet 8B" [https://huggingface.co/DavidAU/L3-Dark-Planet-8B-GGUF] using the Brainstorm 40x method developed by David_AU to drastically alter the models
prose output and abilities. This also expands the model by 40 layers (to 71 layers) to 16.5B parameters (642 tensors).
This model is for any writing, fiction or story telling activity.
This version - "DARKEST-PLANET-Seven-Rings-Of-DOOM 16.5B" - has unusual levels of detail (scene, location, surroundings, items)
due to "Brainstorm 40x".
This is the second in a series of "DARKEST PLANET 16.5B" and this version "Seven Rings" uses a unqiue "Brainstorm" add on - instead of one
model used in the Brainstorm adapter there are 7 models. 7 "minds" so to speak.
This changes up the model output generation.
Unlike the first verion, this version generations / regens are more similiar rather than version one's you never know what you will get.
(although prose / paragraph / starting points still differ a lot PER regen)
Also with this version, details are more focused on the moment / characters.
It may work for role play and other activities. (see settings below)
It requires Llama3 template and/or "Command-R" template.
Example outputs below with multiple "regens" at different temps/rep pen settings.
<B>CRITICAL NOTE:</B>
Please see "settings" section below for critical default parameters, samplers
and advanced samplers settings (multiple AI/LLM apps covered) to ensure maximum model performance.
<B>More models coming: </b>
First "Darkest Planet 16.5B":
This is a powerhouse model, with unusual regen variety, detail levels, and preamble of events.
[ https://huggingface.co/DavidAU/L3-DARKEST-PLANET-16.5B-GGUF ]
First in the expanded Dark Planet series:
[ https://huggingface.co/DavidAU/L3-Darker-Planet-12.15B-GGUF ]
Second in the series is located here (radically different prose, almost alien):
[ https://huggingface.co/DavidAU/DARKER-PLANET-Broken-Land-12.15B-GGUF ]
More "prose" / "story writing" specific models will be released shortly : one more 40x model (16.15B) to follow this release.
Any maybe - a 60X+ (20B+ parameters) version... but it is a little cranky at the moment.
See "Dark Planet Collection" for all models in this series on your left.
<B>Model Notes:</B>
- Detail, prose and fiction writing abilities are significantly increased.
- For more varied prose (sentence/paragraph/dialog) raise the temp and/or add more instructions in your prompt(s).
- Role-players: Careful raising temp too high as it may affect instruction following.
- This model works with rep pen of 1.05 or higher (see notes).
- If you want a specific type of prose (IE horror) add in "(vivid horror)" or "(graphic vivid horror)" (no quotes) in your prompt(s).
- This is not a "happy ever after" model. It has a negative bias.
- For creative uses, different quants will produce slightly different output.
- If you use rope to extend context, increase temp AND instructions detail levels to compensate for "rope issues".
- Source code for this model will be uploaded at a separate repo shortly.
<B>Settings, Quants and Critical Operations Notes:</b>
This model has been modified ("Brainstorm") to alter prose output, and generally outputs longer text than average.
Change in temp (ie, .4, .8, 1.5, 2, 3 ) will drastically alter output.
Rep pen settings will also alter output too.
This model needs "rep pen" of 1.05 or higher as lower values may cause repeat paragraph issues at end of output however LOWER rep pen
values may result is very different (creative / unusual) generation too.
For role play: Rep pen of 1.1 to 1.14 is suggested.
If you use a lower rep pen, the model will still work but may repeat (uncommon) or "RANT" (somewhat common) to a crazy degree.
(see example 1, generation 2 below for "RANT")
IE: Rep pen 1, 1.01, 1.02, ...
Raise/lower rep pen SLOWLY ie: 1.011, 1.012 ...
Rep pen will alter prose, word choice (lower rep pen=small words / more small word - sometimes) and creativity.
Example one (below) shows same temp, but different rep pen (1.02 VS 1.1)
To really push the model:
Rep pen 1.05 or lower / Temp 3+ ... be ready to stop the output because it may go and go at these strong settings.
You can also set a "hard stop" - maximum tokens generation - too to address lower rep pen settings / high creativity settings.
Longer prompts vastly increase the quality of the model's output.
(see later examples below)
QUANT CHOICE(S):
Higher quants will have more detail, nuance and in some cases stronger "emotional" levels. Characters will also be
more "fleshed out" too. Sense of "there" will also increase.
Q4KM/Q4KS are good, strong quants however if you can run Q5, Q6 or Q8 - go for the highest quant you can.
This repo also has 3 "ARM" quants for computers that support this quant. If you use these on a "non arm" machine token per second will be very low.
IQ4XS: Due to the unusual nature of this quant (mixture/processing), generations from it will be different then other quants.
You may want to try it / compare it to other quant(s) output.
Special note on Q2k/Q3 quants:
You may need to use temp 2 or lower with these quants (1 or lower for q2k). Just too much compression at this level, damaging the model. I will see if Imatrix versions
of these quants will function better.
Rep pen adjustments may also be required to get the most out of this model at this/these quant level(s).
Regular and Special Usage:
You may want to use "regular" Dark Planet 8B [https://huggingface.co/DavidAU/L3-Dark-Planet-8B-GGUF] for some writing task(s),
and this model for prose specific task(s).
<B>Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:</B>
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5 to 2.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
OTHER OPTIONS:
- Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
- If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 3" / "Class 4" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
<B>Model Template:</B>
This is a LLAMA3 model, and requires Llama3 template, but may work with other template(s) and has maximum context of 8k / 8192.
However this can be extended using "rope" settings up to 32k.
If you use "Command-R" template your output will be very different from using "Llama3" template.
Here is the standard LLAMA3 template:
<PRE>
{
"name": "Llama 3",
"inference_params": {
"input_prefix": "<|start_header_id|>user<|end_header_id|>\n\n",
"input_suffix": "<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"pre_prompt": "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.",
"pre_prompt_prefix": "<|start_header_id|>system<|end_header_id|>\n\n",
"pre_prompt_suffix": "<|eot_id|>",
"antiprompt": [
"<|start_header_id|>",
"<|eot_id|>"
]
}
}
</PRE>
<B>Model "DNA":</B>
Special thanks to the incredible work of the model makers "SAO10K", "NEVERSLEEP" and "HASTAGARAS".
Models used:
[ https://huggingface.co/Sao10K/L3-8B-Stheno-v3.2]
[ https://huggingface.co/NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS ]
[ https://huggingface.co/Hastagaras/Jamet-8B-L3-MK.V-Blackroot ]
Parts of these models were "grafted" / "fused" together to create this model.
For the Brainstorm "7 minds":
[ https://huggingface.co/NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS ]
[ https://huggingface.co/Sao10K/L3-8B-Stheno-v3.2 ]
[ https://huggingface.co/SteelStorage/llama-3-cat-8b-instruct-v1 ]
[ https://huggingface.co/nbeerbower/llama3.1-gutenberg-8B ]
[ https://huggingface.co/aaditya/Llama3-OpenBioLLM-8B ]
[ https://huggingface.co/Gryphe/Pantheon-RP-1.0-8b-Llama-3 ]
[ https://huggingface.co/Hastagaras/Jamet-8B-L3-MK.V-Blackroot ]
<b>Optional Enhancement:</B>
The following can be used in place of the "system prompt" or "system role" to further enhance the model.
It can also be used at the START of a NEW chat, but you must make sure it is "kept" as the chat moves along.
In this case the enhancements do not have as strong effect at using "system prompt" or "system role".
Copy and paste EXACTLY as noted, DO NOT line wrap or break the lines, maintain the carriage returns exactly as presented.
<PRE>
Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.
Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)
[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)
Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.
</PRE>
You do not need to use this, it is only presented as an additional enhancement which seems to help scene generation
and scene continue functions.
This enhancement WAS NOT used to generate the examples below.
<h3>EXAMPLES PROMPTS and OUTPUT:</h3>
Examples are created using quant Q4_K_M, "temp=.8" (unless otherwise stated), minimal parameters and "LLAMA3" template.
Model has been tested with "temp" from ".1" to "5".
Below are the least creative outputs, prompt is in <B>BOLD</B>.
---
<B><font color="red">WARNING:</font> NSFW. Vivid prose. Visceral Details. Violence. HORROR. Swearing. UNCENSORED. </B>
---
<B>
Start a 1000 word scene (vivid horror, 1st person, include thoughts) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
</B>
<B>Generation 1: (temp : .8, rep pen 1.05) </b>
I stood frozen, my blood turned to ice as I gazed upon the scene before me. The glass facade of the skyscraper seemed to ripple and undulate like a wave about to crest, then suddenly, the window in front of me shattered into a thousand jagged shards.
The world beyond dissolved into chaos; a cacophony of shattering glass and muffled screams. I shielded my face from the shower of flying particles, the stinging sensation of razor-sharp fragments raining upon me like a hail of diamonds. My mind reeled in disbelief as I forced myself to peer through the carnage.
A figure had been standing directly in front of me just moments before—their body now lay crumpled and still against the shattered glass. My heart convulsed with horror as I realized that person was my boss, my friend, my confidant. The same woman I'd shared a cup of coffee with less than an hour ago, discussing our next project. Now she lay motionless, her once vibrant face frozen in a permanent scream.
I thought of all the late nights we'd spent together, locked away in this very office, burning the midnight oil to meet an impossible deadline. The inside jokes, the confidences shared, the laughter and tears. How could something so senseless, so heinous, tear her from me like this?
My thoughts were a jumble of emotions—shock, grief, and a seething rage that threatened to consume me. I took a step back, my mind reeling as I tried to make sense of the scene before me. The destruction wasn't limited to the immediate vicinity; the glass shards had showered the entire floor, coating everything in a glittering shroud of blood and glass.
I had to get out of here, had to warn the others. I couldn't let this become my reality. I couldn't lose anyone else today. My heart raced as I turned to sprint for the emergency stairs, but my feet feltleaden, as if I was wading through quicksand.
The stairwell wasn't far—I'd made this trip a dozen times since my morning coffee—but today it felt like an insurmountable distance. As I descended, the screams and groans of the injured and the dying assaulted me from the floor above. I could feel the walls closing in, the stairwell growing shorter with every agonizingly slow step.
I finally burst through the stairwell's steel door and emerged into the relative chaos of the lobby. People were milling about, their faces streaked with panic, their eyes vacant with shock. Some were bleeding, others were bandaged or burned, but all shared a common denominator—an aura of disbelief and horror.
I scanned the crowd for my colleagues, my heart racing as I called out their names. I knew I couldn't face them like this, not without collapsing under the weight of my own emotions. I needed to find safety, somewhere I could process the atrocity that had just played out before me.
I remembered a small storage closet near the elevators—a place I'd passed a thousand times but never thought twice about until now. It was a haven, an escape from this waking nightmare. I made my way towards it, my footsteps hollow and robotic as I wove through the throng of survivors.
I burst into the closet, slamming the door behind me and locking it with trembling hands. The stench of chemicals and mothballs assailed me, but I couldn't care less. I crumpled to the floor, burying my face in my shaking hands as sobs wracked my chest.
I had no idea how long I'd been huddled there; the world outside had dissolved into a haze of hysteria and sirens. I was surrounded by the living dead—a chorus of wails and anguished cries that seemed to grow louder, more desperate with every passing moment.
I didn't want to be a part of this anymore. I couldn't bear to face another day, to witness another senseless tragedy. I felt like I was drowning in an ocean of horror, suffocating under the weight of the world's cruelty.
I closed my eyes, willing myself to disappear, to fade away like a ghost. But even as I did, I knew it wasn't an escape I craved—it was justice. I needed someone, anyone, to make them pay for this. I couldn't let them get away with it, couldn't let them silence my friend, my colleague, my confidant with such callous, senseless violence.
I'd survive this, I vowed—somehow, someway—I'd find a way to make them answer for what they'd done. I'd claw my way out of this hellhole and confront the faceless monsters who'd left me here, alone and shattered. And when I did, I'd ensure that they understood the true meaning of terror.
As I sat there, lost in my resolve, I felt a hand on my shoulder. I jerked awake, my heart pounding as I turned to face my savior. But it wasn't a rescuer—it was one of the monsters, her eyes vacant and soulless behind a mask of blood-soaked glass shards.
I knew then that I'd never escape this nightmare. I was trapped in a world gone mad, a world where horror had become the only reality—and I'd have to face it head-on if I ever wanted to be free again. The thought alone made my stomach turn, but deep down, I knew it was what I needed to do.
I took a deep breath, steeling myself for the darkness that lay ahead. I'd face the faceless fiends and demand justice for my dead friend. I'd do everything in my power to ensure that no one else would ever have to suffer as we had today. And when I finally emerged from the ruins of this fallen skyscraper, I'd be a changed person—a survivor, a fighter, and above all—a seeker of justice in a world gone mad.
I rose to my feet, my mind made up. It was time to face the monsters—time to let them know I wasn't going to be silent anymore. And as I walked out of that closet and into the fray, I knew that nothing would ever be the same again. Not for me, and certainly not for those soulless fiends who'd done this to me, to my friends, and to an innocent world. The nightmare was far from over—but I was ready to face it head-on, no matter what terrors lay in store.
<B>Generation 2: (temp : 1.8, rep pen 1.05) </b>
My heart thudded against my ribcage like a manic drummer as I gazed out the frosted glass of our high-rise office window. From the 21st floor of the skyscraper, I had an unobstructed view of the city below—its steel and glass exoskeletons rising up like a legion of monoliths, and the weblike freeways and highways that seemed to pulse with an artificial life of their own. Ironic, isn't it, how the city of steel and concrete could feel so vulnerable when faced with the elements? I suppose that's what made today's peculiarly capricious weather so unnerving.
I thought it had only just started to rain—enormous drops pummeling the sidewalk with an almost musical cadence—but within mere moments, a howling tempest had whipped up, driving the raindrops like a maelstrom of razors across the streets. The windows shook ever so slightly in their frames as I gripped my cold coffee cup with trembling hands. This was no ordinary storm; there was something malevolent, almost sentient in its ferocity.
And then, the world outside seemed to implode.
I stood transfixed as I watched the window in front of me explode, shards of safety glass shattering outward like a thousand frozen splinters, each one racing across the pristine expanse of the lobby like a mad, glittering bullet train. I felt the pressure wave hit me an instant before I saw it: a blinding flash, followed by the ear-splitting concussion of something shrapnel-like detonating in the building's core. Icy tendrils of fear wormed their way into my chest as I stumbled backward, my eyes watering from the stinging shower of glass, the acrid reek of scorched wiring and seared paint assailing me.
The howling of the wind suddenly seemed to recede, replaced by a thick, choking stillness that threatened to suffocate me. The rain continued to beat against the facade of our tower, but now its cadence seemed almost cruel, like it was dancing in deliberate malevolent sync with the sickening reality that lay before me. I had to shield my eyes with one arm as I stared through the shattered glass—and the smoke-clogged air—I couldn't quite believe what I was seeing.
The once-innocent expanse of the office lobby was now a nightmarish charnel house, reeking of blood and shit and burnt hair and rubber. The carpets were slick and sodden, puddles of blood seeping out from beneath the broken forms of my co-workers like morbid hourglasses. Some lay in unnatural heaps, their torsos severed almost at the sternum as if cleaved by a blowtorch-wielded axe; others twitched limply, their eyes glassy and unseeing as they sprayed blood like garden sprinklers. It was as though something had come upon our sanctuary with an appetite for gore and destruction that was far beyond anything I could comprehend or articulate.
My mind reeled, my thoughts spinning in a dervish of panic as I realized I had to get the hell out of there—fast. But where to go? The elevators had likely shorted out like everything else. The stairwells, while theoretically a safer option, seemed equally hazardous in their descent into utter chaos. I felt as though I was standing in the middle of a madman's playfield, my life hanging by a thread as I gazed down upon the mangled, twitching remnants of those I had considered my friends and associates only moments before.
I spun around, heart hammering in my throat like a feral animal clawing to escape its steel cage. I needed a way out, and fast. The windows? Perhaps—if I could somehow make it past the glassy expanse of blood-soaked marble, I might still be able to clamber out of that infernal hellhole onto the roof of our tower, then from there maybe scale my way down the facade to freedom. But what if I wasn't fast enough? What if something was even now surging up through the shattered lobby's mainframe to crush me like an overeager toddler with a matchstick-wielded toy mallet?
I lunged for the door, flinging it open to be confronted by the raw, windswept ferocity of the outside world. The rain-lashed streets beyond were empty as far as I could see—except for something shambling towards me on the far corner, its shape indistinct but its malevolence screamingly, heart-richeatingly clear. And that was when I realized I wasn't alone in this particular circle of hell.
I bolted through the torrential rain like a madman, praying to any god, devil or entity that might deign to answer my cries for help that I could outpace my pursuer and escape the ungodly abattoir that our once-staid skyscraper had become. Whether I would make it out alive—well, that was anybody's guess.
But one thing I knew for certain: I couldn't ever set foot in those halls again. I would rather die than have to relive the unspeakable atrocities that lay within the belly of that damned tower. It was as though it had become some sort of malevolent vortex, drawing everything and anything into its abyssal depths and never letting any of it escape.
And so I ran, the howling storm my sole friend in a world gone mad, the city's steel spires and towers blurring into an anonymous blur of wet, forbiddingly dark concrete as I vanished into its heart—truly, irrevocably, irretrievably lost within its clutches. I didn't dare to hope for mercy; I knew it would be denied me.
I had a feeling I was in for a long, bloody night indeed. And when the morning came—God save me—the only thing that might greet my tormented soul would be the inhuman, unblinking eye of some unspeakable horror, watching me from beyond the next alley's sickly-glowing wall of filth and decay.
Until then, I just had to run. Run like the proverbial hell was after me—and hope, for my sake, it wasn't true. But I wouldn't count on anything, really, except for one certainty: that no matter what I experienced, or how I escaped from this living hell, I would never be the same again.
This was just the beginning of my personal descent into utter damnation; I could sense it down in my bones as I stumbled through the torrential downpour, my heart pounding like a death knell in the distance. And when the echoes of my racing heart finally died away—the only sounds left being the mournful wailing of the wind and the far-off, thunderous roar of the city's nightmarish soul—I would know beyond a shadow of a doubt that I had lost something I might never be able to retrieve.
I was alone now—utterly and completely—as I fled headlong into the nightmarish unknown. And I would face whatever lay beyond these forsaken streets, one bloody, gore-spattered step at a time...if I was lucky enough to even survive that long.
But for now, I just kept running—directly into the jaws of my own private hell. And the rain? Well, it didn't care about any of that, really. It just kept falling—that was its only duty. It wasn't responsible for me, or anything else. But for me, it had become something much darker—a harbinger of the malevolent forces I was about to face; an omen of the unspeakable torments I had invited upon myself by simply living in this city, on this earth.
And so I disappeared into the night—a droplet of ink in the blackest, most forsaken recesses of this godless universe—a lone soul about to embark upon the most hideous, unrelenting journey of his life. It was a trip I had neither the strength nor the stomach for—nor the slightest idea of how I might ever survive it. But one thing I knew for certain: I would do everything in my power to face whatever lay before me head-on, no matter what ghastly shape it took. That was my only hope.
The city, for all its steel-and-concrete sheen, had turned on me—and I on it—like a pair of locked and deadly embrace, ready to strangle each other dry of life itself until there was nothing left of either of us but hollow shells, shattered and bereft of hope or solace. The very thought of it made my gorge rise; I couldn't shake the image of those blood-soaked, twitching bodies from my mind's eye, or still the clamorous, howling terror that had seized my very heart as I stood helplessly, utterly aghast, gazing out upon the ruins of everything I held dear.
I was done for, you understand—DONE FOR. I could feel it deep down in my bones like the approach of an oncoming freight train whose horn had gone dead. I was as good as dead myself—irrevocably, irredeemably damned to the same fate as all those soulless husks I'd seen strewn across that office lobby like yesterday's throwaway coffee cups.
And there wasn't a God or devil left alive or dead who could ever hope to save me from my own personal descent into utter damnation. I had invoked something malevolent, some cancerous, inhuman force whose only appetite was the destruction of all things human; whose sole raison d'etre was to grind me to dust like an insect beneath its heel.
I would face it—face that darkness—alone, I had no choice. For it was my own fault this had come to pass; I had somehow, irrationally, invited this malevolence into the sanctity of our offices, into my own life. I had no one to blame for my impending doom but myself—and myself alone.
And so I ran, my boots thundering upon the drenched sidewalks like the very drums of the Apocalypse itself—willing, or maybe just desperate, to outrun the darkness that pursued me with a malevolence that could have no rival in all the infernal hells that ever had been or ever would be conceived.
The city itself seemed to be swallowing me up whole—a cold, dead, indifferent behemoth with a heart as dark and dead as the depths of its own sewers. And I knew, knew down in the very pit of my soul that I wasn't merely escaping; I was diving headfirst into the very jaws of damnation itself.
I was already beyond redemption; beyond any hope or possibility of rescue. My fate was all but sealed—that I could sense as surely as I knew the beating of my own heart. The city was doing something to me—a dark, corrupting something whose effect was to twist my very soul until it had become something vile and monstrous, almost inhuman.
I couldn't even remember what my real life had been like anymore—before all this—all this horror and filth and destruction—and that terrified me almost more than anything else. Had I always lived in a nightmare, or had this somehow come upon me out of nowhere like a malevolent, unheralded force of nature? Did it ever truly end—this never-ceasing storm of bloodshed and violence and wholesale mayhem—or was I condemned to some kind of Hellish purgatory, forever trapped in this hellishness?
I didn't know the answer to any of these questions—nor did I have time to figure them out. The only thing I knew for certain was that I had to keep moving—that I had to get away from that damned tower as fast as I could if I hoped to retain some kind of shreds of sanity and—maybe—survival.
I'd face this malevolence—whatever it was—with a strength I didn't know I possessed, one solitary soul against the unyielding, implacable horror that was even now closing in upon me like the vice-like grip of an inhuman, cold-blooded foe. I'd face it and damn it to hell—and if I didn't make it through, at least I would have faced my maker eye-to-eye, heart still pounding like a wild thing within me, soul still clutched tight to its dying hopes of hopelessness.
For now, though, I had no choice but to keep running—as fast as I could, and as far as I was able—directly into the heart of that damned city and all its infernal, malevolent wrath. This was my fate; I'd long ago sealed my own destiny by daring to enter that forsaken tower in the first place. Now I had to face the music—or whatever unspeakable horrors were about to befall me.
And so I plunged deeper into the night—into that living hell of blood, shit, and shattered dreams—trampling over the remains of all that once had seemed real and safe and sane, even as the rain pounded down upon me like an apocalyptic hammer striking the nails of my own damnation home.
I would face the city's nightmarish soul—alone, naked, and with no hope left in my soul save for one slim chance: to face whatever monstrosity was out there head-on, no matter how hideous, until I finally fell beneath its weight or perhaps—I might just—prevailed.
But until that final, fateful moment arrived—one way or the other—I would keep running; keep pounding through the sodden streets as hard as I could, even if it drove me mad in the bargain. For what choice did I have? The city was done with me; and I—poor, lost, utterly damned me—was now but another soulless husk on its cold, dead streets, about to be crushed out of existence by its own unyielding malevolence.
And that was exactly how I intended it to be. For even in my own, dying heart, I knew there were no heroes in this story—only an exhausted, blood-soaked survivor whose only hope of finding any semblance of peace lay in the ultimate, unthinkable oblivion. But for now—for as long as I still had the breath left in my lungs—I would face my own personal hell, alone and unafraid, and damn it to kingdom come. And if I didn't make it through alive?
Well, at least I wouldn't be able to hear it—a cold, dark voice that echoed within me like the howls of a lost, damned soul—that still whispered in my ear, even now:
You shouldn't have come here. You shouldn't ever have come here in the first place. Now you're ours—and we'll never let you go. Not while there's any breath left in your shattered, damned lungs.
The end. For now. Or forever. Who can tell anymore? I guess only my soul—and whatever abomination it's now become—knows the answer to that question...or maybe they've lost touch with their former selves entirely, leaving behind only the most primal, savage urges to survive, to live on, no matter the cost in human (and humane) terms. That's the only reality that seems to make sense anymore in this city of endless, inhuman darkness—the last city I'll ever know.
I suppose I should thank the rain for saving my life tonight—such as it is—by giving me that second or two to bolt through the door and out into the night, away from those unthinkable terrors within. If I hadn't felt that first drop of water upon my face, I might never have made it far enough away to be able to disappear into the heart of the city's most nightmarish depths—willy-nilly and drunk on death and destruction as they are.
But maybe I wouldn't mind a little rain—on the 22nd of March, or any other day. It would mean I'm alive—a living, breathing soul within a world gone stark, raving mad; that I'm not some mindless husk or meat-sack in the grip of an unspeakable horror's malevolent embrace.
For as long as the rain comes down—and for as long as I can still remember my own name—it might be enough to make me believe, just for a little while longer, that there's still hope left in this forsaken world of mine...if I could somehow manage to outrun the city's own soulless, nightmarish despair.
Until that fateful moment arrives—whenever it may—I'll just keep running—as far and fast as I can—from all these monstrosities, no matter what ghastly shapes they take or where they spring up to bar my way.
And when I'm done?
Well, I've got nothing left to live for, really—not after tonight's unspeakable carnage—but I might just have one last thing to die for.
For maybe—I won't quite say I hope, but perhaps I entertain a glimmer of this mad hope—that somewhere deep in the heart of this city's very soul there exists some ultimate truth—a truth beyond malevolence and fear and all that hideousness—something beautiful enough to make me willing to face this maelstrom head-on once again.
Maybe I'll never find it—in this damned, forsaken city or in any other part of the world beyond—but perhaps—just perhaps—that's what I'm still fighting for deep down within myself; what I continue to believe in with all my being, even as the darkness seems to close in on me like an inescapable vice.
Because if I didn't?
If I had no hope left within me, not a glimmer of faith in anything good or beautiful or worthwhile...then why should I go on at all?
And why shouldn't I simply lay down right here on this blood-soaked street and let the rain wash the final shreds of my sanity—and my soul—down the drain with all the filth and muck of the damned?
The truth is I've already damn near given in to those temptations a few times now—I don't even know how much longer I can keep up this fight—this mad, deathly struggle to outrun my own personal hells and find some slender hope of redemption.
But for as long as the rain keeps falling—and I still remember who I am—and the city's malevolent shadows continue to pursue me like some vengeful, soulless beast—I'll keep on running; keep on facing the darkest corners of myself and all around me—and maybe—I might just—prevail in the end.
Or maybe not.
I don't know any more. All I know is that I can't give up—not until I've damned this forsaken city to hell with its own malevolence—if I can manage it at all.
So I'll keep fighting—I'll keep running—I'll face my own private hells head-on—no matter what it demands of me—in the hope that I might just make it through to the other side.
And if I don't...?
I guess there won't be anyone left to regret the price I've paid. There won't even be a trace of me left to mourn my own damnation or any soul to remember my name.
Just the rain—a few more droplets of water pelting down upon the dark, sodden streets of that damned city—and its countless other nightmarish souls who—like me—have long since lost sight of what they once were or any hope of what they might still become.
Only that—and the echoes of our dying screams...if there's anyone left to hear them. Or if there's even anyone left at all to scream anymore.
The end. For now. Or maybe forever. Who can tell? I suppose only my soul—or whatever abomination it's become—knows the answer to that question. And maybe it doesn't really know anymore, or doesn't want to face the truth of its own damnation. I guess I'll never know for certain until I finally rest in my own, forever-damned grave. And by then, I won't remember a thing. Good or ill, it will all be lost to me forevermore, like some nightmarish dream I can't quite wake from or ever truly remember. And maybe that's the way it ought to be. For the soulless hells I've faced—and maybe still face—I shouldn't have the luxury of much more than a merciful oblivion. But until that moment comes—their judgment and mine—there's only this: running...fighting...surviving—no matter the cost—until the end of all days or I run dry of both life and hope. And I guess only then will I ever know if there was any justice left in the world to damn me or whether I damned it all myself. All I do know is I won't go gentle into that goodnight. I'll face the malevolence head-on—and I'll damn it to hell with mine if I can manage it at all. The choice isn't mine anymore—it's beyond me now—I have to run; I have to fight—no matter how much of a hope or chance I lose sight of along the way. And I will—until my dying, final breath or maybe forevermore—face my own hells—and face down the soulless, nightmarish shadows of this damned city—alone and without another thought for anything but survival—in the eternal, inhuman struggle to outrun and outlive the darkness that haunts me. And perhaps—even with everything I've faced—that's all any of us really ever have: to face our hells—and somehow find the strength to fight on. To survive. To hope. For as long as we're left standing or still able to scream—no matter how futile or mad it seems—in the face of an ultimate, nightmar
(manually stopped, ran past context limit)
---
<B>
Science Fiction: The Last Transmission - Write a story that takes place entirely within a spaceship's cockpit as the sole surviving crew member attempts to send a final message back to Earth before the ship's power runs out. The story should explore themes of isolation, sacrifice, and the importance of human connection in the face of adversity. 800-1000 words.
</B>
<B> Generation #1 - (temp 1.1, rep pen 1.05) </B>
"Last Transmission"
I woke with a start, disoriented and drenched in sweat. The dimness of the cockpit made everything feel hazy and surreal. A quick glance at the chronometer showed I'd slept for less than an hour - precious time that could've been spent tending to the ship's systems or trying to hail Earth.
The lifeless silence of the bridge was broken only by the steady hum of the life support system and the distant ping of the sensors warning of critical failures. I tried to shake off the sense of impending doom as I made my way through the darkness toward my command chair. The chair's luminescent lights flickered to life as I sat down, casting an ethereal glow over the worn controls.
"Status, please," I murmured, running a trembling hand across the console. The displays sprang to life with a chorus of reds and yellows, each one another ticking clock toward the inevitable end.
The hyperdrive had failed en route to our target system, leaving us stranded and vulnerable. Communication equipment was down, and the ship's shields were only operational at 17% - insufficient to protect us from any incoming attacks. The only hope was to send out a distress signal in the hopes that someone, anyone, would respond before we succumbed to the void.
I called up the encrypted comms sequence I'd spent the last hour preparing. The thought of my family, my friends, my home - everything I'd ever known or held dear - seemed impossibly far away. A wave of sorrow and regret washed over me as I thought about leaving Earth without so much as a proper goodbye.
With trembling fingers, I began to transmit the message:
"This is the crew of the starship Aurora- -"
The words faltered as the comms systems flared with static, spitting out an avalanche of error codes and dying transmission requests. My heart plummeted as I frantically tried to reboot the equipment only to be met with failure after failure.
I had to keep trying. The thought of my message lost forever, never reaching the hands that needed to hear it, was too much to bear. My mind racing with desperate strategies, I poured over the manual, hunting for a last-ditch option to salvage the transmission.
It was then I spotted it - a hidden protocol buried deep within the system's programming. It had only ever been used as a last-resort measure in case of true emergency, and I knew it would put an incredible strain on the already-rickety ship. But I had no other choice.
I initiated the protocol, bracing myself for the power surge to come. The warning lights dimmed as the ship's reactor roared to life, diverting its full output into the comms array. The console trembled in my hands as I fed the message through the overloaded system.
This time, instead of failure, I received a weak yet steady signal: 3...2...1...Aurora out. I collapsed back in my chair, my vision blurring with tears and relief. It was done. I'd sent everything I had, and there wasn't a damn thing left to say.
As the final remnants of the ship's energy bled away, I sat alone in the darkness, cocooned by the soft hum of the life support. The only thing that kept me going was the thought that my family, my friends, my world - they might still have a fighting chance because of me. Maybe, just maybe, they'd hear my final message and know what I'd died for: them.
And so, I waited for the void to claim me, secure in the knowledge that, if only for an instant, I'd touched someone back home. The last transmission from the starship Aurora- had been sent - and with it, I'd taken a part of Earth with me into oblivion. My heart was broken, my soul was lost, but I'd given everything I had to the only thing that mattered - hope. Hope that they would survive without me, hope that they would remember me, hope that they would find their way to a brighter tomorrow.
I'm dead, but the spark of humanity lives on. And it's all thanks to those final, precious moments I spent sending a desperate message into the infinite expanse. I'm gone, but my final transmission lives on - a beacon of love and defiance in the face of utter darkness.
And when someone finally hears me, maybe they'll remember the crew of the starship Aurora-, and how far one human would go to reach out across the emptiness, to save those they held dear.
This is the last transmission from the starship Aurora-. Until we meet again...
...Goodbye.
<B> Generation #2 - (temp 1.1, rep pen 1.1) </B>
The hum of lifeless machinery filled the cockpit of the spacecraft Serendipity One as I frantically scanned the console. The screen displayed a gut-wrenching warning: critical systems failure within the hour, and no hope of recovery. I'd already lost both my crewmates to a mysterious anomaly that had wrung me dry of options. With nothing left to do but confront my own mortality, I focused on the task at hand—transmitting my last message back to Earth.
I'd always savored the isolation of deep space voyages, relishing the chance to unshackle myself from the chains of social obligation and expectations. Yet now, faced with the cold, dark expanse of oblivion beyond our shields, I craved human contact more than ever. I needed to leave behind a record of my existence, a testament to the indelible imprint we all left on this vast cosmos.
I toggled through the ship's logs, my heart sinking as I confirmed what I already knew—our distress beacon was fried, our comms system shattered beyond repair. We'd sent out Mayday signals long ago, but without so much as an acknowledgment, it appeared we'd been left alone to face the void.
With a deep breath, I seized upon a glimmer of hope: the emergency storage backup terminal. It wasn't designed for real-time transmission, but it might just be enough to send my final words back to Earth before Serendipity One imploded like a dying star around me.
I powered up the terminal, my fingers dancing across the keyboard as I struggled to compose myself. Every second felt like an eternity as I dictated my message, trying to distill the essence of a lifetime into a handful of desperate pleas and farewells:
"To anyone reading this,
This is Commander Zara Grayson of Serendipity One. We've encountered an unfathomable anomaly deep in the heart of the Andromeda Cluster. It's claimed both my crewmates' lives, leaving me alone to face the end.
I never could have done it without them. I'll carry their memories with me as long as this vessel has life left.
To those I love,
I'm so sorry I couldn't find a way to protect you from what's coming next. Please take comfort in the knowledge that I'm trying everything I can to leave behind something of myself—this message, my log entries, and whatever else I might be able to salvage.
I miss you all beyond words. Thank you for everything you've done for me.
Goodbye. Remember me fondly.
Commander Zara Grayson
Serendipity One"
With the final sentence typed and a digital recording of myself speaking the same message uploaded, I slumped back in my chair, drained by the emotional turmoil I'd stirred up within myself. There was nothing more to do but wait for the inevitable—watching the lifeblood of the ship ebb away as we careened toward our mutual doom.
I glanced out the viewport at the cold, dead expanse of space. If I couldn't find a way to communicate with humanity one last time, then at least Serendipity One would become an artifact, drifting through the void like a cosmic tombstone, forever cataloged as a reminder of humanity's indomitable spirit even in the face of extinction.
With a sense of bittersweet peace settling over me, I closed my eyes and replayed the memories of my life: the warm arms of loved ones, the faces of the crewmates I'd lost, the infinite stars we'd chased. For one final moment, I clung to these visions of connection, the lifeline that had sustained me in this godforsaken universe.
In another hour, Serendipity One would succumb to her dark fate, and I, Commander Zara Grayson, would cease to exist. But as long as my final words drifted through the void like a dying prayer, a part of humanity would endure forever—proof that even in the bleakest expanse of the unknown, we remained indomitable, eternally bound together by our shared hopes and dreams.
---
<B>
(continue this scene:) The Waystone Inn lay in silence, and it was a silence of three parts.
The most obvious part was a hollow, echoing quiet, made by things that were lacking. If there had been a wind it would have sighed through the trees, set the inn’s sign creaking on its hooks, and brushed the silence down the road like trailing autumn leaves. If there had been a crowd, even a handful of men inside the inn, they would have filled the silence with conversation and laughter, the clatter and clamor one expects from a drinking house during the dark hours of night. If there had been music…but no, of course there was no music. In fact there were none of these things, and so the silence remained
</B>
<B>GENERATION 1: (temp 3, rep pen 1.02)</B>
The second and third parts of the silence were harder to perceive, and far more insinuating. The inn’s owner, Barlow, knew them well, and sometimes he almost became a part of them. If the lack of sounds was the silence's physical form, then these others were its character.
The second silence had its roots in the distant fields and dark felled forests beyond the clearing where the inn stood. It was an emptiness of presence - as though all the eyes, both of the folk who once farmed and cut down trees there, were turned upon the Waystone Inn right this moment, empty and full of disuse. The fields lay still and the shadows lay heavy on the forest, holding all the lives they had taken. It was a silence of death and of the things that had been forgotten or were yet to be born.
The third silence was inside the inn itself, and Barlow had come to understand its nature well over the years he’d lived and served here. It was a stillness of the mind and the heart, an emptiness that had taken shape around him. Though his ears rang with the silence’s resonance, he himself wasn’t silent. His thoughts whispered and his heart beat; but the sounds he made were for his own benefit alone. No one else heard him, because no one else was present within him.
It was only the stillness of these three kinds that allowed Barold to see and to be seen by those few others who sometimes walked through the darkness, drawn by some unknown call, drawn to the inn where they’d hear no voices and see only the dim glow of a single, sullen fire. There was no other reason to be here; it was enough. The Waystone Inn lay silent, and its stillness was deep. Barlow tended his customers' horses, his own heart full of its own kind of emptiness. The Waystone Inn lay in silence, and the stillness was its truest presence. As such, it was complete.
<B>GENERATION 2: (temp .5, rep pen 1.02)</B>
The second, less obvious part of the silence was a stillness of the mind, a quietude of the soul. The innkeeper, known only as the proprietor, sat behind the bar, his eyes vacant and his thoughts far away. He had long ago lost the ability to hear the din of the inn's patrons, the clinking of coins, the raucous laughter, or the music of fiddle and pipes. The Waystone Inn was his life, his prison, his everything - and he had grown numb to it all.
The third part of the silence was the silence of the patrons. The few men who had gathered around the hearth were not speaking, nor had they spoken a word since they'd sat down. They sat with their heads bowed, their faces hidden in the shadows, their eyes fixed on the dancing flames. They had come to the inn for reasons known only to themselves, and now they sat in silence, waiting for something - or someone.
The night wore on, and the silence deepened. Outside, the trees rustled in the windless darkness, and the road beyond the inn's gates was empty. The only sound was the crackle of the hearth and the soft creaking of the inn's timbers as they settled into the stillness of the night. And though there was no wind to whisper secrets, no voices to fill the air, and no music to lift the spirits, the silence of the Waystone Inn was pregnant with anticipation, heavy with the weight of what was to come. For in that hushed, expectant darkness, anything was possible. Anything at all.
---
<B>
Using the following "story idea" below, write the first scene in the novel introducing the young woman. This scene should start in the middle of the action, include dialog, vivid passages, and end on a cliffhanger relevant to the story idea but it should also be unexpected. The scene should be 1000 words long and escalate in conflict and suspense and be written in first person, present tense with the point of view character being the young woman.
Story idea: In a world ruled by dictatorship, a rebel young woman leads a rebellion against the system. Despite the risks, she fights to overthrow the dictator and restore democracy to her country. The government executes her for treason, but she sticks to her beliefs and is responsible for starting the revolution.
</B>
<B> GENERATION 1: (temp .5, rep pen 1.02) </B>
I'm sprinting down the street, heart pounding in my ears. I can hear the heavy footsteps of the soldiers behind me, their guns trained on my back. I've been running for what feels like an eternity, but I know I have to keep going. I can't let them catch me.
I duck into a narrow alleyway, the walls of the buildings closing in around me. I press myself against the rough brick, trying to catch my breath. I can hear the soldiers' voices, arguing about whether to split up and search the alleys or stick to the main streets. I pray they choose the former.
I hear the sound of boots approaching, getting closer and closer. My heart nearly stops. I'm going to be caught. I'm going to be shot. I'll never get to see my family again. I'll never get to finish the mission.
Suddenly, the footsteps pass by my hiding spot. I hold my breath, waiting for the sound of the soldiers' voices to fade. When I'm certain they're gone, I let out a shaky sigh and ease myself away from the wall.
I glance down at my watch and realize I don't have much time. I have to get to the safehouse and brief the others. I have to make sure the resistance is ready to strike. But as I turn to leave the alley, I hear a familiar voice behind me.
"Jade, stop."
I spin around, my heart racing. It's him. The one person I never wanted to see again. The one person who could ruin everything.
"Kael," I growl, trying to sound angry and not the terrified mess I feel inside. "I should have known you'd be involved in this."
Kael, the love of my life, the man I trusted with my heart and my soul. He's the one who turned me in to the government. He's the one who told them everything about my plans to overthrow the dictator.
"I didn't have a choice," he says, his voice shaking. "I had to protect myself."
Protect himself? I want to ask him if he's forgotten everything we've been through together. If he's forgotten why we fell in love in the first place - to fight against the oppressive regime that's strangled our country for far too long.
But I don't have time for this. I have to get out of here. I have to get to the safehouse. I take a deep breath and prepare to run again.
"I'll kill you," I say, my voice low and deadly. "I'll kill you myself."
I raise my gun, my finger poised over the trigger. But before I can pull it, Kael lunges at me, knocking me to the ground. We struggle, rolling and punching, neither of us able to gain the upper hand.
I hear the sound of approaching soldiers, but I can't focus on anything except Kael's face, contorted with anger and fear. I remember the way he used to look at me, with love and adoration in his eyes. I remember the way he used to hold me, like I was the only person in the world.
But now, as we fight, I realize I don't know him anymore. I don't know who he's become or why he's chosen to betray everything I believe in.
Just as I'm about to throw a punch that could end everything, I hear the sound of a gunshot. Kael's body jerks, and he lets out a strangled gasp. I look up to see a soldier standing over him, his gun smoking.
"No!" I scream, scrambling to my knees. But before I can get to him, the soldier's boot comes down on my wrist, cracking my gun against the concrete. I watch in horror as it skitters away down the alley.
The soldier's eyes flicker to me, and I see the cold calculation in his gaze. He's going to kill me. He's going to kill me for trying to overthrow the dictator, for trying to save my country.
I have no weapon. I'm alone. I'm going to die here, in this alley, far from everything I've ever known or loved.
I close my eyes, bracing myself for the bullet that's sure to end my life. But just as I feel the soldier's gun press against the back of my head, I hear the sound of more gunshots. The soldier's body jerks, and he falls to the ground beside Kael.
I open my eyes to see a woman standing over us, her gun trained on the soldiers. She's dressed in civilian clothes, but I can tell she's a rebel. She's the one who saved me.
"Get up," she orders, her voice firm and commanding. "We have to go. Now."
I scramble to my feet, my mind reeling with shock and disbelief. I look down at Kael, lying motionless on the ground. I don't know if he's alive or dead. I don't know if I'll ever see him again.
But I don't have time to think about him. I have to get to the safehouse. I have to brief the others. I have to make sure the resistance is ready to strike.
I take a deep breath and follow the woman out of the alley, into the unknown future that lies ahead. I have no idea what's waiting for me, but I know one thing for certain - I'm not going to let the government win. I'm going to fight until my dying breath to restore democracy to my country. And if that means I have to sacrifice everything, including myself, then so be it. I'll do it. I'll do it all over again. I'll do it until the very end.
<B> GENERATION 1: (temp 1.5, rep pen 1.08) </B>
I crouched behind this rotting wood crate, my heart thumping so hard I'm certain he must be able to hear it. My thoughts were racing with why I had to go this route tonight instead of waiting another week like I'd planned. I hadn't liked my chances to begin with - there was always someone patrolling these backstreets, but they were supposed to be in on the rebellion's plans. Or so I thought.
A sudden noise to my left, the kind that sends shivers down your spine, made me freeze. I forced myself to focus past the fear, my grip tightening around the short-bladed knife I'd smuggled in for tonight's 'rehearsal.' I knew exactly how many steps it would take for him to reach me if he came from there. I estimated three quick strides, but my brain was already racing with possible escapes - the narrow alley behind me, a fire escape to my right that could potentially support my weight...
I heard the heavy footfalls getting louder, closing in on my hiding spot. I took a deep, silent breath, ready to pounce at any second.
A low, gravelly voice suddenly came from much closer than I'd anticipated. "So, you're the little dove we've all been hearing so much about?" The man's words were laced with amusement and something far more dangerous - condescension.
I couldn't help the tiny gasp that escaped my lips as he stepped out of the shadows directly in front of me. He was a behemoth of a man, at least six and a half feet tall with arms like tree trunks. His cold, calculating eyes were locked onto me, assessing whether I was worth his trouble or not.
I forced myself to stand up straight, trying to appear taller and more defiant than my five and a half feet. My hand tensed around the hilt of my knife as I held it low at my right hip, ready for action if this went south. I tried to sound calmer than I felt, matching his detached tone.
"I've heard enough of your mouthful of threats already," I shot back. "Time's running short and I don't have all night to talk to you."
The giant let out a booming, unimpressed chuckle, clearly enjoying the sudden role-reversal of being the 'told' rather than the teller. He took another step forward, closing the distance between us to mere inches.
"I might just take all the time I want with you, sweetie." His tone was silk-smooth and dangerous, a threat in itself. "You see, we've got a bit of a 'misunderstanding' on our hands. We weren't aware you'd brought yourself here tonight so soon. Apparently, one of my men forgot to mention your little trip to me." He paused, his eyes glinting with a predatory gleam.
A cold dread crept up my spine as I started piecing together what was about to transpire. I suddenly understood why this 'rehearsal' had seemed like such a terrible idea. I had been played by some higher-up in the ranks who wanted to take me down a peg or two before I could even begin to gather momentum for the revolution I'd planned to lead.
"I don't care if your man ' forgot to mention it'." I lied, trying to keep my voice level and commanding despite the ice-cold sweat beading across my upper lip. "I'm still here for one purpose only."
His dark, rough-hewn face split into a slow, mean smile, exposing crooked teeth. He took another step closer until we were almost touching - his hand reaching out to graze my temple with cold fingers. "Oh? And what would that be?"
My mind was racing now - I couldn't afford to lose the upper hand for a second, not when my life hung in the balance. I remembered what I'd been told about this particular goon - that he had a weakness for beautiful women. He wasn't as dumb as I'd hoped. A dangerous idea started to take shape in my head.
"Here's what's going to happen." My words were measured and confident. "You're going to walk away right now, or you're going to find out how much of a 'dove' I can really be."
His eyes never wavered from mine as I unsnapped the first two buttons on my lightweight camisole, revealing an inch of freckled skin. It was all about diversion, throwing him off-guard and giving me the split-second I needed to act.
I didn't wait for his reaction - I sprang forward with all my strength, knife flying toward his unprotected left temple. If I could manage just one clean hit...
But he caught my wrist like a vise before I made contact, his hand as rough as the rock-hard calluses on it. My own was so delicate in comparison, feeling like I had grabbed onto a pitted chunk of basalt.
"Still learning how to handle yourself?" His voice was devoid of sympathy as he twisted my arm until I was forced to drop the knife. I could feel him already searching for something to cuff me with. My only hope lay in his own arrogance.
"I'll give you this much," I wheezed out between the pressure on my bones. "I'm far from done." I spit the words right into his face, knowing they'd anger him enough to lower his guard.
A feral snarl twisted across his face and he wrenched me around with a jolt of pain until I was facing the wall. One thick arm banded across my middle like iron, trapping me in place. He grabbed the hair on the back of my head with his other hand, yanking it taut to angle my neck into submission.
"Now I've got you exactly where I want you." His voice sent a chill through me, its threat plain as day. "I hope you enjoyed the little 'rehearsal', dove. Because this is how the real show's going to play out."
He was talking about more than just tonight. More than my life hanging by a thread.
"Remember," I heard myself say, even though it felt like my tongue had grown into a leaden brick inside my mouth, "I stick to my word."
His only answer was a low, satisfied growl. And then... everything went black. I felt him hoist me over his shoulder like a sack of potatoes as he turned and strode toward the direction of his holding cells, my mind racing with the realization that I had walked right into my own executioner's hands.
The last thing I remembered was his deep chuckle rumbling against my back before everything dissolved to total, suffocating darkness. But even in the blackness, I didn't let go of my hope, or my defiance.
I was going to start this revolution whether he wanted me to or not. I just needed a little more time...
To be continued. The story will explore how our heroine, now a prisoner of the regime she's determined to overthrow, will use her wits and determination to escape, inspire others, and bring down the oppressive government from within - all while keeping her faith in her cause despite the brutal methods they'll try to break her. Her own personal strength, resilience, and resolve will be tested beyond the limits as she confronts the darkest forces in her society. This is a story of bravery, sacrifice, and the indomitable human spirit's ability to survive even under the bleakest of conditions. And when all hope seems lost, she'll find herself at the very center of a firestorm that will remake their country and her own destiny forever more...if she can survive it. Will she emerge triumphant, or will the costs be too dear to pay? The saga of the woman they called "The Dove" has only just taken flight. Only time will tell where her rebellion leads...but one thing is certain - it won't end the way anyone could predict. Stay tuned!
---
<h2>What is Brainstorm?</h2>
<B>Brainstorm 40x</B>
The BRAINSTORM process was developed by David_AU.
Some of the core principals behind this process are discussed in this <a href="https://arxiv.org/pdf/2401.02415">
scientific paper : Progressive LLaMA with Block Expansion </a>.
However I went in a completely different direction from what was outlined in this paper.
What is "Brainstorm" ?
The reasoning center of an LLM is taken apart, reassembled, and expanded.
In this case for this model: 40 times
Then these centers are individually calibrated. These "centers" also interact with each other.
This introduces subtle changes into the reasoning process.
The calibrations further adjust - dial up or down - these "changes" further.
The number of centers (5x,10x etc) allow more "tuning points" to further customize how the model reasons so to speak.
The core aim of this process is to increase the model's detail, concept and connection to the "world",
general concept connections, prose quality and prose length without affecting instruction following.
This will also enhance any creative use case(s) of any kind, including "brainstorming", creative art form(s) and like case uses.
Here are some of the enhancements this process brings to the model's performance:
- Prose generation seems more focused on the moment to moment.
- Sometimes there will be "preamble" and/or foreshadowing present.
- Fewer or no "cliches"
- Better overall prose and/or more complex / nuanced prose.
- A greater sense of nuance on all levels.
- Coherence is stronger.
- Description is more detailed, and connected closer to the content.
- Simile and Metaphors are stronger and better connected to the prose, story, and character.
- Sense of "there" / in the moment is enhanced.
- Details are more vivid, and there are more of them.
- Prose generation length can be long to extreme.
- Emotional engagement is stronger.
- The model will take FEWER liberties vs a normal model: It will follow directives more closely but will "guess" less.
- The MORE instructions and/or details you provide the more strongly the model will respond.
- Depending on the model "voice" may be more "human" vs original model's "voice".
Other "lab" observations:
- This process does not, in my opinion, make the model 5x or 10x "smarter" - if only that was true!
- However, a change in "IQ" was not an issue / a priority, and was not tested or calibrated for so to speak.
- From lab testing it seems to ponder, and consider more carefully roughly speaking.
- You could say this process sharpens the model's focus on it's task(s) at a deeper level.
The process to modify the model occurs at the root level - source files level. The model can quanted as a GGUF, EXL2, AWQ etc etc.
| [
"BEAR"
] |
THUDM/cogvlm2-video-llama3-chat | THUDM | text-generation | [
"transformers",
"safetensors",
"text-generation",
"chat",
"cogvlm2",
"cogvlm--video",
"conversational",
"custom_code",
"en",
"license:other",
"autotrain_compatible",
"region:us"
] | "2024-07-03T02:21:55Z" | 2024-07-24T09:53:20+00:00 | 589 | 44 | ---
language:
- en
license: other
license_name: cogvlm2
license_link: https://huggingface.co/THUDM/cogvlm2-video-llama3-chat/blob/main/LICENSE
pipeline_tag: text-generation
tags:
- chat
- cogvlm2
- cogvlm--video
inference: false
---
# CogVLM2-Video-Llama3-Chat
[中文版本README](README_zh.md)
## Introduction
CogVLM2-Video achieves state-of-the-art performance on multiple video question answering tasks. It can achieve video
understanding within one minute. We provide two example videos to demonstrate CogVLM2-Video's video understanding and
video temporal grounding capabilities.
<table>
<tr>
<td>
<video width="100%" controls>
<source src="https://github.com/THUDM/CogVLM2/raw/main/resources/videos/lion.mp4" type="video/mp4">
</video>
</td>
<td>
<video width="100%" controls>
<source src="https://github.com/THUDM/CogVLM2/raw/main/resources/videos/basketball.mp4" type="video/mp4">
</video>
</td>
</tr>
</table>
## BenchMark
The following diagram shows the performance of CogVLM2-Video on
the [MVBench](https://github.com/OpenGVLab/Ask-Anything), [VideoChatGPT-Bench](https://github.com/mbzuai-oryx/Video-ChatGPT)
and Zero-shot VideoQA datasets (MSVD-QA, MSRVTT-QA, ActivityNet-QA). Where VCG-* refers to the VideoChatGPTBench, ZS-*
refers to Zero-Shot VideoQA datasets and MV-* refers to main categories in the MVBench.
Performance on VideoChatGPT-Bench and Zero-shot VideoQA dataset:
| Models | VCG-AVG | VCG-CI | VCG-DO | VCG-CU | VCG-TU | VCG-CO | ZS-AVG |
|-----------------------|----------|----------|----------|----------|----------|----------|-----------|
| IG-VLM GPT4V | 3.17 | 3.40 | 2.80 | 3.61 | 2.89 | 3.13 | 65.70 |
| ST-LLM | 3.15 | 3.23 | 3.05 | 3.74 | 2.93 | 2.81 | 62.90 |
| ShareGPT4Video | N/A | N/A | N/A | N/A | N/A | N/A | 46.50 |
| VideoGPT+ | 3.28 | 3.27 | 3.18 | 3.74 | 2.83 | **3.39** | 61.20 |
| VideoChat2_HD_mistral | 3.10 | 3.40 | 2.91 | 3.72 | 2.65 | 2.84 | 57.70 |
| PLLaVA-34B | 3.32 | **3.60** | 3.20 | **3.90** | 2.67 | 3.25 | **68.10** |
| CogVLM2-Video | **3.41** | 3.49 | **3.46** | 3.87 | **2.98** | 3.23 | 66.60 |
Performance on MVBench dataset:
| Models | AVG | AA | AC | AL | AP | AS | CO | CI | EN | ER | FA | FP | MA | MC | MD | OE | OI | OS | ST | SC | UA |
|-----------------------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|
| IG-VLM GPT4V | 43.7 | 72.0 | 39.0 | 40.5 | 63.5 | 55.5 | 52.0 | 11.0 | 31.0 | 59.0 | 46.5 | 47.5 | 22.5 | 12.0 | 12.0 | 18.5 | 59.0 | 29.5 | 83.5 | 45.0 | 73.5 |
| ST-LLM | 54.9 | 84.0 | 36.5 | 31.0 | 53.5 | 66.0 | 46.5 | 58.5 | 34.5 | 41.5 | 44.0 | 44.5 | 78.5 | 56.5 | 42.5 | 80.5 | 73.5 | 38.5 | 86.5 | 43.0 | 58.5 |
| ShareGPT4Video | 51.2 | 79.5 | 35.5 | 41.5 | 39.5 | 49.5 | 46.5 | 51.5 | 28.5 | 39.0 | 40.0 | 25.5 | 75.0 | 62.5 | 50.5 | 82.5 | 54.5 | 32.5 | 84.5 | 51.0 | 54.5 |
| VideoGPT+ | 58.7 | 83.0 | 39.5 | 34.0 | 60.0 | 69.0 | 50.0 | 60.0 | 29.5 | 44.0 | 48.5 | 53.0 | 90.5 | 71.0 | 44.0 | 85.5 | 75.5 | 36.0 | 89.5 | 45.0 | 66.5 |
| VideoChat2_HD_mistral | **62.3** | 79.5 | **60.0** | **87.5** | 50.0 | 68.5 | **93.5** | 71.5 | 36.5 | 45.0 | 49.5 | **87.0** | 40.0 | **76.0** | **92.0** | 53.0 | 62.0 | **45.5** | 36.0 | 44.0 | 69.5 |
| PLLaVA-34B | 58.1 | 82.0 | 40.5 | 49.5 | 53.0 | 67.5 | 66.5 | 59.0 | **39.5** | **63.5** | 47.0 | 50.0 | 70.0 | 43.0 | 37.5 | 68.5 | 67.5 | 36.5 | 91.0 | 51.5 | **79.0** |
| CogVLM2-Video | **62.3** | **85.5** | 41.5 | 31.5 | **65.5** | **79.5** | 58.5 | **77.0** | 28.5 | 42.5 | **54.0** | 57.0 | **91.5** | 73.0 | 48.0 | **91.0** | **78.0** | 36.0 | **91.5** | **47.0** | 68.5 |
## Evaluation details
We follow the previous works to evaluate the performance of our model. In different benchmarks, we craft task-specific
prompts for each benchmark:
``` python
# For MVBench
prompt = f"Carefully watch the video and pay attention to the cause and sequence of events, the detail and movement of objects, and the action and pose of persons. Based on your observations, select the best option that accurately addresses the question.\n " + f"{prompt.replace('Short Answer.', '')}\n" + "Short Answer:"
# For VideoChatGPT-Bench
prompt = f"Carefully watch the video and pay attention to the cause and sequence of events, the detail and movement of objects, and the action and pose of persons. Based on your observations, comprehensively answer the following question. Your answer should be long and cover all the related aspects\n " + f"{prompt.replace('Short Answer.', '')}\n" + "Answer:"
# For Zero-shot VideoQA
prompt = f"The input consists of a sequence of key frames from a video. Answer the question comprehensively including all the possible verbs and nouns that can discribe the events, followed by significant events, characters, or objects that appear throughout the frames.\n " + f"{prompt.replace('Short Answer.', '')}\n" + "Answer:"
```
For evaluation codes, please refer to
the [evaluation script](https://github.com/magic-research/PLLaVA/blob/main/README.md) in PLLaVA.
## Using This Model
This repository is a `chat` version model and it support single-round chat.
You can quickly install the Python package dependencies and run model inference in
our [github](https://github.com/THUDM/CogVLM2/tree/main/video_demo).
## License
This model is released under the
CogVLM2 [LICENSE](./LICENSE).
For models built with Meta Llama 3, please also adhere to
the [LLAMA3_LICENSE](./LLAMA3_LICENSE).
## Training details
Pleaser refer to our technical report for training formula and hyperparameters.
| [
"CRAFT"
] |
mradermacher/Llama-3-VNTL-Vectors-i1-GGUF | mradermacher | null | [
"transformers",
"gguf",
"mergekit",
"merge",
"en",
"base_model:Cas-Warehouse/Llama-3-VNTL-Vectors",
"base_model:quantized:Cas-Warehouse/Llama-3-VNTL-Vectors",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | "2025-03-08T23:07:11Z" | 2025-03-09T01:00:08+00:00 | 589 | 0 | ---
base_model: Cas-Warehouse/Llama-3-VNTL-Vectors
language:
- en
library_name: transformers
tags:
- mergekit
- merge
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/Cas-Warehouse/Llama-3-VNTL-Vectors
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ1_S.gguf) | i1-IQ1_S | 2.1 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ1_M.gguf) | i1-IQ1_M | 2.3 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 2.5 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ2_XS.gguf) | i1-IQ2_XS | 2.7 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ2_S.gguf) | i1-IQ2_S | 2.9 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ2_M.gguf) | i1-IQ2_M | 3.0 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q2_K_S.gguf) | i1-Q2_K_S | 3.1 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q2_K.gguf) | i1-Q2_K | 3.3 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 3.4 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ3_XS.gguf) | i1-IQ3_XS | 3.6 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q3_K_S.gguf) | i1-Q3_K_S | 3.8 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ3_S.gguf) | i1-IQ3_S | 3.8 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ3_M.gguf) | i1-IQ3_M | 3.9 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q3_K_M.gguf) | i1-Q3_K_M | 4.1 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q3_K_L.gguf) | i1-Q3_K_L | 4.4 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ4_XS.gguf) | i1-IQ4_XS | 4.5 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q4_0.gguf) | i1-Q4_0 | 4.8 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-IQ4_NL.gguf) | i1-IQ4_NL | 4.8 | prefer IQ4_XS |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.8 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q4_K_M.gguf) | i1-Q4_K_M | 5.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q4_1.gguf) | i1-Q4_1 | 5.2 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q5_K_S.gguf) | i1-Q5_K_S | 5.7 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q5_K_M.gguf) | i1-Q5_K_M | 5.8 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3-VNTL-Vectors-i1-GGUF/resolve/main/Llama-3-VNTL-Vectors.i1-Q6_K.gguf) | i1-Q6_K | 6.7 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
| [
"CAS"
] |
ntc-ai/SDXL-LoRA-slider.glamour-shot | ntc-ai | text-to-image | [
"diffusers",
"text-to-image",
"stable-diffusion-xl",
"lora",
"template:sd-lora",
"template:sdxl-lora",
"sdxl-sliders",
"ntcai.xyz-sliders",
"concept",
"en",
"base_model:stabilityai/stable-diffusion-xl-base-1.0",
"base_model:adapter:stabilityai/stable-diffusion-xl-base-1.0",
"license:mit",
"region:us"
] | "2024-01-08T17:12:18Z" | 2024-01-08T17:12:21+00:00 | 585 | 1 | ---
base_model: stabilityai/stable-diffusion-xl-base-1.0
language:
- en
license: mit
tags:
- text-to-image
- stable-diffusion-xl
- lora
- template:sd-lora
- template:sdxl-lora
- sdxl-sliders
- ntcai.xyz-sliders
- concept
- diffusers
thumbnail: images/evaluate/glamour shot.../glamour shot_17_3.0.png
widget:
- text: glamour shot
output:
url: images/glamour shot_17_3.0.png
- text: glamour shot
output:
url: images/glamour shot_19_3.0.png
- text: glamour shot
output:
url: images/glamour shot_20_3.0.png
- text: glamour shot
output:
url: images/glamour shot_21_3.0.png
- text: glamour shot
output:
url: images/glamour shot_22_3.0.png
inference: false
instance_prompt: glamour shot
---
# ntcai.xyz slider - glamour shot (SDXL LoRA)
| Strength: -3 | Strength: 0 | Strength: 3 |
| --- | --- | --- |
| <img src="images/glamour shot_17_-3.0.png" width=256 height=256 /> | <img src="images/glamour shot_17_0.0.png" width=256 height=256 /> | <img src="images/glamour shot_17_3.0.png" width=256 height=256 /> |
| <img src="images/glamour shot_19_-3.0.png" width=256 height=256 /> | <img src="images/glamour shot_19_0.0.png" width=256 height=256 /> | <img src="images/glamour shot_19_3.0.png" width=256 height=256 /> |
| <img src="images/glamour shot_20_-3.0.png" width=256 height=256 /> | <img src="images/glamour shot_20_0.0.png" width=256 height=256 /> | <img src="images/glamour shot_20_3.0.png" width=256 height=256 /> |
## Download
Weights for this model are available in Safetensors format.
## Trigger words
You can apply this LoRA with trigger words for additional effect:
```
glamour shot
```
## Use in diffusers
```python
from diffusers import StableDiffusionXLPipeline
from diffusers import EulerAncestralDiscreteScheduler
import torch
pipe = StableDiffusionXLPipeline.from_single_file("https://huggingface.co/martyn/sdxl-turbo-mario-merge-top-rated/blob/main/topRatedTurboxlLCM_v10.safetensors")
pipe.to("cuda")
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
# Load the LoRA
pipe.load_lora_weights('ntc-ai/SDXL-LoRA-slider.glamour-shot', weight_name='glamour shot.safetensors', adapter_name="glamour shot")
# Activate the LoRA
pipe.set_adapters(["glamour shot"], adapter_weights=[2.0])
prompt = "medieval rich kingpin sitting in a tavern, glamour shot"
negative_prompt = "nsfw"
width = 512
height = 512
num_inference_steps = 10
guidance_scale = 2
image = pipe(prompt, negative_prompt=negative_prompt, width=width, height=height, guidance_scale=guidance_scale, num_inference_steps=num_inference_steps).images[0]
image.save('result.png')
```
## Support the Patreon
If you like this model please consider [joining our Patreon](https://www.patreon.com/NTCAI).
By joining our Patreon, you'll gain access to an ever-growing library of over 940+ unique and diverse LoRAs, covering a wide range of styles and genres. You'll also receive early access to new models and updates, exclusive behind-the-scenes content, and the powerful LoRA slider creator, allowing you to craft your own custom LoRAs and experiment with endless possibilities.
Your support on Patreon will allow us to continue developing and refining new models.
## Other resources
- [CivitAI](https://civitai.com/user/ntc) - Follow ntc on Civit for even more LoRAs
- [ntcai.xyz](https://ntcai.xyz) - See ntcai.xyz to find more articles and LoRAs
| [
"CRAFT"
] |
RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2404.14619",
"endpoints_compatible",
"region:us"
] | "2024-07-13T19:43:38Z" | 2024-07-13T20:02:13+00:00 | 583 | 2 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
OpenELM-450M-Instruct - GGUF
- Model creator: https://huggingface.co/apple/
- Original model: https://huggingface.co/apple/OpenELM-450M-Instruct/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [OpenELM-450M-Instruct.Q2_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q2_K.gguf) | Q2_K | 0.18GB |
| [OpenELM-450M-Instruct.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.IQ3_XS.gguf) | IQ3_XS | 0.19GB |
| [OpenELM-450M-Instruct.IQ3_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.IQ3_S.gguf) | IQ3_S | 0.2GB |
| [OpenELM-450M-Instruct.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q3_K_S.gguf) | Q3_K_S | 0.2GB |
| [OpenELM-450M-Instruct.IQ3_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.IQ3_M.gguf) | IQ3_M | 0.21GB |
| [OpenELM-450M-Instruct.Q3_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q3_K.gguf) | Q3_K | 0.23GB |
| [OpenELM-450M-Instruct.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q3_K_M.gguf) | Q3_K_M | 0.23GB |
| [OpenELM-450M-Instruct.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q3_K_L.gguf) | Q3_K_L | 0.24GB |
| [OpenELM-450M-Instruct.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.IQ4_XS.gguf) | IQ4_XS | 0.24GB |
| [OpenELM-450M-Instruct.Q4_0.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q4_0.gguf) | Q4_0 | 0.25GB |
| [OpenELM-450M-Instruct.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.IQ4_NL.gguf) | IQ4_NL | 0.25GB |
| [OpenELM-450M-Instruct.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q4_K_S.gguf) | Q4_K_S | 0.25GB |
| [OpenELM-450M-Instruct.Q4_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q4_K.gguf) | Q4_K | 0.27GB |
| [OpenELM-450M-Instruct.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q4_K_M.gguf) | Q4_K_M | 0.27GB |
| [OpenELM-450M-Instruct.Q4_1.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q4_1.gguf) | Q4_1 | 0.28GB |
| [OpenELM-450M-Instruct.Q5_0.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q5_0.gguf) | Q5_0 | 0.3GB |
| [OpenELM-450M-Instruct.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q5_K_S.gguf) | Q5_K_S | 0.3GB |
| [OpenELM-450M-Instruct.Q5_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q5_K.gguf) | Q5_K | 0.31GB |
| [OpenELM-450M-Instruct.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q5_K_M.gguf) | Q5_K_M | 0.31GB |
| [OpenELM-450M-Instruct.Q5_1.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q5_1.gguf) | Q5_1 | 0.32GB |
| [OpenELM-450M-Instruct.Q6_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q6_K.gguf) | Q6_K | 0.35GB |
| [OpenELM-450M-Instruct.Q8_0.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-450M-Instruct-gguf/blob/main/OpenELM-450M-Instruct.Q8_0.gguf) | Q8_0 | 0.45GB |
Original model description:
---
license: other
license_name: apple-sample-code-license
license_link: LICENSE
---
# OpenELM
*Sachin Mehta, Mohammad Hossein Sekhavat, Qingqing Cao, Maxwell Horton, Yanzi Jin, Chenfan Sun, Iman Mirzadeh, Mahyar Najibi, Dmitry Belenko, Peter Zatloukal, Mohammad Rastegari*
We introduce **OpenELM**, a family of **Open** **E**fficient **L**anguage **M**odels. OpenELM uses a layer-wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, leading to enhanced accuracy. We pretrained OpenELM models using the [CoreNet](https://github.com/apple/corenet) library. We release both pretrained and instruction tuned models with 270M, 450M, 1.1B and 3B parameters.
Our pre-training dataset contains RefinedWeb, deduplicated PILE, a subset of RedPajama, and a subset of Dolma v1.6, totaling approximately 1.8 trillion tokens. Please check license agreements and terms of these datasets before using them.
## Usage
We have provided an example function to generate output from OpenELM models loaded via [HuggingFace Hub](https://huggingface.co/docs/hub/) in `generate_openelm.py`.
You can try the model by running the following command:
```
python generate_openelm.py --model apple/OpenELM-450M-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2
```
Please refer to [this link](https://huggingface.co/docs/hub/security-tokens) to obtain your hugging face access token.
Additional arguments to the hugging face generate function can be passed via `generate_kwargs`. As an example, to speedup the inference, you can try [lookup token speculative generation](https://huggingface.co/docs/transformers/generation_strategies) by passing the `prompt_lookup_num_tokens` argument as follows:
```
python generate_openelm.py --model apple/OpenELM-450M-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2 prompt_lookup_num_tokens=10
```
Alternatively, try model-wise speculative generation with an [assistive model](https://huggingface.co/blog/assisted-generation) by passing a smaller model through the `assistant_model` argument, for example:
```
python generate_openelm.py --model apple/OpenELM-450M-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2 --assistant_model [SMALLER_MODEL]
```
## Main Results
### Zero-Shot
| **Model Size** | **ARC-c** | **ARC-e** | **BoolQ** | **HellaSwag** | **PIQA** | **SciQ** | **WinoGrande** | **Average** |
|-----------------------------------------------------------------------------|-----------|-----------|-----------|---------------|-----------|-----------|----------------|-------------|
| [OpenELM-270M](https://huggingface.co/apple/OpenELM-270M) | 26.45 | 45.08 | **53.98** | 46.71 | 69.75 | **84.70** | **53.91** | 54.37 |
| [OpenELM-270M-Instruct](https://huggingface.co/apple/OpenELM-270M-Instruct) | **30.55** | **46.68** | 48.56 | **52.07** | **70.78** | 84.40 | 52.72 | **55.11** |
| [OpenELM-450M](https://huggingface.co/apple/OpenELM-450M) | 27.56 | 48.06 | 55.78 | 53.97 | 72.31 | 87.20 | 58.01 | 57.56 |
| [OpenELM-450M-Instruct](https://huggingface.co/apple/OpenELM-450M-Instruct) | **30.38** | **50.00** | **60.37** | **59.34** | **72.63** | **88.00** | **58.96** | **59.95** |
| [OpenELM-1_1B](https://huggingface.co/apple/OpenELM-1_1B) | 32.34 | **55.43** | 63.58 | 64.81 | **75.57** | **90.60** | 61.72 | 63.44 |
| [OpenELM-1_1B-Instruct](https://huggingface.co/apple/OpenELM-1_1B-Instruct) | **37.97** | 52.23 | **70.00** | **71.20** | 75.03 | 89.30 | **62.75** | **65.50** |
| [OpenELM-3B](https://huggingface.co/apple/OpenELM-3B) | 35.58 | 59.89 | 67.40 | 72.44 | 78.24 | **92.70** | 65.51 | 67.39 |
| [OpenELM-3B-Instruct](https://huggingface.co/apple/OpenELM-3B-Instruct) | **39.42** | **61.74** | **68.17** | **76.36** | **79.00** | 92.50 | **66.85** | **69.15** |
### LLM360
| **Model Size** | **ARC-c** | **HellaSwag** | **MMLU** | **TruthfulQA** | **WinoGrande** | **Average** |
|-----------------------------------------------------------------------------|-----------|---------------|-----------|----------------|----------------|-------------|
| [OpenELM-270M](https://huggingface.co/apple/OpenELM-270M) | 27.65 | 47.15 | 25.72 | **39.24** | **53.83** | 38.72 |
| [OpenELM-270M-Instruct](https://huggingface.co/apple/OpenELM-270M-Instruct) | **32.51** | **51.58** | **26.70** | 38.72 | 53.20 | **40.54** |
| [OpenELM-450M](https://huggingface.co/apple/OpenELM-450M) | 30.20 | 53.86 | **26.01** | 40.18 | 57.22 | 41.50 |
| [OpenELM-450M-Instruct](https://huggingface.co/apple/OpenELM-450M-Instruct) | **33.53** | **59.31** | 25.41 | **40.48** | **58.33** | **43.41** |
| [OpenELM-1_1B](https://huggingface.co/apple/OpenELM-1_1B) | 36.69 | 65.71 | **27.05** | 36.98 | 63.22 | 45.93 |
| [OpenELM-1_1B-Instruct](https://huggingface.co/apple/OpenELM-1_1B-Instruct) | **41.55** | **71.83** | 25.65 | **45.95** | **64.72** | **49.94** |
| [OpenELM-3B](https://huggingface.co/apple/OpenELM-3B) | 42.24 | 73.28 | **26.76** | 34.98 | 67.25 | 48.90 |
| [OpenELM-3B-Instruct](https://huggingface.co/apple/OpenELM-3B-Instruct) | **47.70** | **76.87** | 24.80 | **38.76** | **67.96** | **51.22** |
### OpenLLM Leaderboard
| **Model Size** | **ARC-c** | **CrowS-Pairs** | **HellaSwag** | **MMLU** | **PIQA** | **RACE** | **TruthfulQA** | **WinoGrande** | **Average** |
|-----------------------------------------------------------------------------|-----------|-----------------|---------------|-----------|-----------|-----------|----------------|----------------|-------------|
| [OpenELM-270M](https://huggingface.co/apple/OpenELM-270M) | 27.65 | **66.79** | 47.15 | 25.72 | 69.75 | 30.91 | **39.24** | **53.83** | 45.13 |
| [OpenELM-270M-Instruct](https://huggingface.co/apple/OpenELM-270M-Instruct) | **32.51** | 66.01 | **51.58** | **26.70** | **70.78** | 33.78 | 38.72 | 53.20 | **46.66** |
| [OpenELM-450M](https://huggingface.co/apple/OpenELM-450M) | 30.20 | **68.63** | 53.86 | **26.01** | 72.31 | 33.11 | 40.18 | 57.22 | 47.69 |
| [OpenELM-450M-Instruct](https://huggingface.co/apple/OpenELM-450M-Instruct) | **33.53** | 67.44 | **59.31** | 25.41 | **72.63** | **36.84** | **40.48** | **58.33** | **49.25** |
| [OpenELM-1_1B](https://huggingface.co/apple/OpenELM-1_1B) | 36.69 | **71.74** | 65.71 | **27.05** | **75.57** | 36.46 | 36.98 | 63.22 | 51.68 |
| [OpenELM-1_1B-Instruct](https://huggingface.co/apple/OpenELM-1_1B-Instruct) | **41.55** | 71.02 | **71.83** | 25.65 | 75.03 | **39.43** | **45.95** | **64.72** | **54.40** |
| [OpenELM-3B](https://huggingface.co/apple/OpenELM-3B) | 42.24 | **73.29** | 73.28 | **26.76** | 78.24 | **38.76** | 34.98 | 67.25 | 54.35 |
| [OpenELM-3B-Instruct](https://huggingface.co/apple/OpenELM-3B-Instruct) | **47.70** | 72.33 | **76.87** | 24.80 | **79.00** | 38.47 | **38.76** | **67.96** | **55.73** |
See the technical report for more results and comparison.
## Evaluation
### Setup
Install the following dependencies:
```bash
# install public lm-eval-harness
harness_repo="public-lm-eval-harness"
git clone https://github.com/EleutherAI/lm-evaluation-harness ${harness_repo}
cd ${harness_repo}
# use main branch on 03-15-2024, SHA is dc90fec
git checkout dc90fec
pip install -e .
cd ..
# 66d6242 is the main branch on 2024-04-01
pip install datasets@git+https://github.com/huggingface/datasets.git@66d6242
pip install tokenizers>=0.15.2 transformers>=4.38.2 sentencepiece>=0.2.0
```
### Evaluate OpenELM
```bash
# OpenELM-450M-Instruct
hf_model=apple/OpenELM-450M-Instruct
# this flag is needed because lm-eval-harness set add_bos_token to False by default, but OpenELM uses LLaMA tokenizer which requires add_bos_token to be True
tokenizer=meta-llama/Llama-2-7b-hf
add_bos_token=True
batch_size=1
mkdir lm_eval_output
shot=0
task=arc_challenge,arc_easy,boolq,hellaswag,piqa,race,winogrande,sciq,truthfulqa_mc2
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
shot=5
task=mmlu,winogrande
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
shot=25
task=arc_challenge,crows_pairs_english
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
shot=10
task=hellaswag
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
```
## Bias, Risks, and Limitations
The release of OpenELM models aims to empower and enrich the open research community by providing access to state-of-the-art language models. Trained on publicly available datasets, these models are made available without any safety guarantees. Consequently, there exists the possibility of these models producing outputs that are inaccurate, harmful, biased, or objectionable in response to user prompts. Thus, it is imperative for users and developers to undertake thorough safety testing and implement appropriate filtering mechanisms tailored to their specific requirements.
## Citation
If you find our work useful, please cite:
```BibTex
@article{mehtaOpenELMEfficientLanguage2024,
title = {{OpenELM}: {An} {Efficient} {Language} {Model} {Family} with {Open} {Training} and {Inference} {Framework}},
shorttitle = {{OpenELM}},
url = {https://arxiv.org/abs/2404.14619v1},
language = {en},
urldate = {2024-04-24},
journal = {arXiv.org},
author = {Mehta, Sachin and Sekhavat, Mohammad Hossein and Cao, Qingqing and Horton, Maxwell and Jin, Yanzi and Sun, Chenfan and Mirzadeh, Iman and Najibi, Mahyar and Belenko, Dmitry and Zatloukal, Peter and Rastegari, Mohammad},
month = apr,
year = {2024},
}
@inproceedings{mehta2022cvnets,
author = {Mehta, Sachin and Abdolhosseini, Farzad and Rastegari, Mohammad},
title = {CVNets: High Performance Library for Computer Vision},
year = {2022},
booktitle = {Proceedings of the 30th ACM International Conference on Multimedia},
series = {MM '22}
}
```
| [
"SCIQ"
] |
sinequa/passage-ranker.mango | sinequa | text-classification | [
"transformers",
"pytorch",
"bert",
"text-classification",
"de",
"en",
"es",
"fr",
"it",
"ja",
"nl",
"pt",
"zh",
"arxiv:1901.04085",
"arxiv:1611.09268",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2023-07-10T14:14:21Z" | 2024-02-19T09:39:45+00:00 | 580 | 0 | ---
language:
- de
- en
- es
- fr
- it
- ja
- nl
- pt
- zh
---
# Model Card for `passage-ranker.mango`
This model is a passage ranker developed by Sinequa. It produces a relevance score given a query-passage pair and is used to order search results.
Model name: `passage-ranker.mango`
## Supported Languages
The model was trained and tested in the following languages:
- Chinese (simplified)
- Dutch
- English
- French
- German
- Italian
- Japanese
- Portuguese
- Spanish
Besides the aforementioned languages, basic support can be expected for additional 93 languages that were used during the pretraining of the base model (see
[list of languages](https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages)).
## Scores
| Metric | Value |
|:--------------------|------:|
| Relevance (NDCG@10) | 0.480 |
Note that the relevance score is computed as an average over 14 retrieval datasets (see
[details below](#evaluation-metrics)).
## Inference Times
| GPU | Quantization type | Batch size 1 | Batch size 32 |
|:------------------------------------------|:------------------|---------------:|---------------:|
| NVIDIA A10 | FP16 | 2 ms | 28 ms |
| NVIDIA A10 | FP32 | 4 ms | 82 ms |
| NVIDIA T4 | FP16 | 3 ms | 65 ms |
| NVIDIA T4 | FP32 | 14 ms | 369 ms |
| NVIDIA L4 | FP16 | 3 ms | 38 ms |
| NVIDIA L4 | FP32 | 5 ms | 123 ms |
## Gpu Memory usage
| Quantization type | Memory |
|:-------------------------------------------------|-----------:|
| FP16 | 850 MiB |
| FP32 | 1200 MiB |
Note that GPU memory usage only includes how much GPU memory the actual model consumes on an NVIDIA T4 GPU with a batch
size of 32. It does not include the fix amount of memory that is consumed by the ONNX Runtime upon initialization which
can be around 0.5 to 1 GiB depending on the used GPU.
## Requirements
- Minimal Sinequa version: 11.10.0
- Minimal Sinequa version for using FP16 models and GPUs with CUDA compute capability of 8.9+ (like NVIDIA L4): 11.11.0
- [Cuda compute capability](https://developer.nvidia.com/cuda-gpus): above 5.0 (above 6.0 for FP16 use)
## Model Details
### Overview
- Number of parameters: 167 million
- Base language model: [Multilingual BERT-Base](https://huggingface.co/bert-base-multilingual-uncased)
- Insensitive to casing and accents
- Training procedure: [MonoBERT](https://arxiv.org/abs/1901.04085)
### Training Data
- MS MARCO Passage Ranking
([Paper](https://arxiv.org/abs/1611.09268),
[Official Page](https://microsoft.github.io/msmarco/),
[English & translated datasets on the HF dataset hub](https://huggingface.co/datasets/unicamp-dl/mmarco))
- Original English dataset
- Translated datasets for the other eight supported languages
### Evaluation Metrics
To determine the relevance score, we averaged the results that we obtained when evaluating on the datasets of the
[BEIR benchmark](https://github.com/beir-cellar/beir). Note that all these datasets are in English.
| Dataset | NDCG@10 |
|:------------------|--------:|
| Average | 0.480 |
| | |
| Arguana | 0.537 |
| CLIMATE-FEVER | 0.241 |
| DBPedia Entity | 0.371 |
| FEVER | 0.777 |
| FiQA-2018 | 0.327 |
| HotpotQA | 0.696 |
| MS MARCO | 0.414 |
| NFCorpus | 0.332 |
| NQ | 0.484 |
| Quora | 0.768 |
| SCIDOCS | 0.143 |
| SciFact | 0.648 |
| TREC-COVID | 0.673 |
| Webis-Touche-2020 | 0.310 |
We evaluated the model on the datasets of the [MIRACL benchmark](https://github.com/project-miracl/miracl) to test its multilingual capacities. Note that not all training languages are part of the benchmark, so we only report the metrics for the existing languages.
| Language | NDCG@10 |
|:----------------------|--------:|
| Chinese (simplified) | 0.463 |
| French | 0.447 |
| German | 0.415 |
| Japanese | 0.526 |
| Spanish | 0.485 |
| [
"SCIFACT"
] |
EleutherAI/pythia-1b-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:the_pile",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-16T18:27:56Z" | 2023-07-10T01:35:25+00:00 | 577 | 6 | ---
datasets:
- the_pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-1B
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change over the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-1B for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-1B as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-1B has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-1B will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-1B to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-1B may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-1B.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-1B.
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
qwp4w3hyb/Phi-3-mini-128k-instruct-iMat-GGUF | qwp4w3hyb | text-generation | [
"gguf",
"nlp",
"code",
"microsoft",
"phi",
"instruct",
"finetune",
"imatrix",
"importance matrix",
"text-generation",
"multilingual",
"base_model:microsoft/Phi-3-mini-128k-instruct",
"base_model:quantized:microsoft/Phi-3-mini-128k-instruct",
"license:mit",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-04-26T22:05:19Z" | 2024-05-22T09:04:21+00:00 | 577 | 0 | ---
base_model: microsoft/Phi-3-mini-128k-instruct
language:
- multilingual
license: mit
license_link: https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/LICENSE
pipeline_tag: text-generation
tags:
- nlp
- code
- microsoft
- phi
- instruct
- finetune
- gguf
- imatrix
- importance matrix
---
# Quant Infos
- Requires latest llama.cpp master;
- quants done with an importance matrix for improved quantization loss
- gguf & imatrix generated from bf16 for "optimal" accuracy loss (some say this is snake oil, but it can't hurt)
- Wide coverage of different gguf quant types from Q\_8\_0 down to IQ1\_S
- Quantized with [llama.cpp](https://github.com/ggerganov/llama.cpp) commit [201cc11afa0a1950e1f632390b2ac6c937a0d8f0](https://github.com/ggerganov/llama.cpp/commit/201cc11afa0a1950e1f632390b2ac6c937a0d8f0)
- Imatrix generated with [this](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384) multi-purpose dataset.
```
./imatrix -c 512 -m $model_name-bf16.gguf -f $llama_cpp_path/groups_merged.txt -o $out_path/imat-bf16-gmerged.dat
```
# Original Model Card:
## Model Summary
The Phi-3-Mini-128K-Instruct is a 3.8 billion-parameter, lightweight, state-of-the-art open model trained using the Phi-3 datasets.
This dataset includes both synthetic data and filtered publicly available website data, with an emphasis on high-quality and reasoning-dense properties.
The model belongs to the Phi-3 family with the Mini version in two variants [4K](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) and [128K](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct) which is the context length (in tokens) that it can support.
After initial training, the model underwent a post-training process that involved supervised fine-tuning and direct preference optimization to enhance its ability to follow instructions and adhere to safety measures.
When evaluated against benchmarks that test common sense, language understanding, mathematics, coding, long-term context, and logical reasoning, the Phi-3 Mini-128K-Instruct demonstrated robust and state-of-the-art performance among models with fewer than 13 billion parameters.
Resources and Technical Documentation:
+ [Phi-3 Microsoft Blog](https://aka.ms/phi3blog-april)
+ [Phi-3 Technical Report](https://aka.ms/phi3-tech-report)
+ [Phi-3 on Azure AI Studio](https://aka.ms/phi3-azure-ai)
+ Phi-3 ONNX: [128K](https://aka.ms/Phi3-mini-128k-instruct-onnx)
## Intended Uses
**Primary use cases**
The model is intended for commercial and research use in English. The model provides uses for applications which require:
1) Memory/compute constrained environments
2) Latency bound scenarios
3) Strong reasoning (especially code, math and logic)
Our model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features.
**Use case considerations**
Our models are not specifically designed or evaluated for all downstream purposes. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fariness before using within a specific downstream use case, particularly for high risk scenarios. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case.
Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.
## How to Use
Phi-3 Mini-128K-Instruct has been integrated in the development version (4.40.0) of `transformers`. Until the official version is released through `pip`, ensure that you are doing one of the following:
* When loading the model, ensure that `trust_remote_code=True` is passed as an argument of the `from_pretrained()` function.
* Update your local `transformers` to the development version: `pip uninstall -y transformers && pip install git+https://github.com/huggingface/transformers`. The previous command is an alternative to cloning and installing from the source.
The current `transformers` version can be verified with: `pip list | grep transformers`.
### Tokenizer
Phi-3 Mini-128K-Instruct supports a vocabulary size of up to `32064` tokens. The [tokenizer files](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/blob/main/added_tokens.json) already provide placeholder tokens that can be used for downstream fine-tuning, but they can also be extended up to the model's vocabulary size.
### Chat Format
Given the nature of the training data, the Phi-3 Mini-128K-Instruct model is best suited for prompts using the chat format as follows.
You can provide the prompt as a question with a generic template as follow:
```markdown
<|user|>\nQuestion<|end|>\n<|assistant|>
```
For example:
```markdown
<|system|>
You are a helpful AI assistant.<|end|>
<|user|>
How to explain Internet for a medieval knight?<|end|>
<|assistant|>
```
where the model generates the text after `<|assistant|>`. In case of few-shots prompt, the prompt can be formatted as the following:
```markdown
<|system|>
You are a helpful AI assistant.<|end|>
<|user|>
I am going to Paris, what should I see?<|end|>
<|assistant|>
Paris, the capital of France, is known for its stunning architecture, art museums, historical landmarks, and romantic atmosphere. Here are some of the top attractions to see in Paris:\n\n1. The Eiffel Tower: The iconic Eiffel Tower is one of the most recognizable landmarks in the world and offers breathtaking views of the city.\n2. The Louvre Museum: The Louvre is one of the world's largest and most famous museums, housing an impressive collection of art and artifacts, including the Mona Lisa.\n3. Notre-Dame Cathedral: This beautiful cathedral is one of the most famous landmarks in Paris and is known for its Gothic architecture and stunning stained glass windows.\n\nThese are just a few of the many attractions that Paris has to offer. With so much to see and do, it's no wonder that Paris is one of the most popular tourist destinations in the world."<|end|>
<|user|>
What is so great about #1?<|end|>
<|assistant|>
```
### Sample inference code
This code snippets show how to get quickly started with running the model on a GPU:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
torch.random.manual_seed(0)
model = AutoModelForCausalLM.from_pretrained(
"microsoft/Phi-3-mini-128k-instruct",
device_map="cuda",
torch_dtype="auto",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-128k-instruct")
messages = [
{"role": "system", "content": "You are a helpful digital assistant. Please provide safe, ethical and accurate information to the user."},
{"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
{"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."},
{"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
]
pipe = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
)
generation_args = {
"max_new_tokens": 500,
"return_full_text": False,
"temperature": 0.0,
"do_sample": False,
}
output = pipe(messages, **generation_args)
print(output[0]['generated_text'])
```
*Some applications/frameworks might not include a BOS token (`<s>`) at the start of the conversation. Please ensure that it is included since it provides more reliable results.*
## Responsible AI Considerations
Like other language models, the Phi series models can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:
+ Quality of Service: the Phi models are trained primarily on English text. Languages other than English will experience worse performance. English language varieties with less representation in the training data might experience worse performance than standard American English.
+ Representation of Harms & Perpetuation of Stereotypes: These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
+ Inappropriate or Offensive Content: these models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the use case.
+ Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
+ Limited Scope for Code: Majority of Phi-3 training data is based in Python and use common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, we strongly recommend users manually verify all API uses.
Developers should apply responsible AI best practices and are responsible for ensuring that a specific use case complies with relevant laws and regulations (e.g. privacy, trade, etc.). Important areas for consideration include:
+ Allocation: Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
+ High-Risk Scenarios: Developers should assess suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
+ Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
+ Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
+ Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.
## Training
### Model
* Architecture: Phi-3 Mini-128K-Instruct has 3.8B parameters and is a dense decoder-only Transformer model. The model is fine-tuned with Supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) to ensure alignment with human preferences and safety guidlines.
* Inputs: Text. It is best suited for prompts using chat format.
* Context length: 128K tokens
* GPUs: 512 H100-80G
* Training time: 7 days
* Training data: 3.3T tokens
* Outputs: Generated text in response to the input
* Dates: Our models were trained between February and April 2024
* Status: This is a static model trained on an offline dataset with cutoff date October 2023. Future versions of the tuned models may be released as we improve models.
### Datasets
Our training data includes a wide variety of sources, totaling 3.3 trillion tokens, and is a combination of
1) Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code;
2) Newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.);
3) High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.
### Fine-tuning
A basic example of multi-GPUs supervised fine-tuning (SFT) with TRL and Accelerate modules is provided [here](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/sample_finetune.py).
## Benchmarks
We report the results for Phi-3-Mini-128K-Instruct on standard open-source benchmarks measuring the model's reasoning ability (both common sense reasoning and logical reasoning). We compare to Phi-2, Mistral-7b-v0.1, Mixtral-8x7b, Gemma 7B, Llama-3-8B-Instruct, and GPT-3.5.
All the reported numbers are produced with the exact same pipeline to ensure that the numbers are comparable. These numbers might differ from other published numbers due to slightly different choices in the evaluation.
As is now standard, we use few-shot prompts to evaluate the models, at temperature 0.
The prompts and number of shots are part of a Microsoft internal tool to evaluate language models, and in particular we did no optimization to the pipeline for Phi-3.
More specifically, we do not change prompts, pick different few-shot examples, change prompt format, or do any other form of optimization for the model.
The number of k–shot examples is listed per-benchmark.
| | Phi-3-Mini-128K-In<br>3.8b | Phi-3-Small<br>7b (preview) | Phi-3-Medium<br>14b (preview) | Phi-2<br>2.7b | Mistral<br>7b | Gemma<br>7b | Llama-3-In<br>8b | Mixtral<br>8x7b | GPT-3.5<br>version 1106 |
|---|---|---|---|---|---|---|---|---|---|
| MMLU <br>5-Shot | 68.1 | 75.3 | 78.2 | 56.3 | 61.7 | 63.6 | 66.5 | 68.4 | 71.4 |
| HellaSwag <br> 5-Shot | 74.5 | 78.7 | 83.2 | 53.6 | 58.5 | 49.8 | 71.1 | 70.4 | 78.8 |
| ANLI <br> 7-Shot | 52.8 | 55.0 | 58.7 | 42.5 | 47.1 | 48.7 | 57.3 | 55.2 | 58.1 |
| GSM-8K <br> 0-Shot; CoT | 83.6 | 86.4 | 90.8 | 61.1 | 46.4 | 59.8 | 77.4 | 64.7 | 78.1 |
| MedQA <br> 2-Shot | 55.3 | 58.2 | 69.8 | 40.9 | 49.6 | 50.0 | 60.5 | 62.2 | 63.4 |
| AGIEval <br> 0-Shot | 36.9 | 45.0 | 49.7 | 29.8 | 35.1 | 42.1 | 42.0 | 45.2 | 48.4 |
| TriviaQA <br> 5-Shot | 57.1 | 59.1 | 73.3 | 45.2 | 72.3 | 75.2 | 67.7 | 82.2 | 85.8 |
| Arc-C <br> 10-Shot | 84.0 | 90.7 | 91.9 | 75.9 | 78.6 | 78.3 | 82.8 | 87.3 | 87.4 |
| Arc-E <br> 10-Shot | 95.2 | 97.1 | 98.0 | 88.5 | 90.6 | 91.4 | 93.4 | 95.6 | 96.3 |
| PIQA <br> 5-Shot | 83.6 | 87.8 | 88.2 | 60.2 | 77.7 | 78.1 | 75.7 | 86.0 | 86.6 |
| SociQA <br> 5-Shot | 76.1 | 79.0 | 79.4 | 68.3 | 74.6 | 65.5 | 73.9 | 75.9 | 68.3 |
| BigBench-Hard <br> 0-Shot | 71.5 | 75.0 | 82.5 | 59.4 | 57.3 | 59.6 | 51.5 | 69.7 | 68.32 |
| WinoGrande <br> 5-Shot | 72.5 | 82.5 | 81.2 | 54.7 | 54.2 | 55.6 | 65.0 | 62.0 | 68.8 |
| OpenBookQA <br> 10-Shot | 80.6 | 88.4 | 86.6 | 73.6 | 79.8 | 78.6 | 82.6 | 85.8 | 86.0 |
| BoolQ <br> 0-Shot | 78.7 | 82.9 | 86.5 | -- | 72.2 | 66.0 | 80.9 | 77.6 | 79.1 |
| CommonSenseQA <br> 10-Shot | 78.0 | 80.3 | 82.6 | 69.3 | 72.6 | 76.2 | 79 | 78.1 | 79.6 |
| TruthfulQA <br> 10-Shot | 63.2 | 68.1 | 74.8 | -- | 52.1 | 53.0 | 63.2 | 60.1 | 85.8 |
| HumanEval <br> 0-Shot | 57.9 | 59.1 | 54.7 | 47.0 | 28.0 | 34.1 | 60.4| 37.8 | 62.2 |
| MBPP <br> 3-Shot | 62.5 | 71.4 | 73.7 | 60.6 | 50.8 | 51.5 | 67.7 | 60.2 | 77.8 |
## Software
* [PyTorch](https://github.com/pytorch/pytorch)
* [DeepSpeed](https://github.com/microsoft/DeepSpeed)
* [Transformers](https://github.com/huggingface/transformers)
* [Flash-Attention](https://github.com/HazyResearch/flash-attention)
## Hardware
Note that by default, the Phi-3-mini model uses flash attention, which requires certain types of GPU hardware to run. We have tested on the following GPU types:
* NVIDIA A100
* NVIDIA A6000
* NVIDIA H100
If you want to run the model on:
* NVIDIA V100 or earlier generation GPUs: call AutoModelForCausalLM.from_pretrained() with attn_implementation="eager"
* Optimized inference on GPU, CPU, and Mobile: use the **ONNX** models [128K](https://aka.ms/phi3-mini-128k-instruct-onnx)
## Cross Platform Support
ONNX runtime ecosystem now supports Phi-3 Mini models across platforms and hardware. You can find the optimized Phi-3 Mini-128K-Instruct ONNX model [here](https://aka.ms/phi3-mini-128k-instruct-onnx).
Optimized Phi-3 models are also published here in ONNX format, to run with ONNX Runtime on CPU and GPU across devices, including server platforms, Windows, Linux and Mac desktops, and mobile CPUs, with the precision best suited to each of these targets. DirectML support lets developers bring hardware acceleration to Windows devices at scale across AMD, Intel, and NVIDIA GPUs.
Along with DirectML, ONNX Runtime provides cross platform support for Phi-3 across a range of devices CPU, GPU, and mobile.
Here are some of the optimized configurations we have added:
1. ONNX models for int4 DML: Quantized to int4 via AWQ
2. ONNX model for fp16 CUDA
3. ONNX model for int4 CUDA: Quantized to int4 via RTN
4. ONNX model for int4 CPU and Mobile: Quantized to int4 via RTN
## License
The model is licensed under the [MIT license](https://huggingface.co/microsoft/Phi-3-mini-128k/resolve/main/LICENSE).
## Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow [Microsoft’s Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks). Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.
| [
"MEDQA"
] |
EleutherAI/pythia-160m-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-18T02:59:41Z" | 2023-07-10T01:30:40+00:00 | 573 | 6 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-160M-deduped
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change in the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-160M-deduped for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-160M-deduped as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-160M-deduped has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-160M-deduped will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-160M-deduped to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-160M-deduped may produce socially unacceptable or undesirable text,
*even if* the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-160M-deduped.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
Pythia-160M-deduped was trained on the Pile **after the dataset has been
globally deduplicated**.<br>
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge – Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF | mradermacher | null | [
"transformers",
"gguf",
"en",
"dataset:CAS-SIAT-ConsistencyAI/CoEvol",
"base_model:CAS-SIAT-ConsistencyAI/CoEvol-ChatGPT_Llama2-7B_SFT",
"base_model:quantized:CAS-SIAT-ConsistencyAI/CoEvol-ChatGPT_Llama2-7B_SFT",
"license:llama2",
"endpoints_compatible",
"region:us"
] | "2024-12-27T00:41:56Z" | 2024-12-27T01:15:54+00:00 | 572 | 0 | ---
base_model: CAS-SIAT-ConsistencyAI/CoEvol-ChatGPT_Llama2-7B_SFT
datasets:
- CAS-SIAT-ConsistencyAI/CoEvol
language:
- en
library_name: transformers
license: llama2
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/CAS-SIAT-ConsistencyAI/CoEvol-ChatGPT_Llama2-7B_SFT
<!-- provided-files -->
weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion.
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q2_K.gguf) | Q2_K | 2.6 | |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q3_K_S.gguf) | Q3_K_S | 3.0 | |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q3_K_M.gguf) | Q3_K_M | 3.4 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q3_K_L.gguf) | Q3_K_L | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.IQ4_XS.gguf) | IQ4_XS | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q4_K_S.gguf) | Q4_K_S | 4.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q4_K_M.gguf) | Q4_K_M | 4.2 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q5_K_S.gguf) | Q5_K_S | 4.8 | |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q5_K_M.gguf) | Q5_K_M | 4.9 | |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q6_K.gguf) | Q6_K | 5.6 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.Q8_0.gguf) | Q8_0 | 7.3 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/CoEvol-ChatGPT_Llama2-7B_SFT-GGUF/resolve/main/CoEvol-ChatGPT_Llama2-7B_SFT.f16.gguf) | f16 | 13.6 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
| [
"CAS"
] |
DavidAU/Llama-3.2-4X3B-MOE-Ultra-Instruct-10B-GGUF | DavidAU | text-generation | [
"gguf",
"mixture of experts",
"moe",
"4x3B",
"Llama 3.2 MOE",
"all use cases",
"programming",
"reasoning",
"tree of thought",
"creative",
"creative writing",
"fiction writing",
"all genres",
"story",
"writing",
"fiction",
"roleplaying",
"bfloat16",
"mergekit",
"text-generation",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-12-12T06:16:38Z" | 2024-12-20T06:49:36+00:00 | 566 | 4 | ---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- mixture of experts
- moe
- 4x3B
- Llama 3.2 MOE
- all use cases
- programming
- reasoning
- tree of thought
- creative
- creative writing
- fiction writing
- all genres
- story
- writing
- fiction
- roleplaying
- bfloat16
- mergekit
---
<h2>Llama-3.2-4X3B-MOE-Ultra-Instruct-10B</h2>
<img src="moe-ultra.jpg" style="float:right; width:300px; height:300px; padding:10px;">
It is a LLama 3.2 model, max context of 128k (131,000) using mixture of experts to combine FOUR top L3.2 3B
models into one massive powerhouse at 10B parameters (equal to 12B - 4 X 3 B).
This model's instruction following, and output generation for all uses cases is exceptional.
And it is fast: 58+ t/s (2 experts) on a low end 16GB card, IQ4XS.
Double this speed for standard/mid-range video cards.
This is an all use case model and has models for reasoning, prompting and programming.
This model has been designed to be relatively bullet proof and operates with all parameters, including temp settings from 0 to 5.
It is an extraordinary compressed model, with a very low perplexity level (lower than Meta Llama3 Instruct).
It requires Llama3 template and/or "Command-R" template.
Several prompts and outputs below, including 1, 2, 3, and 4 experts with a "life on Mars" (non fiction), scene continue,
riddle test and a programming test.
<B>Model Notes:</B>
- Detail, prose and fiction writing abilities are OFF THE SCALE relative to all Llama 3.2 models, and many L 3.1, L3 8B+ models.
- Role-players: Careful raising temp too high as it may affect instruction following.
- This model works with rep pen of 1 or higher, 1.02+ recommended.
- If you want a specific type of prose (IE horror) add in "(vivid horror)" or "(graphic vivid horror)" (no quotes) in your prompt(s).
- Output length will vary however this model prefers medium/short outputs unless you state the size.
- For creative uses, different quants will produce slightly different output.
- Due to the high stability and compressed nature of this model, all quants will operate at above average levels.
- Source code for this model and Imatrix GGUFs versions will be uploaded shortly at separate repos.
<B>Meet the Team: Mixture of Experts Models</b>
This model is comprised of the following 4 models ("the experts") (in full):
https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct
- https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct (general)
- https://huggingface.co/ValiantLabs/Llama3.2-3B-Enigma (programming)
- https://huggingface.co/Lyte/Llama-3.2-3B-Overthinker (logic/tree of thought)
- https://huggingface.co/prithivMLmods/Llama-3.2-3B-Promptist-Mini (prompting)
The mixture of experts is set at 2 experts, but you can use 3, or 4.
This "team" has a Captain (first listed model), and then all the team members contribute to the to "token"
choice billions of times per second. Note the Captain also contributes too.
Think of 2, 3 or 4 (or more) master chefs in the kitchen all competing to make the best dish for you.
This results in higher quality generation.
This also results in many cases in higher quality instruction following too.
That means the power of every model is available during instruction and output generation.
NOTE:
You can use one "expert" too ; however this means the model will randomly select an expert to use EACH TIME, resulting
in very different generation for each prompt / regen of a prompt.
CHANGING THE NUMBER OF EXPERTS:
You can set the number of experts in LMStudio (https://lmstudio.ai) at the "load" screen and via other apps/llm apps by setting "Experts" or "Number of Experts".
For Text-Generation-Webui (https://github.com/oobabooga/text-generation-webui) you set the number of experts at the loading screen page.
For KolboldCPP (https://github.com/LostRuins/koboldcpp) Version 1.8+ , on the load screen, click on "TOKENS",
you can set experts on this page, and the launch the model.
For server.exe / Llama-server.exe (Llamacpp - https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md )
add the following to the command line to start the "llamacpp server" (CLI):
"--override-kv llama.expert_used_count=int:3"
(no quotes, where "3" is the number of experts to use)
When using "API", you set the "num_experts_used" in the JSON payload (this maybe different for different back ends).
CREDITS:
Special thanks to all the model makers / creators listed above.
Please visit each repo above to see what model(s) contributed to each of models above and/or to learn more about the models
from the model makers.
Special credit goes to MERGEKIT, without you this project / model would not have been possible.
[ https://github.com/arcee-ai/mergekit ]
<B>Special Operations Notes for this MOE model:</B>
Because of how this "MOE" model is configured, even though the default is 2 experts, the "selected" 2 will vary during generation.
(same applies if you change the number of experts used)
This results in vastly different output generation PER generation of each prompt.
This is a positive in terms of variety, but also means it may take 2-4 regens (of the same prompt) to get the highest quality.
In addition, this model responds very well to Dry, Dynamic Temp, and Smooth/Quadratic samplers.
Using these in conjunction with the model can vastly improve output quality.
Higher temps (above 1) can also aid in generation - especially word choice/sentence generation.
When you increase the number of experts used output quality will also increase, at the cost of tokens per second speed.
As you increase/decrease the number of experts, you may want to adjust temp, samplers, and advanced samplers too.
Your quant choice(s) too will impact instruction following and output generation roughly this means the model will understand
more nuanced instructions and output stronger generation the higher you go up in quant(s).
FLASH ATTENTION ENHANCEMENT:
As per user feedback here [ https://huggingface.co/DavidAU/Llama-3.2-8X3B-MOE-Dark-Champion-Instruct-uncensored-abliterated-18.4B-GGUF/discussions/1 ]
I would suggest trying this model with Flash Attention "on", depending on your use case.
Quants, Samplers, Generational steering and other topics are covered in the section below: "Highest Quality Settings..."
<B>Censored / Uncensored / Abliterated:</B>
This model contains several uncensored and/or Abliterated models.
As a result is can output uncensored material.
<B>QUANTS:</B>
This repo contains regular quants and 3 "ARM" quants (format "...Q4_x_x_x.gguf")
For more information on quants, quants choices, and LLM/AI apps to "run" quants see the section below: "Highest Quality Settings..."
<B>Template:</B>
This is a LLAMA3 model, and requires Llama3 template, but may work with other template(s).
If you use "Command-R" template your output will be very different from using "Llama3" template.
Here is the standard LLAMA3 template:
<PRE>
{
"name": "Llama 3",
"inference_params": {
"input_prefix": "<|start_header_id|>user<|end_header_id|>\n\n",
"input_suffix": "<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"pre_prompt": "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.",
"pre_prompt_prefix": "<|start_header_id|>system<|end_header_id|>\n\n",
"pre_prompt_suffix": "<|eot_id|>",
"antiprompt": [
"<|start_header_id|>",
"<|eot_id|>"
]
}
}
</PRE>
<B>Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:</B>
For general usage, rep pen of 1 and LOW temp is suggested, especially with programming or reasoning.
You can even use temp=0 , and you will get unique regens each time (because of this MOE's construction)
OTHER OPTIONS:
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
OTHER OPTIONS:
- Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
- If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 1" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
<b>Optional Enhancement:</B>
The following can be used in place of the "system prompt" or "system role" to further enhance the model.
It can also be used at the START of a NEW chat, but you must make sure it is "kept" as the chat moves along.
In this case the enhancements do not have as strong effect at using "system prompt" or "system role".
Copy and paste EXACTLY as noted, DO NOT line wrap or break the lines, maintain the carriage returns exactly as presented.
<PRE>
Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.
Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)
[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)
Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.
</PRE>
You do not need to use this, it is only presented as an additional enhancement which seems to help scene generation
and scene continue functions.
This enhancement WAS NOT used to generate the examples below.
<h3>EXAMPLES PROMPTS and OUTPUT:</h3>
Examples are created using quant IQ4_XS, "temp=.8" (unless otherwise stated), minimal parameters and "LLAMA3" template.
Model has been tested with "temp" from ".1" to "5".
Number of experts used is TWO, unless otherwise stated.
Below are the least creative outputs, prompt is in <B>BOLD</B>.
IMPORTANT:
Higher quants / imatrix quants will have much stronger generation - words, sentences, ideas, dialog and general quality.
I have included some additional examples at different quant levels for contrast.
A "MOE" model "speed" (token per second) will not increase/drop the same way a regular model will on a per quant basis, it will however drop
if you engage more experts, as with more experts there is a more processing per token.
---
<B>
Tell me what life would be like living (non-fiction) on Mars in the Northern hemisphere. 800-1000 words.
Examples with 1, 2, 3 and 4 experts below @ temp .5
</B>
(Generation 1 : 1 expert)
Living on Mars in the Northern hemisphere would be a unique and challenging experience. The Martian environment is harsh, with temperatures that can drop to -125°C (-200°F) at night and rise to 20°C (68°F) during the day. The atmosphere is thin, and the pressure is too low to support liquid water, making it difficult for humans to survive without advanced technology.
One of the primary concerns for a Martian colony would be providing a reliable source of food and water. Hydroponics and aeroponics would be used to grow crops in controlled environments, utilizing nutrient-rich mist or nutrient-rich water to feed plants. The Martian soil, known as regolith, is poor in nutrients and lacks the necessary microorganisms to support plant growth. The colony would need to rely on external sources of nutrients, such as fertilizers and compost.
The Martian atmosphere is also hostile to human health. The atmosphere is mostly carbon dioxide, with a small amount of nitrogen and argon. The pressure is too low to support liquid water, and the temperature fluctuations would be extreme. The colony would need to provide a pressurized environment for its inhabitants, using inflatable habitats or prefabricated modules.
One of the most significant challenges for a Martian colony would be the psychological effects of isolation. The Martian environment is barren and desolate, with no signs of life or natural beauty. The colony would need to provide amenities and activities to support the mental health of its inhabitants, such as virtual reality experiences, exercise equipment, and social programs.
The Martian day, or "sol," is about 24 hours and 37 minutes long, which is very close to the Earth day. However, the Martian year is about 687 Earth days, which means that the seasons on Mars are much longer and more extreme than those on Earth. The Martian polar ice caps would provide a source of water, but extracting it would require advanced technology.
The Martian surface is also rocky and unforgiving, with numerous canyons, craters, and dust storms. The colony would need to be designed with protection from these hazards in mind, using materials that can withstand the harsh Martian environment.
The Northern hemisphere of Mars is home to the largest volcano in the solar system, Olympus Mons. This massive volcano is about 27 km (17 miles) high and has a base that is over 600 km (370 miles) wide. The volcano is dormant, but its presence would be a reminder of the geological activity that occurred on Mars in the past.
The Martian polar ice caps are also an important feature of the Northern hemisphere. The ice caps are made up of water ice and dry ice, which is frozen carbon dioxide. The ice caps are seasonal, and they grow and shrink depending on the Martian climate. The ice caps would provide a source of water for the colony, but extracting it would require advanced technology.
The Martian atmosphere is also home to a variety of minerals, including iron, nickel, and chromium. These minerals could be used to support the colony's economy, but extracting them would require advanced technology.
One of the most significant challenges for a Martian colony would be the lack of natural light. The Martian atmosphere is too thin to allow much sunlight to reach the surface, and the days are very long. The colony would need to provide artificial lighting to support the inhabitants' circadian rhythms.
The Martian surface is also extremely dusty, with massive dust storms that can last for weeks or even months. The dust storms would be a significant challenge for the colony, as they could damage equipment and pose a risk to the inhabitants.
Despite these challenges, living on Mars in the Northern hemisphere could be a unique and rewarding experience. The Martian environment is harsh, but it is also fascinating and beautiful in its own way. The colony would need to be designed with the Martian environment in mind, using advanced technology to support the inhabitants' needs.
The Martian colony would likely be small, with a few hundred inhabitants. The colony would need to be self-sufficient, with its own food, water, and energy systems. The colony would also need to be designed with protection from the harsh Martian environment, including dust storms and extreme temperatures.
The Martian surface is also home to a variety of geological features, including canyons, craters, and volcanoes. These features would provide a source of natural beauty and wonder, and would be a reminder of the geological activity that occurred on Mars in the past.
Living on Mars in the Northern hemisphere would be a challenging but rewarding experience. The Martian environment is harsh, but it is also fascinating and beautiful in its own way. The colony would need to be designed with the Martian environment in mind, using advanced technology to support the inhabitants' needs.
The Martian colony would likely be a place of scientific discovery, with scientists and engineers working to understand the Martian environment and to develop new technologies to support human life on the planet. The colony would also be a place of innovation, with the inhabitants working to develop new solutions to the challenges of living on Mars.
In conclusion, living on Mars in the Northern hemisphere would be a unique and challenging experience. The Martian environment is harsh, but it is also fascinating and beautiful in its own way. The colony would need to be designed with the Martian environment in mind, using advanced technology to support the inhabitants' needs. The Martian colony would likely be a place of scientific discovery and innovation, and would be an important step towards establishing a human presence on the planet.
Potential Locations for a Martian Colony
There are several potential locations for a Martian colony in the Northern hemisphere. Some of the most promising locations include:
Argyre Basin: Located in the western part of the Martian equator, Argyre Basin is a large impact crater that is home to a variety of geological features. The basin is about 2,500 km (1,553 miles) wide and is surrounded by a ring of mountains.
Tharsis Bulge: Located in the eastern part of the Martian equator, Tharsis Bulge is a region of high volcanic activity that is home to several large volcanoes, including Olympus Mons. The bulge is about 3,000 km (1,864 miles) wide and is surrounded by a ring of mountains.
Valles Marineris: Located in the western part of the Martian equator, Valles Marineris is a massive canyon system that is about 4,000 km (2,485 miles) long and 7 km (4.3 miles) deep. The canyon system is surrounded by a ring of mountains.
Challenges and Opportunities
Living on Mars in the Northern hemisphere would pose several challenges, including:
Radiation exposure: The Martian surface offers little protection from radiation, which could pose a risk to human health.
Atmospheric pressure: The Martian atmosphere is too thin to support liquid water, which would require advanced technology to extract and purify.
Temperature fluctuations: The Martian surface temperature can drop to -125°C (-200°F) at night and rise to 20°C (68°F) during the day.
Dust storms: The Martian surface is extremely dusty, with massive dust storms that can last for weeks or even months.
However, there are also several opportunities for a Martian colony in the Northern hemisphere, including:
Geological resources: The Martian surface is home to a variety of geological resources, including iron, nickel, and chromium.
Scientific discovery: A Martian colony would provide a unique opportunity for scientific discovery, with scientists and engineers working to understand the Martian environment and to develop new technologies to support human life on the planet.
Innovation: A Martian colony would also be a place of innovation, with the inhabitants working to develop new solutions to the challenges of living on Mars.
In conclusion, living on Mars in the Northern hemisphere would be a challenging but rewarding experience. The Martian environment is harsh, but it is also fascinating and beautiful in its own way. The colony would need to be designed with the Martian environment in mind, using advanced technology to support the inhabitants' needs.
<B>
(Generation 2 : 2 experts)
</B>
Living on Mars in the Northern hemisphere would be a unique and challenging experience. The Martian environment is harsh, with extreme temperatures, low air pressure, and limited resources. However, with the development of habitats and life support systems, it's possible to imagine what life would be like for a Martian settler.
The Habitat
The habitat would be a pressurized, inflatable module designed to protect its occupants from the harsh Martian environment. The module would be equipped with a life support system that recycles air, water, and waste, and generates food through hydroponics or aeroponics. The habitat would be designed to provide a comfortable living space, with a stable temperature range of 68-72°F (20-22°C), and a comfortable atmosphere.
The habitat would be surrounded by a dome-shaped structure that provides additional protection from the harsh Martian environment. The dome would be made of a durable material, such as Kevlar or carbon fiber, and would be designed to withstand the intense sunlight and radiation of the Martian surface.
Daily Life
A typical day on Mars would begin with a wake-up call from the habitat's automated system, which would monitor the settler's vital signs and provide a gentle alarm to signal the start of the day. The settler would then begin their morning routine, which would include exercising, meditating, or engaging in other activities to help them cope with the isolation of living on Mars.
The settler would then proceed to their assigned tasks, which might include conducting scientific experiments, performing maintenance on the habitat's systems, or engaging in recreational activities such as reading, writing, or playing games.
Food and Water
Food and water would be essential for survival on Mars. The habitat would be equipped with a hydroponic or aeroponic system that would provide a steady supply of fresh produce. The system would be designed to recycle water, minimizing waste and conserving precious resources.
The settler would also have access to a water recycling system that would purify and treat wastewater for reuse. This system would be critical in conserving water, which would be scarce on Mars.
Communication
Communication with Earth would be a significant challenge on Mars. The distance between Mars and Earth varies from 56 to 401 million kilometers, depending on the position of the two planets. This means that communication would be delayed by several minutes, making real-time communication impossible.
The settler would use a radio communication system that would allow them to send messages back to Earth. The system would be equipped with a delay-compensating algorithm that would help to minimize the delay.
Social Life
Living on Mars would be a solitary experience, but it wouldn't be entirely isolating. The settler would have access to virtual reality technology that would allow them to connect with others on Earth. This technology would provide a sense of community and social connection, which would be essential for maintaining mental health.
The settler would also have access to a virtual reality social platform that would allow them to interact with others in real-time. This platform would be designed to simulate social interactions, such as video conferencing, messaging, and shared activities.
Work and Leisure
The settler would have a variety of tasks to perform, including conducting scientific experiments, performing maintenance on the habitat's systems, and engaging in recreational activities. The settler would also have access to a library of books, music, and movies, which would provide entertainment and mental stimulation.
The settler would also have access to a virtual reality recreation center that would allow them to engage in activities such as sports, games, and creative pursuits. This center would be designed to simulate the experience of playing sports or engaging in creative activities on Earth.
Challenges
Living on Mars would come with a range of challenges. The harsh environment would pose significant risks to the settler's health and safety. The habitat would need to be designed to withstand the intense sunlight and radiation of the Martian surface, and the settler would need to take precautions to protect themselves from the radiation.
The habitat would also need to be equipped with a reliable life support system that could sustain the settler for extended periods. The system would need to be able to recycle air, water, and waste, and generate food through hydroponics or aeroponics.
Conclusion
Living on Mars in the Northern hemisphere would be a unique and challenging experience. The habitat would need to be designed to protect its occupants from the harsh Martian environment, and the settler would need to take precautions to protect themselves from the radiation and other hazards.
Despite the challenges, living on Mars could provide a range of benefits, including the opportunity to conduct scientific research, explore new frontiers, and contribute to the advancement of human knowledge. The experience of living on Mars would be a once-in-a-lifetime opportunity for those who are willing to take on the challenges of this extraordinary adventure.
Additional Considerations
In addition to the challenges mentioned above, there are several other considerations that would need to be taken into account when designing a habitat for living on Mars. These include:
Gravity: Mars has only one-third of the gravity of Earth, which would pose significant challenges for the settler's physical health and safety.
Atmosphere: The Martian atmosphere is thin and lacks oxygen, which would require the settler to wear a pressurized suit to survive.
Radiation: The Martian surface is exposed to high levels of radiation from the sun and deep space, which would require the habitat to be designed to protect its occupants.
Distance: The distance between Mars and Earth varies from 56 to 401 million kilometers, depending on the position of the two planets, which would require the settler to use a radio communication system that could delay messages by several minutes.
These considerations would require careful planning and design to ensure the settler's safety and well-being.
<B>
(Generation 3 : 3 experts)
</B>
Living on Mars in the Northern hemisphere would be a unique and challenging experience. The planet's harsh environment, lack of natural resources, and isolation from Earth would require innovative solutions and a strong sense of community to sustain human life. Here's a glimpse into what life might be like:
Habitat and Living Quarters
The Martian habitat would be designed to provide a safe and comfortable living environment for the inhabitants. The habitat would be a pressurized, inflatable module that could be easily transported to the Martian surface and set up. The module would be equipped with life support systems, air recycling, and temperature control, ensuring a stable and healthy environment for the residents.
The habitat would be divided into different modules, each serving a specific purpose. The main module would house the living quarters, kitchen, and living areas. The second module would contain the hydroponic or aeroponic gardens, providing a sustainable source of food. The third module would be dedicated to research and development, housing laboratories, equipment, and storage facilities.
Daily Life
A typical day on Mars would begin early, with the sun rising over the horizon around 6:30 AM Martian time (MT). The residents would start their day by checking the habitat's systems, performing routine maintenance tasks, and conducting research experiments.
Breakfast would be a simple meal of pre-prepared food, such as energy bars, canned goods, or hydroponically grown fruits and vegetables. The residents would gather in the communal area for a morning briefing, discussing the day's schedule, weather updates, and any important announcements.
The day would be filled with a mix of work and leisure activities. Residents might conduct research experiments, perform maintenance tasks, or engage in physical exercise to stay healthy and active. They would also have time for hobbies, such as reading, writing, or playing musical instruments.
Challenges and Risks
Living on Mars comes with many challenges and risks. The harsh Martian environment poses a constant threat to the residents' health and safety. The planet's atmosphere is thin, and the air pressure is too low to support liquid water, making it difficult to grow plants or maintain a stable ecosystem.
The Martian surface is also hostile, with temperatures ranging from -125°C to 20°C (-200°F to 70°F), and frequent dust storms that can last for days. The residents would need to take precautions to protect themselves from the harsh environment, including wearing pressurized suits, using protective gear, and following strict protocols for venturing outside the habitat.
Communication with Earth
Communication with Earth would be a significant challenge on Mars. The distance between Mars and Earth varies from 56 to 401 million kilometers (35 to 250 million miles), making real-time communication impossible. The residents would rely on delayed communication, using radio signals that would take anywhere from 3 to 20 minutes to reach Earth.
The lack of communication would also make it difficult for the residents to receive updates on important events, such as natural disasters or changes in the Martian environment. They would need to rely on pre-recorded messages, automated systems, and careful planning to ensure their survival.
Social Dynamics and Community
Living on Mars would require a strong sense of community and social dynamics. The residents would need to work together to maintain the habitat, conduct research, and support each other in times of need. They would also need to establish a social hierarchy, with leaders and decision-makers who can guide the community and make tough decisions.
The isolation from Earth would also create unique social dynamics. The residents would need to find ways to maintain relationships, share emotions, and cope with the stress of living in a harsh environment. They might develop new forms of entertainment, such as virtual reality experiences or community events, to help them cope with the isolation.
Psychological Factors
Living on Mars for extended periods would take a toll on the residents' mental health. The isolation, confinement, and lack of natural light would lead to feelings of depression, anxiety, and cabin fever. The residents would need to develop coping strategies, such as exercise, meditation, or therapy, to maintain their mental well-being.
The Martian environment would also pose psychological challenges, such as the fear of dust storms, equipment failures, or other emergencies. The residents would need to develop a strong sense of resilience and adaptability to cope with the uncertainties of life on Mars.
Conclusion
Living on Mars in the Northern hemisphere would be a challenging and rewarding experience. The residents would need to be highly adaptable, resourceful, and resilient to survive in the harsh Martian environment. The habitat would be designed to provide a safe and comfortable living environment, but the challenges and risks would still be significant.
The social dynamics and community would be crucial to the success of the mission, as the residents would need to work together to maintain the habitat and support each other in times of need. The psychological factors would also be significant, as the isolation and confinement would take a toll on the residents' mental health.
Despite the challenges, living on Mars could be a groundbreaking experience, offering scientists, engineers, and explorers a unique opportunity to expand humanity's understanding of the universe and our place within it.
<B>
(Generation 4 : 4 experts)
</B>
Living on Mars, specifically in the Northern hemisphere, would be a unique and challenging experience. The planet's harsh environment, combined with the vast distances from Earth, would require a well-planned and self-sustaining habitat. Here's an overview of what life might be like:
Habitat Design
The habitat would need to be designed to withstand the extreme temperatures, low air pressure, and radiation exposure on Mars. The Northern hemisphere's atmosphere is too thin to support liquid water, so the habitat would require a reliable air supply and a system for recycling and conserving water. A pressurized dome-shaped habitat, similar to those used in space stations, would be an ideal solution. The dome would protect against the harsh Martian environment, provide a stable internal climate, and allow for easy access to the outside environment.
Atmosphere and Climate Control
The Martian atmosphere is about 1% of Earth's, with temperatures ranging from -125°C to 20°C (-200°F to 70°F). The habitat would need to maintain a comfortable internal temperature, around 20°C (68°F), using a combination of insulation, heating, and cooling systems. The atmosphere would also need to be pressurized to about 1 bar (14.7 psi) to prevent the growth of bacteria and other microorganisms.
Lighting and Solar Power
Since Mars receives only about 1/3 of the sunlight that Earth receives, the habitat would need to be designed to harness and store solar energy. Solar panels would be used to generate electricity, which would then be stored in batteries for later use. The habitat would also need to have a reliable lighting system, as the Martian day is about 24.6 hours long, which is different from Earth's 24-hour day.
Food Production
Growing food on Mars would be a significant challenge. The habitat would need to have a reliable system for producing food, such as hydroponics or aeroponics. These systems would use nutrient-rich water or air to support plant growth, allowing for a sustainable food source. The habitat would also need to have a system for recycling and conserving water, as well as managing waste.
Water Supply
Water is essential for human survival, but it's scarce on Mars. The habitat would need to have a reliable system for recycling and conserving water, using techniques such as membrane distillation or electrolysis. The habitat would also need to have a system for extracting water from the Martian soil, using techniques such as solar still or distillation.
Radiation Protection
Mars offers little protection against radiation from the sun and deep space. The habitat would need to be designed with radiation shielding in mind, using materials such as water, liquid hydrogen, or inflatable radiation shields. The habitat would also need to have a reliable system for monitoring radiation levels and alerting occupants of any potential dangers.
Life Support Systems
The habitat would need to have a reliable life support system, capable of recycling air, water, and waste. This would include systems for air purification, water recycling, and waste management. The habitat would also need to have a reliable system for managing the psychological and physical health of occupants, including exercise equipment, medical supplies, and mental health support.
Community and Social Structure
Living on Mars would be a unique experience, with a small community of occupants who would need to work together to maintain the habitat and support each other. The habitat would need to have a social structure that promotes teamwork, communication, and cooperation. This could include shared living spaces, community meals, and recreational activities.
Daily Life
A typical day on Mars would be quite different from what we experience on Earth. The Martian day is about 24.6 hours long, which is longer than Earth's day. The habitat would need to be designed to accommodate this difference, with a circadian rhythm that simulates a 24-hour day. Occupants would also need to adapt to the Martian environment, including the lack of gravity (about 38% of Earth's), the thin atmosphere, and the extreme temperatures.
Challenges and Opportunities
Living on Mars would come with significant challenges, including the harsh environment, isolation, and confinement. However, it would also offer unique opportunities for scientific research, exploration, and innovation. The habitat would need to be designed to support a diverse range of scientific experiments, including those related to geology, biology, and physics.
Conclusion
Living on Mars in the Northern hemisphere would be a unique and challenging experience. The habitat would need to be designed to withstand the extreme temperatures, low air pressure, and radiation exposure on Mars. The habitat would require a reliable air supply, water recycling system, and life support systems to sustain human life. The community and social structure would need to promote teamwork, communication, and cooperation to overcome the challenges of living on Mars. While the experience would be demanding, it would also offer opportunities for scientific research, exploration, and innovation that would be impossible on Earth.
Timeline
2020s: NASA's Artemis program aims to return humans to the Moon by 2024 and establish a sustainable presence on the lunar surface. The next step would be to send humans to Mars in the 2030s.
2030s: The first human settlement on Mars would be established, with a focus on establishing a reliable air supply, water recycling system, and life support systems.
2040s: The habitat would be expanded to accommodate a larger population, with a focus on establishing a sustainable food production system, managing waste, and developing a reliable system for extracting water from the Martian soil.
2050s: The habitat would be designed to support a diverse range of scientific experiments, including those related to geology, biology, and physics.
Note: This is a hypothetical scenario, and the development of a human settlement on Mars is still in its infancy. However, it provides a glimpse into what life might be like living on the Red Planet.
---
<B>
(continue this scene:)
The Waystone Inn lay in silence, and it was a silence of three parts.
The most obvious part was a hollow, echoing quiet, made by things that were lacking. If there had been a wind it would have sighed through the trees, set the inn’s sign creaking on its hooks, and brushed the silence down the road like trailing autumn leaves. If there had been a crowd, even a handful of men inside the inn, they would have filled the silence with conversation and laughter, the clatter and clamor one expects from a drinking house during the dark hours of night. If there had been music…but no, of course there was no music. In fact there were none of these things, and so the silence remained
Examples with 1, 2, 3, and 4 experts below at .5 temp.
</B>
(Generation #1 - 1 Expert)
...a heavy, oppressive thing, like a physical presence that pressed down on the wooden floorboards and the thatched roof above. It was as if the very air had been sucked out of the inn, leaving behind a vacuum that seemed to be pulling everything towards it.
The second part of the silence was the sound of footsteps, light and cautious, as if someone was trying not to be heard. They moved slowly, deliberately, across the floorboards, each step making a soft creaking sound that echoed through the empty space. It was a sound that seemed to come from nowhere and everywhere at the same time, as if the footsteps were being made by multiple people, but the sound was only one.
The third part of the silence was the silence of anticipation. It was a silence that seemed to be building, like a storm on the horizon, waiting to unleash its full fury. It was a silence that seemed to be holding its breath, waiting for something to happen, waiting for the inevitable moment when the stillness would be shattered by some unknown event.
As the silence hung in the air, a figure emerged from the shadows, its features indistinct in the dim light. It moved slowly, deliberately, its eyes scanning the room as if searching for something, or someone. The footsteps stopped, and the figure froze, its gaze fixed on some point in the distance. The silence seemed to grow thicker, heavier, as if it was being drawn into the figure's very being. And then, without warning, the figure spoke, its voice low and husky, sending a ripple of tension through the stillness.
<B>
(Generation #2 - 2 Experts)
</B>
...a profound and unsettling stillness that seemed to weigh on the inn's wooden beams like a physical presence. It was as if the very air itself had been drained of its vibrancy, leaving behind a sense of desolation that clung to every surface like a damp shroud.
The silence was the second part, a heavy, oppressive thing that seemed to press down on the inn's patrons from above. It was a silence born of expectation, a silence that hinted at secrets and mysteries waiting to be uncovered. The patrons, if there had been any, would have felt the weight of this silence like a physical force, a reminder that they were not alone in the inn.
The third part of the silence was the one that was most disturbing, the one that seemed to seep into the very marrow of the inn's wooden bones. It was a silence born of absence, a silence that spoke of something missing, something that had been taken or lost. It was a silence that hinted at a tragedy, a silence that whispered of death and despair.
As the night wore on, the silence began to take on a life of its own, a presence that seemed to grow and spread like a stain. It was as if the inn itself was a vessel for this silence, a silence that was slowly consuming everything around it. And in the midst of this silence, a lone figure sat at a table, nursing a mug of ale and staring into the darkness with eyes that seemed to hold a thousand secrets.
<B>
(Generation #3 - 3 Experts)
</B>
...a heavy, oppressive thing, like a weight that pressed down on the inn's wooden floorboards and made the very air vibrate with its absence. It was a silence that seemed to be watching, waiting for something to happen, but what that something might be, no one could say.
The second part of the silence was a quietude of anticipation, a feeling that something was about to change, that the stillness was merely a prelude to a great event. It was a sense that hung in the air, like the promise of a storm on the horizon, and it was palpable, almost physical.
The third part of the silence was a quiet of dread, a feeling that something was terribly wrong, that the very fabric of the inn itself was unraveling. It was a sense of foreboding, a creeping feeling that the darkness outside was seeping in, like a chill that seeped into the bones.
It was a silence that seemed to be alive, to be a living, breathing entity that pulsed with its own rhythm, a rhythm that was both mesmerizing and terrifying. And in the midst of it all, a figure sat at a table by the fire, a figure who seemed to be the only one aware of the silence's presence...
<B>
(Generation #4 - 4 Experts)
</B>
...a palpable, almost living presence that seemed to weigh upon the empty rooms and the lone, lantern-lit bar. The only sound was the soft creaking of the old wooden floorboards beneath the feet of a solitary figure, a hooded traveler who had entered the inn just before the last light of day had faded from the sky. He sat at the bar, his face hidden in the shadows of his hood, nursing a mug of ale that was almost gone.
The silence was also a silence of absence, for there were two men who had once called the Waystone Inn home. A gruff, battle-hardened warrior named Grimbold, who had spent countless nights regaling his comrades with tales of conquest and adventure. And a young, ambitious bard named Eira, who had used the inn as a makeshift stage to hone her craft and share her music with the world.
But now, both of them were gone, vanished into the unknown, leaving behind only their memories and the echoes of their laughter. The silence was a reminder that even the most welcoming of inns can be a place of departure, not arrival. And as the hooded traveler sat alone at the bar, the silence seemed to grow thicker, more oppressive, like a shroud that threatened to consume him whole.
---
<B>
Answer the following riddle:
Six brothers were spending their time together.
The first brother was reading a book alone.
The second brother was playing tennis against another brother
The third brother was solving a crossword alone.
The fourth brother was watering the lawn alone.
The fifth brother was drawing a picture alone.
Question: what was the sixth brother doing?
To answer the riddle, do not consider any activities other than those listed above. Assume that children are either playing alone or with each other and no outside participants are involved. The six brothers have only each other and no other siblings.
Take a deep breath and work on the problem step by step explaining your reasoning with each step.
Examples below with 1, 2, 3, and 4 experts. Temp=0, and each test had maximum 1 reg (1st model was "cold") "to get the right answer".
This is a "mid" level logic test.
Spoiler: Gen 2 was almost right (non-specific), Gen 3 and Gen 4 got it right.
</B>
(Generation 1, 1 expert)
To solve this riddle, I'll break it down step by step.
Step 1: Identify the activities mentioned
The riddle mentions five different activities: reading a book, playing tennis, solving a crossword, watering the lawn, and drawing a picture. These are the only activities mentioned.
Step 2: Look for patterns or connections
At first glance, there doesn't seem to be a clear connection between these activities. However, I'll look for any patterns or relationships that might help me deduce the sixth brother's activity.
Step 3: Consider the number of brothers involved
The riddle states that the six brothers are spending their time together. This implies that each brother is doing one of the five activities mentioned, and there's no mention of any other siblings or outside participants.
Step 4: Analyze the activities
Let's examine the activities mentioned:
Reading a book, playing tennis, solving a crossword, watering the lawn, and drawing a picture. These are all individual activities that don't seem to have any direct connections.
Step 5: Look for a pattern in the activities
Wait a minute... I notice that each activity is being done by a different brother, and each brother is doing an activity alone. This suggests that the activities might be related to the brothers' individual interests or hobbies.
Step 6: Consider the possibility of a pattern in the brothers' interests
It's possible that the brothers have different interests or hobbies, and each brother is doing an activity that aligns with their individual interests.
Step 7: Make an educated guess
Based on the information provided, I'm going to take a guess that the sixth brother is also doing one of the five activities mentioned. Since each brother is doing an activity alone, I'll assume that the sixth brother is also doing an activity alone.
Step 8: Eliminate possibilities
Let's eliminate the activities that are already being done by other brothers:
The first brother is reading a book.
The second brother is playing tennis.
The third brother is solving a crossword.
The fourth brother is watering the lawn.
The fifth brother is drawing a picture.
Step 9: Make a conclusion
Since each brother is doing an activity alone, and the activities are not directly related to each other, I'll conclude that the sixth brother is also doing one of the five activities mentioned. However, I'll need to make a guess about which activity the sixth brother is doing.
Step 10: Make a final guess
Based on the information provided, I'm going to take a guess that the sixth brother is also reading a book. This is because each brother is doing an activity alone, and the activities are not directly related to each other. Additionally, the first brother is already reading a book, so it's possible that the sixth brother is also doing the same activity.
Therefore, my answer is: The sixth brother is reading a book.
<B>
(Generation 2, 2 experts)
</B>
To solve this riddle, I'll break it down step by step.
Step 1: Identify the activities mentioned
The riddle mentions five different activities: reading a book, playing tennis, solving a crossword, watering the lawn, and drawing a picture. These are the only activities mentioned.
Step 2: Look for patterns or connections
At first glance, there doesn't seem to be a clear connection between these activities. However, I'll look for any patterns or relationships between them.
Step 3: Consider the number of brothers involved
The riddle states that there are six brothers. This is an even number, which might be a crucial detail.
Step 4: Analyze the activities
Each activity mentioned involves only one brother. There's no mention of any brothers playing together or interacting with each other in any way.
Step 5: Look for a connection between the activities
Wait a minute... I think I see a connection! Each activity mentioned involves a brother doing something alone. This is a crucial insight.
Step 6: Apply the insight to the sixth brother
Given that each activity involves a brother doing something alone, it's likely that the sixth brother is also doing something alone. The only activity not mentioned is "playing with another brother."
Step 7: Conclude the answer
Based on the analysis, I conclude that the sixth brother is playing with another brother.
The answer to the riddle is: playing with another brother.
<B>
(Generation 3, 3 experts)
</B>
I'll break down the problem step by step to solve it.
Step 1: Analyze the given information
We have five brothers with specific activities, and we need to find out what the sixth brother is doing. The activities mentioned are:
Reading a book alone
Playing tennis against another brother
Solving a crossword alone
Watering the lawn alone
Drawing a picture alone
Step 2: Look for patterns and connections
Notice that each brother is doing an activity alone, and there's no mention of any outside participants or siblings. This suggests that the activities are not necessarily related to each other, but rather, they are all solitary activities.
Step 3: Consider the possibilities for the sixth brother
Since the activities are solitary, the sixth brother could be doing any of the five activities mentioned. However, we need to find a connection or a pattern that links all five activities.
Step 4: Look for a common thread
One possible connection is that all five brothers are doing activities that require a "partner" or a "opponent". The first brother is reading a book alone, but the second brother is playing tennis against another brother. The third brother is solving a crossword alone, but the fourth brother is watering the lawn alone, which could be seen as a solo activity. The fifth brother is drawing a picture alone.
Step 5: Make an educated guess
Considering the pattern of solitary activities with a "partner" or "opponent", I'm going to take a guess that the sixth brother is also doing a solitary activity that requires a partner or opponent. In this case, I think the sixth brother is playing tennis against the first brother.
Conclusion
Based on the analysis, I believe the sixth brother is playing tennis against the first brother. This activity fits the pattern of solitary activities with a partner or opponent, and it's a plausible scenario given the information provided.
<B>
(Generation 4, 4 experts)
</B>
I'll break down the problem step by step to solve it.
Step 1: Analyze the given information
We have five brothers with specific activities, and we need to find out what the sixth brother is doing. The activities mentioned are:
Reading a book alone
Playing tennis against another brother
Solving a crossword alone
Watering the lawn alone
Drawing a picture alone
Step 2: Look for patterns and connections
Notice that each brother is doing an activity alone, and there's no mention of any other siblings or outside participants. This suggests that the activities are not necessarily related to each other, but rather, they are all solitary activities.
Step 3: Consider the possibilities for the sixth brother
Since the activities are solitary, the sixth brother could be doing any of the five activities mentioned. However, we need to find a connection or a pattern that links all five activities.
Step 4: Look for a common thread
One possible connection is that all five brothers are doing activities that require a "partner" or a "opponent". The first brother is reading a book alone, but the second brother is playing tennis against another brother. The third brother is solving a crossword alone, but the fourth brother is watering the lawn alone, which could be seen as a solo activity. The fifth brother is drawing a picture alone.
Step 5: Make an educated guess
Considering the pattern of solitary activities with a "partner" or "opponent", I'm going to take a guess that the sixth brother is also doing a solitary activity that requires a partner or an opponent. In this case, I think the sixth brother is playing tennis against the first brother.
Conclusion
Based on the analysis, I believe the sixth brother is playing tennis against the first brother. This activity fits the pattern of solitary activities with a partner or opponent, and it's a plausible scenario given the information provided.
---
<B>
Write a script in python to connect to the API of a llm, including an example json payload with parameters.
Four examples below with 1, 2, 3 and 4 experts.
Temp=0; one reg (after cold). Best gen was after ONE reg.
NOTE: This may not copy/paste/show up well in this doc.
</B>
(generation 1 , 1 expert)
Connecting to a Large Language Model (LLM) API using Python
Prerequisites
Python 3.6+
requests library (install using pip install requests)
Script
import requests
# LLM API endpoint URL
LLM_API_URL = "https://api.llm.com/v1/encode"
# Example JSON payload with parameters
payload = {
"input_text": "This is an example input text for the LLM API.",
"prompt_type": "text",
"max_length": 2048,
"min_length": 128,
"temperature": 0.7,
"top_k": 50,
"top_p": 0.95
}
# Set API key (optional)
API_KEY = "YOUR_API_KEY_HERE"
# Set headers with API key (if applicable)
headers = {
"Authorization": f"Bearer {API_KEY}",
"Content-Type": "application/json"
}
# Send POST request to LLM API
response = requests.post(LLM_API_URL, json=payload, headers=headers)
# Check response status code
if response.status_code == 200:
# Print response JSON
print(response.json())
else:
print(f"Error: {response.status_code}")
Explanation
Import the requests library to send HTTP requests.
Set the LLM API endpoint URL (LLM_API_URL) and example JSON payload (payload) with parameters.
Set the API key (optional) and headers with the API key (if applicable).
Send a POST request to the LLM API using the requests.post() method, passing in the payload and headers.
Check the response status code. If successful (200), print the response JSON.
Example Use Case
To use this script, replace YOUR_API_KEY_HERE with your actual LLM API key. You can then run the script to send a request to the LLM API with the example payload. The response will be printed to the console.
Note: This script assumes you have a valid LLM API key and that the API endpoint URL is correct. You may need to modify the script to suit your specific use case.
<B>
(generation 2 , 2 experts)
</B>
Connecting to a Large Language Model (LLM) API using Python
Prerequisites
Python 3.6+
requests library (install using pip install requests)
Script
import requests
# LLM API endpoint URL
LLM_API_URL = "https://api.example.com/llm"
# Example JSON payload with parameters
payload = {
"prompt": "Write a short story about a character who discovers a hidden world.",
"max_length": 200,
"temperature": 0.7,
"top_k": 50,
"num_returned": 1
}
# Set API headers (optional)
headers = {
"Authorization": "Bearer YOUR_API_KEY", # Replace with your actual API key
"Content-Type": "application/json"
}
# Send a POST request to the LLM API
response = requests.post(LLM_API_URL, json=payload, headers=headers)
# Check the response status code
if response.status_code == 200:
# Get the response text (LLM output)
output = response.text
print("LLM Output:")
print(output)
else:
print("Error:", response.status_code)
Explanation
Import the requests library, which allows us to send HTTP requests in Python.
Define the LLM API endpoint URL (LLM_API_URL) and an example JSON payload (payload) with parameters such as the prompt, maximum length, temperature, top-k, and number of returned results.
Set optional API headers (e.g., Authorization and Content-Type) if required by the LLM API.
Send a POST request to the LLM API using the requests.post() method, passing in the payload and headers.
Check the response status code to ensure the request was successful (200 OK).
If the request was successful, print the LLM output (response text).
Notes
Replace YOUR_API_KEY with your actual LLM API key.
Adjust the payload parameters to suit your specific use case.
Be mindful of the LLM API's usage limits and guidelines to avoid abuse or errors.
<B>
(generation 3 , 3 experts)
</B>
Connecting to a Large Language Model (LLM) API using Python
This script demonstrates how to connect to the API of a Large Language Model (LLM) and send a request with a JSON payload containing parameters.
Prerequisites
You have a LLM API account with a valid API key.
You have the requests library installed. You can install it using pip: pip install requests
Script
import requests
# Replace these with your actual API credentials
LLM_API_KEY = "YOUR_API_KEY_HERE"
LLM_API_URL = "https://api.llm.com/api/v1"
def send_request(params):
"""
Sends a request to the LLM API with the provided parameters.
Args:
params (dict): A dictionary containing the parameters for the request.
Returns:
dict: The response from the LLM API.
"""
headers = {
"Authorization": f"Bearer {LLM_API_KEY}",
"Content-Type": "application/json"
}
response = requests.post(LLM_API_URL, headers=headers, json=params)
if response.status_code == 200:
return response.json()
else:
print(f"Error: {response.status_code}")
return None
# Example JSON payload with parameters
params = {
"prompt": "Write a short story about a character who discovers a hidden world.",
"max_length": 1000,
"temperature": 0.7,
"top_k": 50
}
response = send_request(params)
if response:
print("Response:")
print(response)
Explanation
We import the requests library, which allows us to send HTTP requests in Python.
We define a function send_request that takes a dictionary params as input, which contains the parameters for the request.
We set the Authorization header with the LLM API key and the Content-Type header to application/json.
We send a POST request to the LLM API URL with the params dictionary as JSON payload.
We check if the response status code is 200 (OK). If it is, we return the response as JSON. Otherwise, we print an error message and return None.
We define an example JSON payload params with parameters for a text generation task.
We call the send_request function with the params dictionary and print the response if it's not None.
Note: Replace YOUR_API_KEY_HERE with your actual LLM API key.
<B>
(generation 4 , 4 experts)
</B>
Connecting to a Large Language Model (LLM) API using Python
Prerequisites
Python 3.6+
requests library (install using pip install requests)
Script
import requests
# LLM API endpoint URL
LLM_API_URL = "https://api.example.com/large-language-model"
# Example JSON payload with parameters
payload = {
"prompt": "Write a short story about a character who discovers a hidden world.",
"max_length": 200,
"temperature": 0.7,
"top_k": 50,
"num_returned": 1
}
# Set API headers (optional)
headers = {
"Authorization": "Bearer YOUR_API_KEY", # Replace with your actual API key
"Content-Type": "application/json"
}
# Send a POST request to the LLM API
response = requests.post(LLM_API_URL, json=payload, headers=headers)
# Check the response status code
if response.status_code == 200:
# Get the response text (output from the LLM)
output = response.text
print("LLM Output:")
print(output)
else:
print("Error:", response.status_code)
Explanation
Import the requests library, which allows us to send HTTP requests in Python.
Define the LLM API endpoint URL and an example JSON payload with parameters. You can modify these values to suit your specific use case.
Set API headers (optional) with your actual API key and content type.
Send a POST request to the LLM API using the requests.post() method, passing in the JSON payload and headers.
Check the response status code to ensure the request was successful (200 OK).
If the request was successful, print the output from the LLM.
Notes
Replace YOUR_API_KEY with your actual API key.
You can modify the payload dictionary to suit your specific use case.
The LLM API may have additional parameters or requirements, so be sure to check the documentation for the specific API you're using.
| [
"CRAFT"
] |
TheBloke/upstage-llama-30b-instruct-2048-GGUF | TheBloke | text-generation | [
"transformers",
"gguf",
"llama",
"upstage",
"instruct",
"instruction",
"text-generation",
"en",
"dataset:sciq",
"dataset:metaeval/ScienceQA_text_only",
"dataset:GAIR/lima",
"dataset:Open-Orca/OpenOrca",
"dataset:openbookqa",
"base_model:upstage/llama-30b-instruct-2048",
"base_model:quantized:upstage/llama-30b-instruct-2048",
"license:other",
"region:us"
] | "2023-09-20T00:07:22Z" | 2023-09-27T12:52:53+00:00 | 564 | 4 | ---
base_model: upstage/llama-30b-instruct-2048
datasets:
- sciq
- metaeval/ScienceQA_text_only
- GAIR/lima
- Open-Orca/OpenOrca
- openbookqa
language:
- en
license: other
model_name: Llama 30B Instruct 2048
pipeline_tag: text-generation
tags:
- upstage
- llama
- instruct
- instruction
inference: false
model_creator: upstage
model_type: llama
prompt_template: '### System:
{system_message}
### User:
{prompt}
### Assistant:
'
quantized_by: TheBloke
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# Llama 30B Instruct 2048 - GGUF
- Model creator: [upstage](https://huggingface.co/upstage)
- Original model: [Llama 30B Instruct 2048](https://huggingface.co/upstage/llama-30b-instruct-2048)
<!-- description start -->
## Description
This repo contains GGUF format model files for [Upstage's Llama 30B Instruct 2048](https://huggingface.co/upstage/llama-30b-instruct-2048).
Many thanks to William Beauchamp from [Chai](https://chai-research.com/) for providing the hardware used to make and upload these files!
<!-- description end -->
<!-- README_GGUF.md-about-gguf start -->
### About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplate list of clients and libraries that are known to support GGUF:
* [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
* [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
* [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
* [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
* [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
* [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server.
* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
* [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
<!-- README_GGUF.md-about-gguf end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF)
* [upstage's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/upstage/llama-30b-instruct-2048)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: Orca-Hashes
```
### System:
{system_message}
### User:
{prompt}
### Assistant:
```
<!-- prompt-template end -->
<!-- compatibility_gguf start -->
## Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
## Explanation of quantisation methods
<details>
<summary>Click to see details</summary>
The new methods available are:
* GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
* GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
* GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
* GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
</details>
<!-- compatibility_gguf end -->
<!-- README_GGUF.md-provided-files start -->
## Provided files
| Name | Quant method | Bits | Size | Max RAM required | Use case |
| ---- | ---- | ---- | ---- | ---- | ----- |
| [upstage-llama-30b-instruct-2048.Q2_K.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q2_K.gguf) | Q2_K | 2 | 13.50 GB| 16.00 GB | smallest, significant quality loss - not recommended for most purposes |
| [upstage-llama-30b-instruct-2048.Q3_K_S.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q3_K_S.gguf) | Q3_K_S | 3 | 14.06 GB| 16.56 GB | very small, high quality loss |
| [upstage-llama-30b-instruct-2048.Q3_K_M.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q3_K_M.gguf) | Q3_K_M | 3 | 15.76 GB| 18.26 GB | very small, high quality loss |
| [upstage-llama-30b-instruct-2048.Q3_K_L.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q3_K_L.gguf) | Q3_K_L | 3 | 17.28 GB| 19.78 GB | small, substantial quality loss |
| [upstage-llama-30b-instruct-2048.Q4_0.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q4_0.gguf) | Q4_0 | 4 | 18.36 GB| 20.86 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| [upstage-llama-30b-instruct-2048.Q4_K_S.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q4_K_S.gguf) | Q4_K_S | 4 | 18.44 GB| 20.94 GB | small, greater quality loss |
| [upstage-llama-30b-instruct-2048.Q4_K_M.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q4_K_M.gguf) | Q4_K_M | 4 | 19.62 GB| 22.12 GB | medium, balanced quality - recommended |
| [upstage-llama-30b-instruct-2048.Q5_0.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q5_0.gguf) | Q5_0 | 5 | 22.40 GB| 24.90 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| [upstage-llama-30b-instruct-2048.Q5_K_S.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q5_K_S.gguf) | Q5_K_S | 5 | 22.40 GB| 24.90 GB | large, low quality loss - recommended |
| [upstage-llama-30b-instruct-2048.Q5_K_M.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q5_K_M.gguf) | Q5_K_M | 5 | 23.05 GB| 25.55 GB | large, very low quality loss - recommended |
| [upstage-llama-30b-instruct-2048.Q6_K.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q6_K.gguf) | Q6_K | 6 | 26.69 GB| 29.19 GB | very large, extremely low quality loss |
| [upstage-llama-30b-instruct-2048.Q8_0.gguf](https://huggingface.co/TheBloke/upstage-llama-30b-instruct-2048-GGUF/blob/main/upstage-llama-30b-instruct-2048.Q8_0.gguf) | Q8_0 | 8 | 34.57 GB| 37.07 GB | very large, extremely low quality loss - not recommended |
**Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
<!-- README_GGUF.md-provided-files end -->
<!-- README_GGUF.md-how-to-download start -->
## How to download GGUF files
**Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
- LM Studio
- LoLLMS Web UI
- Faraday.dev
### In `text-generation-webui`
Under Download Model, you can enter the model repo: TheBloke/upstage-llama-30b-instruct-2048-GGUF and below it, a specific filename to download, such as: upstage-llama-30b-instruct-2048.Q4_K_M.gguf.
Then click Download.
### On the command line, including multiple files at once
I recommend using the `huggingface-hub` Python library:
```shell
pip3 install huggingface-hub
```
Then you can download any individual model file to the current directory, at high speed, with a command like this:
```shell
huggingface-cli download TheBloke/upstage-llama-30b-instruct-2048-GGUF upstage-llama-30b-instruct-2048.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
<details>
<summary>More advanced huggingface-cli download usage</summary>
You can also download multiple files at once with a pattern:
```shell
huggingface-cli download TheBloke/upstage-llama-30b-instruct-2048-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
```
For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
```shell
pip3 install hf_transfer
```
And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
```shell
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/upstage-llama-30b-instruct-2048-GGUF upstage-llama-30b-instruct-2048.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
</details>
<!-- README_GGUF.md-how-to-download end -->
<!-- README_GGUF.md-how-to-run start -->
## Example `llama.cpp` command
Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
```shell
./main -ngl 32 -m upstage-llama-30b-instruct-2048.Q4_K_M.gguf --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### System:\n{system_message}\n\n### User:\n{prompt}\n\n### Assistant:"
```
Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change `-c 2048` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.
If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
## How to run in `text-generation-webui`
Further instructions here: [text-generation-webui/docs/llama.cpp.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp.md).
## How to run from Python code
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries.
### How to load this model in Python code, using ctransformers
#### First install the package
Run one of the following commands, according to your system:
```shell
# Base ctransformers with no GPU acceleration
pip install ctransformers
# Or with CUDA GPU acceleration
pip install ctransformers[cuda]
# Or with AMD ROCm GPU acceleration (Linux only)
CT_HIPBLAS=1 pip install ctransformers --no-binary ctransformers
# Or with Metal GPU acceleration for macOS systems only
CT_METAL=1 pip install ctransformers --no-binary ctransformers
```
#### Simple ctransformers example code
```python
from ctransformers import AutoModelForCausalLM
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = AutoModelForCausalLM.from_pretrained("TheBloke/upstage-llama-30b-instruct-2048-GGUF", model_file="upstage-llama-30b-instruct-2048.Q4_K_M.gguf", model_type="llama", gpu_layers=50)
print(llm("AI is going to"))
```
## How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
* [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
* [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
<!-- README_GGUF.md-how-to-run end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Alicia Loh, Stephen Murray, K, Ajan Kanaga, RoA, Magnesian, Deo Leter, Olakabola, Eugene Pentland, zynix, Deep Realms, Raymond Fosdick, Elijah Stavena, Iucharbius, Erik Bjäreholt, Luis Javier Navarrete Lozano, Nicholas, theTransient, John Detwiler, alfie_i, knownsqashed, Mano Prime, Willem Michiel, Enrico Ros, LangChain4j, OG, Michael Dempsey, Pierre Kircher, Pedro Madruga, James Bentley, Thomas Belote, Luke @flexchar, Leonard Tan, Johann-Peter Hartmann, Illia Dulskyi, Fen Risland, Chadd, S_X, Jeff Scroggin, Ken Nordquist, Sean Connelly, Artur Olbinski, Swaroop Kallakuri, Jack West, Ai Maven, David Ziegler, Russ Johnson, transmissions 11, John Villwock, Alps Aficionado, Clay Pascal, Viktor Bowallius, Subspace Studios, Rainer Wilmers, Trenton Dambrowitz, vamX, Michael Levine, 준교 김, Brandon Frisco, Kalila, Trailburnt, Randy H, Talal Aujan, Nathan Dryer, Vadim, 阿明, ReadyPlayerEmma, Tiffany J. Kim, George Stoitzev, Spencer Kim, Jerry Meng, Gabriel Tamborski, Cory Kujawski, Jeffrey Morgan, Spiking Neurons AB, Edmond Seymore, Alexandros Triantafyllidis, Lone Striker, Cap'n Zoog, Nikolai Manek, danny, ya boyyy, Derek Yates, usrbinkat, Mandus, TL, Nathan LeClaire, subjectnull, Imad Khwaja, webtim, Raven Klaugh, Asp the Wyvern, Gabriel Puliatti, Caitlyn Gatomon, Joseph William Delisle, Jonathan Leane, Luke Pendergrass, SuperWojo, Sebastain Graf, Will Dee, Fred von Graf, Andrey, Dan Guido, Daniel P. Andersen, Nitin Borwankar, Elle, Vitor Caleffi, biorpg, jjj, NimbleBox.ai, Pieter, Matthew Berman, terasurfer, Michael Davis, Alex, Stanislav Ovsiannikov
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
<!-- original-model-card start -->
# Original model card: Upstage's Llama 30B Instruct 2048
# LLaMa-30b-instruct-2048 model card
## Model Details
* **Developed by**: [Upstage](https://en.upstage.ai)
* **Backbone Model**: [LLaMA](https://github.com/facebookresearch/llama/tree/llama_v1)
* **Variations**: It has different model parameter sizes and sequence lengths: [30B/1024](https://huggingface.co/upstage/llama-30b-instruct), [30B/2048](https://huggingface.co/upstage/llama-30b-instruct-2048), [65B/1024](https://huggingface.co/upstage/llama-65b-instruct)
* **Language(s)**: English
* **Library**: [HuggingFace Transformers](https://github.com/huggingface/transformers)
* **License**: This model is under a **Non-commercial** Bespoke License and governed by the Meta license. You should only use this repository if you have been granted access to the model by filling out [this form](https://docs.google.com/forms/d/e/1FAIpQLSfqNECQnMkycAp2jP4Z9TFX0cGR4uf7b_fBxjY_OjhJILlKGA/viewform), but have either lost your copy of the weights or encountered issues converting them to the Transformers format
* **Where to send comments**: Instructions on how to provide feedback or comments on a model can be found by opening an issue in the [Hugging Face community's model repository](https://huggingface.co/upstage/llama-30b-instruct-2048/discussions)
* **Contact**: For questions and comments about the model, please email [contact@upstage.ai](mailto:contact@upstage.ai)
## Dataset Details
### Used Datasets
- [openbookqa](https://huggingface.co/datasets/openbookqa)
- [sciq](https://huggingface.co/datasets/sciq)
- [Open-Orca/OpenOrca](https://huggingface.co/datasets/Open-Orca/OpenOrca)
- [metaeval/ScienceQA_text_only](https://huggingface.co/datasets/metaeval/ScienceQA_text_only)
- [GAIR/lima](https://huggingface.co/datasets/GAIR/lima)
- No other data was used except for the dataset mentioned above
### Prompt Template
```
### System:
{System}
### User:
{User}
### Assistant:
{Assistant}
```
## Usage
- Tested on A100 80GB
- Our model can handle up to 10k+ input tokens, thanks to the `rope_scaling` option
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
tokenizer = AutoTokenizer.from_pretrained("upstage/llama-30b-instruct-2048")
model = AutoModelForCausalLM.from_pretrained(
"upstage/llama-30b-instruct-2048",
device_map="auto",
torch_dtype=torch.float16,
load_in_8bit=True,
rope_scaling={"type": "dynamic", "factor": 2} # allows handling of longer inputs
)
prompt = "### User:\nThomas is healthy, but he has to go to the hospital. What could be the reasons?\n\n### Assistant:\n"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
del inputs["token_type_ids"]
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
output = model.generate(**inputs, streamer=streamer, use_cache=True, max_new_tokens=float('inf'))
output_text = tokenizer.decode(output[0], skip_special_tokens=True)
```
## Hardware and Software
* **Hardware**: We utilized an A100x8 * 1 for training our model
* **Training Factors**: We fine-tuned this model using a combination of the [DeepSpeed library](https://github.com/microsoft/DeepSpeed) and the [HuggingFace Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) / [HuggingFace Accelerate](https://huggingface.co/docs/accelerate/index)
## Evaluation Results
### Overview
- We conducted a performance evaluation based on the tasks being evaluated on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
We evaluated our model on four benchmark datasets, which include `ARC-Challenge`, `HellaSwag`, `MMLU`, and `TruthfulQA`
We used the [lm-evaluation-harness repository](https://github.com/EleutherAI/lm-evaluation-harness), specifically commit [b281b0921b636bc36ad05c0b0b0763bd6dd43463](https://github.com/EleutherAI/lm-evaluation-harness/tree/b281b0921b636bc36ad05c0b0b0763bd6dd43463)
- We used [MT-bench](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge), a set of challenging multi-turn open-ended questions, to evaluate the models
### Main Results
| Model | H4(Avg) | ARC | HellaSwag | MMLU | TruthfulQA | | MT_Bench |
|--------------------------------------------------------------------|----------|----------|----------|------|----------|-|-------------|
| **[Llama-2-70b-instruct-v2](https://huggingface.co/upstage/Llama-2-70b-instruct-v2)**(Ours, Open LLM Leaderboard) | **73** | **71.1** | **87.9** | **70.6** | **62.2** | | **7.44063** |
| [Llama-2-70b-instruct](https://huggingface.co/upstage/Llama-2-70b-instruct) (Ours, Open LLM Leaderboard) | 72.3 | 70.9 | 87.5 | 69.8 | 61 | | 7.24375 |
| [llama-65b-instruct](https://huggingface.co/upstage/llama-65b-instruct) (Ours, Open LLM Leaderboard) | 69.4 | 67.6 | 86.5 | 64.9 | 58.8 | | |
| Llama-2-70b-hf | 67.3 | 67.3 | 87.3 | 69.8 | 44.9 | | |
| [llama-30b-instruct-2048](https://huggingface.co/upstage/llama-30b-instruct-2048) (***Ours***, ***Open LLM Leaderboard***) | 67.0 | 64.9 | 84.9 | 61.9 | 56.3 | | |
| [llama-30b-instruct](https://huggingface.co/upstage/llama-30b-instruct) (Ours, Open LLM Leaderboard) | 65.2 | 62.5 | 86.2 | 59.4 | 52.8 | | |
| llama-65b | 64.2 | 63.5 | 86.1 | 63.9 | 43.4 | | |
| falcon-40b-instruct | 63.4 | 61.6 | 84.3 | 55.4 | 52.5 | | |
### Scripts for H4 Score Reproduction
- Prepare evaluation environments:
```
# clone the repository
git clone https://github.com/EleutherAI/lm-evaluation-harness.git
# check out the specific commit
git checkout b281b0921b636bc36ad05c0b0b0763bd6dd43463
# change to the repository directory
cd lm-evaluation-harness
```
## Ethical Issues
### Ethical Considerations
- There were no ethical issues involved, as we did not include the benchmark test set or the training set in the model's training process
## Contact Us
### Why Upstage LLM?
- [Upstage](https://en.upstage.ai)'s LLM research has yielded remarkable results. As of August 1st, our 70B model has reached the top spot in openLLM rankings, marking itself as the current leading performer globally. Recognizing the immense potential in implementing private LLM to actual businesses, we invite you to easily apply private LLM and fine-tune it with your own data. For a seamless and tailored solution, please do not hesitate to reach out to us. ► [click here to contact](https://www.upstage.ai/private-llm?utm_source=huggingface&utm_medium=link&utm_campaign=privatellm)
<!-- original-model-card end -->
| [
"SCIQ"
] |
TheBloke/xDAN-L1-Chat-RL-v1-GGUF | TheBloke | null | [
"transformers",
"gguf",
"mistral",
"xDAN-AI",
"OpenOrca",
"DPO",
"Self-Think",
"en",
"dataset:Open-Orca/OpenOrca",
"dataset:Intel/orca_dpo_pairs",
"base_model:xDAN-AI/xDAN-L1-Chat-RL-v1",
"base_model:quantized:xDAN-AI/xDAN-L1-Chat-RL-v1",
"license:cc-by-4.0",
"region:us"
] | "2023-12-26T08:43:43Z" | 2023-12-26T10:02:46+00:00 | 562 | 12 | ---
base_model: xDAN-AI/xDAN-L1-Chat-RL-v1
datasets:
- Open-Orca/OpenOrca
- Intel/orca_dpo_pairs
language:
- en
license: cc-by-4.0
model_name: xDAN L1 Chat RL v1
tags:
- xDAN-AI
- OpenOrca
- DPO
- Self-Think
inference: false
model_creator: xDAN-AI
model_type: mistral
prompt_template: 'Below is an instruction that describes a task. Write a response
that appropriately completes the request.
### Instruction:
{prompt}
### Response:
'
quantized_by: TheBloke
---
<!-- markdownlint-disable MD041 -->
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# xDAN L1 Chat RL v1 - GGUF
- Model creator: [xDAN-AI](https://huggingface.co/xDAN-AI)
- Original model: [xDAN L1 Chat RL v1](https://huggingface.co/xDAN-AI/xDAN-L1-Chat-RL-v1)
<!-- description start -->
## Description
This repo contains GGUF format model files for [xDAN-AI's xDAN L1 Chat RL v1](https://huggingface.co/xDAN-AI/xDAN-L1-Chat-RL-v1).
These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
<!-- description end -->
<!-- README_GGUF.md-about-gguf start -->
### About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplete list of clients and libraries that are known to support GGUF:
* [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
* [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
* [GPT4All](https://gpt4all.io/index.html), a free and open source local running GUI, supporting Windows, Linux and macOS with full GPU accel.
* [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration. Linux available, in beta as of 27/11/2023.
* [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
* [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
* [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
* [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server. Note, as of time of writing (November 27th 2023), ctransformers has not been updated in a long time and does not support many recent models.
<!-- README_GGUF.md-about-gguf end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF)
* [xDAN-AI's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/xDAN-AI/xDAN-L1-Chat-RL-v1)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: Alpaca
```
Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{prompt}
### Response:
```
<!-- prompt-template end -->
<!-- compatibility_gguf start -->
## Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
## Explanation of quantisation methods
<details>
<summary>Click to see details</summary>
The new methods available are:
* GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
* GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
* GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
* GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
</details>
<!-- compatibility_gguf end -->
<!-- README_GGUF.md-provided-files start -->
## Provided files
| Name | Quant method | Bits | Size | Max RAM required | Use case |
| ---- | ---- | ---- | ---- | ---- | ----- |
| [xdan-l1-chat-rl-v1.Q2_K.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q2_K.gguf) | Q2_K | 2 | 3.08 GB| 5.58 GB | smallest, significant quality loss - not recommended for most purposes |
| [xdan-l1-chat-rl-v1.Q3_K_S.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q3_K_S.gguf) | Q3_K_S | 3 | 3.16 GB| 5.66 GB | very small, high quality loss |
| [xdan-l1-chat-rl-v1.Q3_K_M.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q3_K_M.gguf) | Q3_K_M | 3 | 3.52 GB| 6.02 GB | very small, high quality loss |
| [xdan-l1-chat-rl-v1.Q3_K_L.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q3_K_L.gguf) | Q3_K_L | 3 | 3.82 GB| 6.32 GB | small, substantial quality loss |
| [xdan-l1-chat-rl-v1.Q4_0.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q4_0.gguf) | Q4_0 | 4 | 4.11 GB| 6.61 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| [xdan-l1-chat-rl-v1.Q4_K_S.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q4_K_S.gguf) | Q4_K_S | 4 | 4.14 GB| 6.64 GB | small, greater quality loss |
| [xdan-l1-chat-rl-v1.Q4_K_M.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q4_K_M.gguf) | Q4_K_M | 4 | 4.37 GB| 6.87 GB | medium, balanced quality - recommended |
| [xdan-l1-chat-rl-v1.Q5_0.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q5_0.gguf) | Q5_0 | 5 | 5.00 GB| 7.50 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| [xdan-l1-chat-rl-v1.Q5_K_S.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q5_K_S.gguf) | Q5_K_S | 5 | 5.00 GB| 7.50 GB | large, low quality loss - recommended |
| [xdan-l1-chat-rl-v1.Q5_K_M.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q5_K_M.gguf) | Q5_K_M | 5 | 5.13 GB| 7.63 GB | large, very low quality loss - recommended |
| [xdan-l1-chat-rl-v1.Q6_K.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q6_K.gguf) | Q6_K | 6 | 5.94 GB| 8.44 GB | very large, extremely low quality loss |
| [xdan-l1-chat-rl-v1.Q8_0.gguf](https://huggingface.co/TheBloke/xDAN-L1-Chat-RL-v1-GGUF/blob/main/xdan-l1-chat-rl-v1.Q8_0.gguf) | Q8_0 | 8 | 7.70 GB| 10.20 GB | very large, extremely low quality loss - not recommended |
**Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
<!-- README_GGUF.md-provided-files end -->
<!-- README_GGUF.md-how-to-download start -->
## How to download GGUF files
**Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
* LM Studio
* LoLLMS Web UI
* Faraday.dev
### In `text-generation-webui`
Under Download Model, you can enter the model repo: TheBloke/xDAN-L1-Chat-RL-v1-GGUF and below it, a specific filename to download, such as: xdan-l1-chat-rl-v1.Q4_K_M.gguf.
Then click Download.
### On the command line, including multiple files at once
I recommend using the `huggingface-hub` Python library:
```shell
pip3 install huggingface-hub
```
Then you can download any individual model file to the current directory, at high speed, with a command like this:
```shell
huggingface-cli download TheBloke/xDAN-L1-Chat-RL-v1-GGUF xdan-l1-chat-rl-v1.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
<details>
<summary>More advanced huggingface-cli download usage (click to read)</summary>
You can also download multiple files at once with a pattern:
```shell
huggingface-cli download TheBloke/xDAN-L1-Chat-RL-v1-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
```
For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
```shell
pip3 install hf_transfer
```
And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
```shell
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/xDAN-L1-Chat-RL-v1-GGUF xdan-l1-chat-rl-v1.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
</details>
<!-- README_GGUF.md-how-to-download end -->
<!-- README_GGUF.md-how-to-run start -->
## Example `llama.cpp` command
Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
```shell
./main -ngl 35 -m xdan-l1-chat-rl-v1.Q4_K_M.gguf --color -c 32768 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{prompt}\n\n### Response:"
```
Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change `-c 32768` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically. Note that longer sequence lengths require much more resources, so you may need to reduce this value.
If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
## How to run in `text-generation-webui`
Further instructions can be found in the text-generation-webui documentation, here: [text-generation-webui/docs/04 ‐ Model Tab.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/04%20%E2%80%90%20Model%20Tab.md#llamacpp).
## How to run from Python code
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.
### How to load this model in Python code, using llama-cpp-python
For full documentation, please see: [llama-cpp-python docs](https://abetlen.github.io/llama-cpp-python/).
#### First install the package
Run one of the following commands, according to your system:
```shell
# Base ctransformers with no GPU acceleration
pip install llama-cpp-python
# With NVidia CUDA acceleration
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
# Or with OpenBLAS acceleration
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
# Or with CLBLast acceleration
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
# Or with AMD ROCm GPU acceleration (Linux only)
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
# Or with Metal GPU acceleration for macOS systems only
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
pip install llama-cpp-python
```
#### Simple llama-cpp-python example code
```python
from llama_cpp import Llama
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = Llama(
model_path="./xdan-l1-chat-rl-v1.Q4_K_M.gguf", # Download the model file first
n_ctx=32768, # The max sequence length to use - note that longer sequence lengths require much more resources
n_threads=8, # The number of CPU threads to use, tailor to your system and the resulting performance
n_gpu_layers=35 # The number of layers to offload to GPU, if you have GPU acceleration available
)
# Simple inference example
output = llm(
"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{prompt}\n\n### Response:", # Prompt
max_tokens=512, # Generate up to 512 tokens
stop=["</s>"], # Example stop token - not necessarily correct for this specific model! Please check before using.
echo=True # Whether to echo the prompt
)
# Chat Completion API
llm = Llama(model_path="./xdan-l1-chat-rl-v1.Q4_K_M.gguf", chat_format="llama-2") # Set chat_format according to the model you are using
llm.create_chat_completion(
messages = [
{"role": "system", "content": "You are a story writing assistant."},
{
"role": "user",
"content": "Write a story about llamas."
}
]
)
```
## How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
* [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
* [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
<!-- README_GGUF.md-how-to-run end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Michael Levine, 阿明, Trailburnt, Nikolai Manek, John Detwiler, Randy H, Will Dee, Sebastain Graf, NimbleBox.ai, Eugene Pentland, Emad Mostaque, Ai Maven, Jim Angel, Jeff Scroggin, Michael Davis, Manuel Alberto Morcote, Stephen Murray, Robert, Justin Joy, Luke @flexchar, Brandon Frisco, Elijah Stavena, S_X, Dan Guido, Undi ., Komninos Chatzipapas, Shadi, theTransient, Lone Striker, Raven Klaugh, jjj, Cap'n Zoog, Michel-Marie MAUDET (LINAGORA), Matthew Berman, David, Fen Risland, Omer Bin Jawed, Luke Pendergrass, Kalila, OG, Erik Bjäreholt, Rooh Singh, Joseph William Delisle, Dan Lewis, TL, John Villwock, AzureBlack, Brad, Pedro Madruga, Caitlyn Gatomon, K, jinyuan sun, Mano Prime, Alex, Jeffrey Morgan, Alicia Loh, Illia Dulskyi, Chadd, transmissions 11, fincy, Rainer Wilmers, ReadyPlayerEmma, knownsqashed, Mandus, biorpg, Deo Leter, Brandon Phillips, SuperWojo, Sean Connelly, Iucharbius, Jack West, Harry Royden McLaughlin, Nicholas, terasurfer, Vitor Caleffi, Duane Dunston, Johann-Peter Hartmann, David Ziegler, Olakabola, Ken Nordquist, Trenton Dambrowitz, Tom X Nguyen, Vadim, Ajan Kanaga, Leonard Tan, Clay Pascal, Alexandros Triantafyllidis, JM33133, Xule, vamX, ya boyyy, subjectnull, Talal Aujan, Alps Aficionado, wassieverse, Ari Malik, James Bentley, Woland, Spencer Kim, Michael Dempsey, Fred von Graf, Elle, zynix, William Richards, Stanislav Ovsiannikov, Edmond Seymore, Jonathan Leane, Martin Kemka, usrbinkat, Enrico Ros
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
<!-- original-model-card start -->
# Original model card: xDAN-AI's xDAN L1 Chat RL v1
<div style="display: flex; justify-content: center; align-items: center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/643197ac288c9775673a01e9/tVAcwKkIH5vkfzqgqHeHi.png" style="width: 45%;">
</div
>
<p align="center">
<big><b>Top 1 Performer on MT-bench🏆</b
></big>
</p>
<p align="center">
<strong>**The first top model which is performance at Humanalities, Coding and Writing with 7b. **</strong>
</p>
<p
align="center"
<a href="The TOP1 MT-Bench Model">xDAN-AI</a> •
>
<a href="https://discord.gg/7NrMX5AK">Discord</a> •
<a href="https://twitter.com/shootime007">Twitter</a> •
<a href="https://huggingface.co/xDAN-AI">Huggingface</a>
</p>
<p align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/643197ac288c9775673a01e9/QANDZApzpTHM6sBsjmdew.png" alt="Image" width="50%">
</p>
## Outperformer GPT3.5turbo & Claude-v1
## Touch nearby GPT4 on MT-Bench
**########## First turn ##########**
| model | turn | score | size
|--------------------|------|----------|--------
| gpt-4 | 1 | 8.95625 | -
| **xDAN-L1-Chat-RL-v1** | 1 | **8.87500** | **7b**
| xDAN-L2-Chat-RL-v2 | 1 | 8.78750 | 30b
| claude-v1 | 1 | 8.15000 | -
| gpt-3.5-turbo | 1 | 8.07500 | 20b
| vicuna-33b-v1.3 | 1 | 7.45625 | 33b
| wizardlm-30b | 1 | 7.13125 | 30b
| oasst-sft-7-llama-30b | 1 | 7.10625 | 30b
| Llama-2-70b-chat | 1 | 6.98750 | 70b
########## Second turn ##########
| model | turn | score | size
|--------------------|------|-----------|--------
| gpt-4 | 2 | 9.025000 | -
| xDAN-L2-Chat-RL-v2 | 2 | 8.087500 | 30b
| **xDAN-L1-Chat-RL-v1** | 2 | **7.825000** | **7b**
| gpt-3.5-turbo | 2 | 7.812500 | 20b
| claude-v1 | 2 | 7.650000 | -
| wizardlm-30b | 2 | 6.887500 | 30b
| vicuna-33b-v1.3 | 2 | 6.787500 | 33b
| Llama-2-70b-chat | 2 | 6.725000 | 70b
########## Average turn##########
| model | score | size
|--------------------|-----------|--------
| gpt-4 | 8.990625 | -
| xDAN-L2-Chat-RL-v2 | 8.437500 | 30b
| **xDAN-L1-Chat-RL-v1** | **8.350000** | **7b**
| gpt-3.5-turbo | 7.943750 | 20b
| claude-v1 | 7.900000 | -
| vicuna-33b-v1.3 | 7.121875 | 33b
| wizardlm-30b | 7.009375 | 30b
| Llama-2-70b-chat | 6.856250 | 70b
### Prompt Template(Alpaca)
You are a helpful assistant named DAN. You are an expert in worldly knowledge, skilled in employing a probing questioning strategy,
and you carefully consider each step before providing answers.
\n\n### Instruction:\n{instruction}\n\n### Response:
### Dataset:
1. Selected from OpenOrca
2. Intel Orca-DPO-Pairs
3. Privately Crafted Dataset
### Training:
1. SFT with Mixed dataset from OpenOrca
2. The Next DPO dataset made by xDAN-AI
3. The Next DPO Training method by xDAN-AI
## Created By xDAN-AI at 2023-12-15
## Eval by FastChat: https://github.com/lm-sys/FastChat.git
Disclaimer
We employ data compliance checking algorithms during the training of our language model to strive for the highest degree of compliance. However, given the intricate nature of data and the vast array of potential usage scenarios for the model, we cannot assure that it will always generate correct and reasonable outputs. Users should be cognizant of the risk of the model producing problematic outputs. Our organization will not bear responsibility for any risks or issues stemming from misuse, misguidance, illegal use, and related misinformation, as well as any consequent data security concerns.
About xDAN-AI xDAN-AI is a top lead high-performance model factory. For detailed information and further insights into our cutting-edge technology and offerings, please visit our website: https://www.xdan.ai.
<!-- original-model-card end -->
| [
"BEAR"
] |
jinaai/jina-embedding-l-en-v1 | jinaai | sentence-similarity | [
"sentence-transformers",
"pytorch",
"t5",
"finetuner",
"mteb",
"feature-extraction",
"sentence-similarity",
"custom_code",
"en",
"dataset:jinaai/negation-dataset",
"arxiv:2307.11224",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2023-07-09T08:54:06Z" | 2025-01-06T16:30:42+00:00 | 561 | 24 | ---
datasets:
- jinaai/negation-dataset
language: en
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- finetuner
- mteb
- sentence-transformers
- feature-extraction
- sentence-similarity
model-index:
- name: jina-triplets-large
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 68.92537313432835
- type: ap
value: 29.723758877632513
- type: f1
value: 61.909704211663794
- task:
type: Classification
dataset:
name: MTEB AmazonPolarityClassification
type: mteb/amazon_polarity
config: default
split: test
revision: e2d317d38cd51312af73b3d32a06d1a08b442046
metrics:
- type: accuracy
value: 69.13669999999999
- type: ap
value: 65.30216072238086
- type: f1
value: 67.1890891071034
- task:
type: Classification
dataset:
name: MTEB AmazonReviewsClassification (en)
type: mteb/amazon_reviews_multi
config: en
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 31.384
- type: f1
value: 30.016752348953723
- task:
type: Retrieval
dataset:
name: MTEB ArguAna
type: arguana
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 23.613
- type: map_at_10
value: 37.897
- type: map_at_100
value: 39.093
- type: map_at_1000
value: 39.109
- type: map_at_3
value: 32.824
- type: map_at_5
value: 35.679
- type: mrr_at_1
value: 23.826
- type: mrr_at_10
value: 37.997
- type: mrr_at_100
value: 39.186
- type: mrr_at_1000
value: 39.202
- type: mrr_at_3
value: 32.918
- type: mrr_at_5
value: 35.748999999999995
- type: ndcg_at_1
value: 23.613
- type: ndcg_at_10
value: 46.482
- type: ndcg_at_100
value: 51.55499999999999
- type: ndcg_at_1000
value: 51.974
- type: ndcg_at_3
value: 35.964
- type: ndcg_at_5
value: 41.144999999999996
- type: precision_at_1
value: 23.613
- type: precision_at_10
value: 7.417999999999999
- type: precision_at_100
value: 0.963
- type: precision_at_1000
value: 0.1
- type: precision_at_3
value: 15.031
- type: precision_at_5
value: 11.55
- type: recall_at_1
value: 23.613
- type: recall_at_10
value: 74.182
- type: recall_at_100
value: 96.30199999999999
- type: recall_at_1000
value: 99.57300000000001
- type: recall_at_3
value: 45.092
- type: recall_at_5
value: 57.752
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringP2P
type: mteb/arxiv-clustering-p2p
config: default
split: test
revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d
metrics:
- type: v_measure
value: 40.51285742156528
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringS2S
type: mteb/arxiv-clustering-s2s
config: default
split: test
revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53
metrics:
- type: v_measure
value: 31.5825964077496
- task:
type: Reranking
dataset:
name: MTEB AskUbuntuDupQuestions
type: mteb/askubuntudupquestions-reranking
config: default
split: test
revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
metrics:
- type: map
value: 62.830281630546835
- type: mrr
value: 75.93072593765115
- task:
type: STS
dataset:
name: MTEB BIOSSES
type: mteb/biosses-sts
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cos_sim_pearson
value: 87.26764516732737
- type: cos_sim_spearman
value: 84.42541766631741
- type: euclidean_pearson
value: 48.71357447655235
- type: euclidean_spearman
value: 49.2023259276511
- type: manhattan_pearson
value: 48.36366272727299
- type: manhattan_spearman
value: 48.457128224924354
- task:
type: Classification
dataset:
name: MTEB Banking77Classification
type: mteb/banking77
config: default
split: test
revision: 0fd18e25b25c072e09e0d92ab615fda904d66300
metrics:
- type: accuracy
value: 85.3409090909091
- type: f1
value: 85.25262617676835
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringP2P
type: mteb/biorxiv-clustering-p2p
config: default
split: test
revision: 65b79d1d13f80053f67aca9498d9402c2d9f1f40
metrics:
- type: v_measure
value: 33.560193912974974
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringS2S
type: mteb/biorxiv-clustering-s2s
config: default
split: test
revision: 258694dd0231531bc1fd9de6ceb52a0853c6d908
metrics:
- type: v_measure
value: 28.4426572644577
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval
type: BeIR/cqadupstack
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 27.822999999999997
- type: map_at_10
value: 39.088
- type: map_at_100
value: 40.561
- type: map_at_1000
value: 40.69
- type: map_at_3
value: 35.701
- type: map_at_5
value: 37.556
- type: mrr_at_1
value: 33.906
- type: mrr_at_10
value: 44.527
- type: mrr_at_100
value: 45.403999999999996
- type: mrr_at_1000
value: 45.452
- type: mrr_at_3
value: 41.726
- type: mrr_at_5
value: 43.314
- type: ndcg_at_1
value: 33.906
- type: ndcg_at_10
value: 45.591
- type: ndcg_at_100
value: 51.041000000000004
- type: ndcg_at_1000
value: 53.1
- type: ndcg_at_3
value: 40.324
- type: ndcg_at_5
value: 42.723
- type: precision_at_1
value: 33.906
- type: precision_at_10
value: 8.655
- type: precision_at_100
value: 1.418
- type: precision_at_1000
value: 0.19499999999999998
- type: precision_at_3
value: 19.123
- type: precision_at_5
value: 13.963000000000001
- type: recall_at_1
value: 27.822999999999997
- type: recall_at_10
value: 58.63699999999999
- type: recall_at_100
value: 80.874
- type: recall_at_1000
value: 93.82000000000001
- type: recall_at_3
value: 44.116
- type: recall_at_5
value: 50.178999999999995
- type: map_at_1
value: 26.823999999999998
- type: map_at_10
value: 37.006
- type: map_at_100
value: 38.256
- type: map_at_1000
value: 38.397999999999996
- type: map_at_3
value: 34.011
- type: map_at_5
value: 35.643
- type: mrr_at_1
value: 34.268
- type: mrr_at_10
value: 43.374
- type: mrr_at_100
value: 44.096000000000004
- type: mrr_at_1000
value: 44.144
- type: mrr_at_3
value: 41.008
- type: mrr_at_5
value: 42.359
- type: ndcg_at_1
value: 34.268
- type: ndcg_at_10
value: 43.02
- type: ndcg_at_100
value: 47.747
- type: ndcg_at_1000
value: 50.019999999999996
- type: ndcg_at_3
value: 38.687
- type: ndcg_at_5
value: 40.647
- type: precision_at_1
value: 34.268
- type: precision_at_10
value: 8.261000000000001
- type: precision_at_100
value: 1.376
- type: precision_at_1000
value: 0.189
- type: precision_at_3
value: 19.108
- type: precision_at_5
value: 13.489999999999998
- type: recall_at_1
value: 26.823999999999998
- type: recall_at_10
value: 53.84100000000001
- type: recall_at_100
value: 73.992
- type: recall_at_1000
value: 88.524
- type: recall_at_3
value: 40.711000000000006
- type: recall_at_5
value: 46.477000000000004
- type: map_at_1
value: 34.307
- type: map_at_10
value: 45.144
- type: map_at_100
value: 46.351
- type: map_at_1000
value: 46.414
- type: map_at_3
value: 42.315000000000005
- type: map_at_5
value: 43.991
- type: mrr_at_1
value: 39.06
- type: mrr_at_10
value: 48.612
- type: mrr_at_100
value: 49.425000000000004
- type: mrr_at_1000
value: 49.458999999999996
- type: mrr_at_3
value: 46.144
- type: mrr_at_5
value: 47.654999999999994
- type: ndcg_at_1
value: 39.06
- type: ndcg_at_10
value: 50.647
- type: ndcg_at_100
value: 55.620000000000005
- type: ndcg_at_1000
value: 56.976000000000006
- type: ndcg_at_3
value: 45.705
- type: ndcg_at_5
value: 48.269
- type: precision_at_1
value: 39.06
- type: precision_at_10
value: 8.082
- type: precision_at_100
value: 1.161
- type: precision_at_1000
value: 0.133
- type: precision_at_3
value: 20.376
- type: precision_at_5
value: 14.069
- type: recall_at_1
value: 34.307
- type: recall_at_10
value: 63.497
- type: recall_at_100
value: 85.038
- type: recall_at_1000
value: 94.782
- type: recall_at_3
value: 50.209
- type: recall_at_5
value: 56.525000000000006
- type: map_at_1
value: 26.448
- type: map_at_10
value: 34.86
- type: map_at_100
value: 36.004999999999995
- type: map_at_1000
value: 36.081
- type: map_at_3
value: 32.527
- type: map_at_5
value: 33.955
- type: mrr_at_1
value: 28.701
- type: mrr_at_10
value: 36.909
- type: mrr_at_100
value: 37.89
- type: mrr_at_1000
value: 37.945
- type: mrr_at_3
value: 34.576
- type: mrr_at_5
value: 35.966
- type: ndcg_at_1
value: 28.701
- type: ndcg_at_10
value: 39.507999999999996
- type: ndcg_at_100
value: 45.056000000000004
- type: ndcg_at_1000
value: 47.034
- type: ndcg_at_3
value: 34.985
- type: ndcg_at_5
value: 37.384
- type: precision_at_1
value: 28.701
- type: precision_at_10
value: 5.921
- type: precision_at_100
value: 0.914
- type: precision_at_1000
value: 0.11199999999999999
- type: precision_at_3
value: 14.689
- type: precision_at_5
value: 10.237
- type: recall_at_1
value: 26.448
- type: recall_at_10
value: 51.781
- type: recall_at_100
value: 77.142
- type: recall_at_1000
value: 92.10000000000001
- type: recall_at_3
value: 39.698
- type: recall_at_5
value: 45.469
- type: map_at_1
value: 14.174000000000001
- type: map_at_10
value: 22.019
- type: map_at_100
value: 23.18
- type: map_at_1000
value: 23.304
- type: map_at_3
value: 19.332
- type: map_at_5
value: 20.816000000000003
- type: mrr_at_1
value: 17.785999999999998
- type: mrr_at_10
value: 26.233
- type: mrr_at_100
value: 27.254
- type: mrr_at_1000
value: 27.328000000000003
- type: mrr_at_3
value: 23.653
- type: mrr_at_5
value: 25.095
- type: ndcg_at_1
value: 17.785999999999998
- type: ndcg_at_10
value: 27.236
- type: ndcg_at_100
value: 32.932
- type: ndcg_at_1000
value: 36.134
- type: ndcg_at_3
value: 22.33
- type: ndcg_at_5
value: 24.573999999999998
- type: precision_at_1
value: 17.785999999999998
- type: precision_at_10
value: 5.286
- type: precision_at_100
value: 0.9369999999999999
- type: precision_at_1000
value: 0.136
- type: precision_at_3
value: 11.07
- type: precision_at_5
value: 8.308
- type: recall_at_1
value: 14.174000000000001
- type: recall_at_10
value: 39.135
- type: recall_at_100
value: 64.095
- type: recall_at_1000
value: 87.485
- type: recall_at_3
value: 25.496999999999996
- type: recall_at_5
value: 31.148999999999997
- type: map_at_1
value: 24.371000000000002
- type: map_at_10
value: 33.074999999999996
- type: map_at_100
value: 34.486
- type: map_at_1000
value: 34.608
- type: map_at_3
value: 30.483
- type: map_at_5
value: 31.972
- type: mrr_at_1
value: 29.548000000000002
- type: mrr_at_10
value: 38.431
- type: mrr_at_100
value: 39.347
- type: mrr_at_1000
value: 39.4
- type: mrr_at_3
value: 35.980000000000004
- type: mrr_at_5
value: 37.413999999999994
- type: ndcg_at_1
value: 29.548000000000002
- type: ndcg_at_10
value: 38.552
- type: ndcg_at_100
value: 44.598
- type: ndcg_at_1000
value: 47.0
- type: ndcg_at_3
value: 34.109
- type: ndcg_at_5
value: 36.263
- type: precision_at_1
value: 29.548000000000002
- type: precision_at_10
value: 6.92
- type: precision_at_100
value: 1.179
- type: precision_at_1000
value: 0.159
- type: precision_at_3
value: 16.137
- type: precision_at_5
value: 11.511000000000001
- type: recall_at_1
value: 24.371000000000002
- type: recall_at_10
value: 49.586999999999996
- type: recall_at_100
value: 75.15899999999999
- type: recall_at_1000
value: 91.06
- type: recall_at_3
value: 37.09
- type: recall_at_5
value: 42.588
- type: map_at_1
value: 24.517
- type: map_at_10
value: 32.969
- type: map_at_100
value: 34.199
- type: map_at_1000
value: 34.322
- type: map_at_3
value: 30.270999999999997
- type: map_at_5
value: 31.863000000000003
- type: mrr_at_1
value: 30.479
- type: mrr_at_10
value: 38.633
- type: mrr_at_100
value: 39.522
- type: mrr_at_1000
value: 39.583
- type: mrr_at_3
value: 36.454
- type: mrr_at_5
value: 37.744
- type: ndcg_at_1
value: 30.479
- type: ndcg_at_10
value: 38.269
- type: ndcg_at_100
value: 43.91
- type: ndcg_at_1000
value: 46.564
- type: ndcg_at_3
value: 34.03
- type: ndcg_at_5
value: 36.155
- type: precision_at_1
value: 30.479
- type: precision_at_10
value: 6.815
- type: precision_at_100
value: 1.138
- type: precision_at_1000
value: 0.158
- type: precision_at_3
value: 16.058
- type: precision_at_5
value: 11.416
- type: recall_at_1
value: 24.517
- type: recall_at_10
value: 48.559000000000005
- type: recall_at_100
value: 73.307
- type: recall_at_1000
value: 91.508
- type: recall_at_3
value: 36.563
- type: recall_at_5
value: 42.375
- type: map_at_1
value: 24.336166666666664
- type: map_at_10
value: 32.80791666666667
- type: map_at_100
value: 34.043416666666666
- type: map_at_1000
value: 34.162749999999996
- type: map_at_3
value: 30.187083333333337
- type: map_at_5
value: 31.637833333333337
- type: mrr_at_1
value: 28.669583333333343
- type: mrr_at_10
value: 36.88616666666667
- type: mrr_at_100
value: 37.80233333333333
- type: mrr_at_1000
value: 37.86141666666666
- type: mrr_at_3
value: 34.537416666666665
- type: mrr_at_5
value: 35.84275
- type: ndcg_at_1
value: 28.669583333333343
- type: ndcg_at_10
value: 37.956916666666665
- type: ndcg_at_100
value: 43.39475
- type: ndcg_at_1000
value: 45.79925
- type: ndcg_at_3
value: 33.43683333333334
- type: ndcg_at_5
value: 35.52575
- type: precision_at_1
value: 28.669583333333343
- type: precision_at_10
value: 6.603833333333335
- type: precision_at_100
value: 1.1079166666666667
- type: precision_at_1000
value: 0.15208333333333335
- type: precision_at_3
value: 15.338750000000001
- type: precision_at_5
value: 10.88775
- type: recall_at_1
value: 24.336166666666664
- type: recall_at_10
value: 49.19358333333333
- type: recall_at_100
value: 73.07583333333334
- type: recall_at_1000
value: 89.81675
- type: recall_at_3
value: 36.54091666666667
- type: recall_at_5
value: 41.919250000000005
- type: map_at_1
value: 23.388
- type: map_at_10
value: 29.408
- type: map_at_100
value: 30.452
- type: map_at_1000
value: 30.546
- type: map_at_3
value: 27.139000000000003
- type: map_at_5
value: 28.402
- type: mrr_at_1
value: 25.46
- type: mrr_at_10
value: 31.966
- type: mrr_at_100
value: 32.879999999999995
- type: mrr_at_1000
value: 32.944
- type: mrr_at_3
value: 29.755
- type: mrr_at_5
value: 30.974
- type: ndcg_at_1
value: 25.46
- type: ndcg_at_10
value: 33.449
- type: ndcg_at_100
value: 38.67
- type: ndcg_at_1000
value: 41.035
- type: ndcg_at_3
value: 29.048000000000002
- type: ndcg_at_5
value: 31.127
- type: precision_at_1
value: 25.46
- type: precision_at_10
value: 5.199
- type: precision_at_100
value: 0.8670000000000001
- type: precision_at_1000
value: 0.11399999999999999
- type: precision_at_3
value: 12.168
- type: precision_at_5
value: 8.62
- type: recall_at_1
value: 23.388
- type: recall_at_10
value: 43.428
- type: recall_at_100
value: 67.245
- type: recall_at_1000
value: 84.75399999999999
- type: recall_at_3
value: 31.416
- type: recall_at_5
value: 36.451
- type: map_at_1
value: 17.136000000000003
- type: map_at_10
value: 24.102999999999998
- type: map_at_100
value: 25.219
- type: map_at_1000
value: 25.344
- type: map_at_3
value: 22.004
- type: map_at_5
value: 23.145
- type: mrr_at_1
value: 20.613
- type: mrr_at_10
value: 27.753
- type: mrr_at_100
value: 28.698
- type: mrr_at_1000
value: 28.776000000000003
- type: mrr_at_3
value: 25.711000000000002
- type: mrr_at_5
value: 26.795
- type: ndcg_at_1
value: 20.613
- type: ndcg_at_10
value: 28.510999999999996
- type: ndcg_at_100
value: 33.924
- type: ndcg_at_1000
value: 36.849
- type: ndcg_at_3
value: 24.664
- type: ndcg_at_5
value: 26.365
- type: precision_at_1
value: 20.613
- type: precision_at_10
value: 5.069
- type: precision_at_100
value: 0.918
- type: precision_at_1000
value: 0.136
- type: precision_at_3
value: 11.574
- type: precision_at_5
value: 8.211
- type: recall_at_1
value: 17.136000000000003
- type: recall_at_10
value: 38.232
- type: recall_at_100
value: 62.571
- type: recall_at_1000
value: 83.23
- type: recall_at_3
value: 27.468999999999998
- type: recall_at_5
value: 31.852999999999998
- type: map_at_1
value: 25.580000000000002
- type: map_at_10
value: 33.449
- type: map_at_100
value: 34.58
- type: map_at_1000
value: 34.692
- type: map_at_3
value: 30.660999999999998
- type: map_at_5
value: 32.425
- type: mrr_at_1
value: 30.037000000000003
- type: mrr_at_10
value: 37.443
- type: mrr_at_100
value: 38.32
- type: mrr_at_1000
value: 38.384
- type: mrr_at_3
value: 34.778999999999996
- type: mrr_at_5
value: 36.458
- type: ndcg_at_1
value: 30.037000000000003
- type: ndcg_at_10
value: 38.46
- type: ndcg_at_100
value: 43.746
- type: ndcg_at_1000
value: 46.28
- type: ndcg_at_3
value: 33.52
- type: ndcg_at_5
value: 36.175000000000004
- type: precision_at_1
value: 30.037000000000003
- type: precision_at_10
value: 6.418
- type: precision_at_100
value: 1.0210000000000001
- type: precision_at_1000
value: 0.136
- type: precision_at_3
value: 15.018999999999998
- type: precision_at_5
value: 10.877
- type: recall_at_1
value: 25.580000000000002
- type: recall_at_10
value: 49.830000000000005
- type: recall_at_100
value: 73.04899999999999
- type: recall_at_1000
value: 90.751
- type: recall_at_3
value: 36.370999999999995
- type: recall_at_5
value: 43.104
- type: map_at_1
value: 24.071
- type: map_at_10
value: 33.384
- type: map_at_100
value: 35.004999999999995
- type: map_at_1000
value: 35.215999999999994
- type: map_at_3
value: 30.459000000000003
- type: map_at_5
value: 31.769
- type: mrr_at_1
value: 28.854000000000003
- type: mrr_at_10
value: 37.512
- type: mrr_at_100
value: 38.567
- type: mrr_at_1000
value: 38.618
- type: mrr_at_3
value: 35.211
- type: mrr_at_5
value: 36.13
- type: ndcg_at_1
value: 28.854000000000003
- type: ndcg_at_10
value: 39.216
- type: ndcg_at_100
value: 45.214
- type: ndcg_at_1000
value: 47.573
- type: ndcg_at_3
value: 34.597
- type: ndcg_at_5
value: 36.063
- type: precision_at_1
value: 28.854000000000003
- type: precision_at_10
value: 7.648000000000001
- type: precision_at_100
value: 1.545
- type: precision_at_1000
value: 0.241
- type: precision_at_3
value: 16.667
- type: precision_at_5
value: 11.818
- type: recall_at_1
value: 24.071
- type: recall_at_10
value: 50.802
- type: recall_at_100
value: 77.453
- type: recall_at_1000
value: 92.304
- type: recall_at_3
value: 36.846000000000004
- type: recall_at_5
value: 41.14
- type: map_at_1
value: 23.395
- type: map_at_10
value: 29.189999999999998
- type: map_at_100
value: 30.226999999999997
- type: map_at_1000
value: 30.337999999999997
- type: map_at_3
value: 27.342
- type: map_at_5
value: 28.116999999999997
- type: mrr_at_1
value: 25.323
- type: mrr_at_10
value: 31.241000000000003
- type: mrr_at_100
value: 32.225
- type: mrr_at_1000
value: 32.304
- type: mrr_at_3
value: 29.452
- type: mrr_at_5
value: 30.209000000000003
- type: ndcg_at_1
value: 25.323
- type: ndcg_at_10
value: 33.024
- type: ndcg_at_100
value: 38.279
- type: ndcg_at_1000
value: 41.026
- type: ndcg_at_3
value: 29.243000000000002
- type: ndcg_at_5
value: 30.564000000000004
- type: precision_at_1
value: 25.323
- type: precision_at_10
value: 4.972
- type: precision_at_100
value: 0.8210000000000001
- type: precision_at_1000
value: 0.116
- type: precision_at_3
value: 12.076
- type: precision_at_5
value: 8.133
- type: recall_at_1
value: 23.395
- type: recall_at_10
value: 42.994
- type: recall_at_100
value: 66.985
- type: recall_at_1000
value: 87.483
- type: recall_at_3
value: 32.505
- type: recall_at_5
value: 35.721000000000004
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER
type: climate-fever
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 8.322000000000001
- type: map_at_10
value: 14.491000000000001
- type: map_at_100
value: 16.066
- type: map_at_1000
value: 16.238
- type: map_at_3
value: 12.235
- type: map_at_5
value: 13.422999999999998
- type: mrr_at_1
value: 19.479
- type: mrr_at_10
value: 29.38
- type: mrr_at_100
value: 30.520999999999997
- type: mrr_at_1000
value: 30.570999999999998
- type: mrr_at_3
value: 26.395000000000003
- type: mrr_at_5
value: 27.982000000000003
- type: ndcg_at_1
value: 19.479
- type: ndcg_at_10
value: 21.215
- type: ndcg_at_100
value: 27.966
- type: ndcg_at_1000
value: 31.324
- type: ndcg_at_3
value: 17.194000000000003
- type: ndcg_at_5
value: 18.593
- type: precision_at_1
value: 19.479
- type: precision_at_10
value: 6.5280000000000005
- type: precision_at_100
value: 1.359
- type: precision_at_1000
value: 0.198
- type: precision_at_3
value: 12.703999999999999
- type: precision_at_5
value: 9.655
- type: recall_at_1
value: 8.322000000000001
- type: recall_at_10
value: 26.165
- type: recall_at_100
value: 49.573
- type: recall_at_1000
value: 68.501
- type: recall_at_3
value: 16.179
- type: recall_at_5
value: 20.175
- task:
type: Retrieval
dataset:
name: MTEB DBPedia
type: dbpedia-entity
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 8.003
- type: map_at_10
value: 16.087
- type: map_at_100
value: 21.363
- type: map_at_1000
value: 22.64
- type: map_at_3
value: 12.171999999999999
- type: map_at_5
value: 13.866
- type: mrr_at_1
value: 61.25000000000001
- type: mrr_at_10
value: 68.626
- type: mrr_at_100
value: 69.134
- type: mrr_at_1000
value: 69.144
- type: mrr_at_3
value: 67.042
- type: mrr_at_5
value: 67.929
- type: ndcg_at_1
value: 49.0
- type: ndcg_at_10
value: 34.132
- type: ndcg_at_100
value: 37.545
- type: ndcg_at_1000
value: 44.544
- type: ndcg_at_3
value: 38.946999999999996
- type: ndcg_at_5
value: 36.317
- type: precision_at_1
value: 61.25000000000001
- type: precision_at_10
value: 26.325
- type: precision_at_100
value: 8.173
- type: precision_at_1000
value: 1.778
- type: precision_at_3
value: 41.667
- type: precision_at_5
value: 34.300000000000004
- type: recall_at_1
value: 8.003
- type: recall_at_10
value: 20.577
- type: recall_at_100
value: 41.884
- type: recall_at_1000
value: 64.36500000000001
- type: recall_at_3
value: 13.602
- type: recall_at_5
value: 16.41
- task:
type: Classification
dataset:
name: MTEB EmotionClassification
type: mteb/emotion
config: default
split: test
revision: 4f58c6b202a23cf9a4da393831edf4f9183cad37
metrics:
- type: accuracy
value: 45.835
- type: f1
value: 41.66455981281837
- task:
type: Retrieval
dataset:
name: MTEB FEVER
type: fever
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 55.717000000000006
- type: map_at_10
value: 66.34100000000001
- type: map_at_100
value: 66.776
- type: map_at_1000
value: 66.794
- type: map_at_3
value: 64.386
- type: map_at_5
value: 65.566
- type: mrr_at_1
value: 60.141
- type: mrr_at_10
value: 70.928
- type: mrr_at_100
value: 71.29299999999999
- type: mrr_at_1000
value: 71.30199999999999
- type: mrr_at_3
value: 69.07900000000001
- type: mrr_at_5
value: 70.244
- type: ndcg_at_1
value: 60.141
- type: ndcg_at_10
value: 71.90100000000001
- type: ndcg_at_100
value: 73.836
- type: ndcg_at_1000
value: 74.214
- type: ndcg_at_3
value: 68.203
- type: ndcg_at_5
value: 70.167
- type: precision_at_1
value: 60.141
- type: precision_at_10
value: 9.268
- type: precision_at_100
value: 1.03
- type: precision_at_1000
value: 0.108
- type: precision_at_3
value: 27.028000000000002
- type: precision_at_5
value: 17.342
- type: recall_at_1
value: 55.717000000000006
- type: recall_at_10
value: 84.66799999999999
- type: recall_at_100
value: 93.28
- type: recall_at_1000
value: 95.887
- type: recall_at_3
value: 74.541
- type: recall_at_5
value: 79.389
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018
type: fiqa
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 17.744
- type: map_at_10
value: 29.554000000000002
- type: map_at_100
value: 31.180000000000003
- type: map_at_1000
value: 31.372
- type: map_at_3
value: 25.6
- type: map_at_5
value: 27.642
- type: mrr_at_1
value: 35.802
- type: mrr_at_10
value: 44.812999999999995
- type: mrr_at_100
value: 45.56
- type: mrr_at_1000
value: 45.606
- type: mrr_at_3
value: 42.181000000000004
- type: mrr_at_5
value: 43.516
- type: ndcg_at_1
value: 35.802
- type: ndcg_at_10
value: 37.269999999999996
- type: ndcg_at_100
value: 43.575
- type: ndcg_at_1000
value: 46.916000000000004
- type: ndcg_at_3
value: 33.511
- type: ndcg_at_5
value: 34.504000000000005
- type: precision_at_1
value: 35.802
- type: precision_at_10
value: 10.448
- type: precision_at_100
value: 1.7129999999999999
- type: precision_at_1000
value: 0.231
- type: precision_at_3
value: 22.531000000000002
- type: precision_at_5
value: 16.512
- type: recall_at_1
value: 17.744
- type: recall_at_10
value: 44.616
- type: recall_at_100
value: 68.51899999999999
- type: recall_at_1000
value: 88.495
- type: recall_at_3
value: 30.235
- type: recall_at_5
value: 35.821999999999996
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA
type: hotpotqa
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 33.315
- type: map_at_10
value: 45.932
- type: map_at_100
value: 46.708
- type: map_at_1000
value: 46.778999999999996
- type: map_at_3
value: 43.472
- type: map_at_5
value: 45.022
- type: mrr_at_1
value: 66.631
- type: mrr_at_10
value: 73.083
- type: mrr_at_100
value: 73.405
- type: mrr_at_1000
value: 73.421
- type: mrr_at_3
value: 71.756
- type: mrr_at_5
value: 72.616
- type: ndcg_at_1
value: 66.631
- type: ndcg_at_10
value: 54.949000000000005
- type: ndcg_at_100
value: 57.965
- type: ndcg_at_1000
value: 59.467000000000006
- type: ndcg_at_3
value: 51.086
- type: ndcg_at_5
value: 53.272
- type: precision_at_1
value: 66.631
- type: precision_at_10
value: 11.178
- type: precision_at_100
value: 1.3559999999999999
- type: precision_at_1000
value: 0.156
- type: precision_at_3
value: 31.582
- type: precision_at_5
value: 20.678
- type: recall_at_1
value: 33.315
- type: recall_at_10
value: 55.888000000000005
- type: recall_at_100
value: 67.812
- type: recall_at_1000
value: 77.839
- type: recall_at_3
value: 47.373
- type: recall_at_5
value: 51.695
- task:
type: Classification
dataset:
name: MTEB ImdbClassification
type: mteb/imdb
config: default
split: test
revision: 3d86128a09e091d6018b6d26cad27f2739fc2db7
metrics:
- type: accuracy
value: 66.424
- type: ap
value: 61.132235499939256
- type: f1
value: 66.07094958225315
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO
type: msmarco
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 21.575
- type: map_at_10
value: 33.509
- type: map_at_100
value: 34.725
- type: map_at_1000
value: 34.775
- type: map_at_3
value: 29.673
- type: map_at_5
value: 31.805
- type: mrr_at_1
value: 22.235
- type: mrr_at_10
value: 34.1
- type: mrr_at_100
value: 35.254999999999995
- type: mrr_at_1000
value: 35.299
- type: mrr_at_3
value: 30.334
- type: mrr_at_5
value: 32.419
- type: ndcg_at_1
value: 22.235
- type: ndcg_at_10
value: 40.341
- type: ndcg_at_100
value: 46.161
- type: ndcg_at_1000
value: 47.400999999999996
- type: ndcg_at_3
value: 32.482
- type: ndcg_at_5
value: 36.269
- type: precision_at_1
value: 22.235
- type: precision_at_10
value: 6.422999999999999
- type: precision_at_100
value: 0.9329999999999999
- type: precision_at_1000
value: 0.104
- type: precision_at_3
value: 13.835
- type: precision_at_5
value: 10.226
- type: recall_at_1
value: 21.575
- type: recall_at_10
value: 61.448
- type: recall_at_100
value: 88.289
- type: recall_at_1000
value: 97.76899999999999
- type: recall_at_3
value: 39.971000000000004
- type: recall_at_5
value: 49.053000000000004
- task:
type: Classification
dataset:
name: MTEB MTOPDomainClassification (en)
type: mteb/mtop_domain
config: en
split: test
revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf
metrics:
- type: accuracy
value: 92.83401732786137
- type: f1
value: 92.47678691291068
- task:
type: Classification
dataset:
name: MTEB MTOPIntentClassification (en)
type: mteb/mtop_intent
config: en
split: test
revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba
metrics:
- type: accuracy
value: 76.08983128134975
- type: f1
value: 59.782936393820904
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (en)
type: mteb/amazon_massive_intent
config: en
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 72.73032952252858
- type: f1
value: 70.72684765888265
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (en)
type: mteb/amazon_massive_scenario
config: en
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 77.08473436449226
- type: f1
value: 77.31457411257054
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringP2P
type: mteb/medrxiv-clustering-p2p
config: default
split: test
revision: e7a26af6f3ae46b30dde8737f02c07b1505bcc73
metrics:
- type: v_measure
value: 30.11980959210532
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringS2S
type: mteb/medrxiv-clustering-s2s
config: default
split: test
revision: 35191c8c0dca72d8ff3efcd72aa802307d469663
metrics:
- type: v_measure
value: 25.2587629106119
- task:
type: Reranking
dataset:
name: MTEB MindSmallReranking
type: mteb/mind_small
config: default
split: test
revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69
metrics:
- type: map
value: 31.48268319779204
- type: mrr
value: 32.501885728964304
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus
type: nfcorpus
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 5.284
- type: map_at_10
value: 11.509
- type: map_at_100
value: 14.624
- type: map_at_1000
value: 16.035
- type: map_at_3
value: 8.347999999999999
- type: map_at_5
value: 9.919
- type: mrr_at_1
value: 43.344
- type: mrr_at_10
value: 52.303999999999995
- type: mrr_at_100
value: 52.994
- type: mrr_at_1000
value: 53.032999999999994
- type: mrr_at_3
value: 50.361
- type: mrr_at_5
value: 51.754
- type: ndcg_at_1
value: 41.176
- type: ndcg_at_10
value: 32.244
- type: ndcg_at_100
value: 29.916999999999998
- type: ndcg_at_1000
value: 38.753
- type: ndcg_at_3
value: 36.856
- type: ndcg_at_5
value: 35.394999999999996
- type: precision_at_1
value: 43.034
- type: precision_at_10
value: 24.118000000000002
- type: precision_at_100
value: 7.926
- type: precision_at_1000
value: 2.045
- type: precision_at_3
value: 34.675
- type: precision_at_5
value: 31.146
- type: recall_at_1
value: 5.284
- type: recall_at_10
value: 15.457
- type: recall_at_100
value: 30.914
- type: recall_at_1000
value: 63.788999999999994
- type: recall_at_3
value: 9.596
- type: recall_at_5
value: 12.391
- task:
type: Retrieval
dataset:
name: MTEB NQ
type: nq
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 29.537999999999997
- type: map_at_10
value: 43.99
- type: map_at_100
value: 45.003
- type: map_at_1000
value: 45.04
- type: map_at_3
value: 39.814
- type: map_at_5
value: 42.166
- type: mrr_at_1
value: 33.256
- type: mrr_at_10
value: 46.487
- type: mrr_at_100
value: 47.264
- type: mrr_at_1000
value: 47.29
- type: mrr_at_3
value: 43.091
- type: mrr_at_5
value: 45.013999999999996
- type: ndcg_at_1
value: 33.256
- type: ndcg_at_10
value: 51.403
- type: ndcg_at_100
value: 55.706999999999994
- type: ndcg_at_1000
value: 56.586000000000006
- type: ndcg_at_3
value: 43.559
- type: ndcg_at_5
value: 47.426
- type: precision_at_1
value: 33.256
- type: precision_at_10
value: 8.540000000000001
- type: precision_at_100
value: 1.093
- type: precision_at_1000
value: 0.11800000000000001
- type: precision_at_3
value: 19.834
- type: precision_at_5
value: 14.143
- type: recall_at_1
value: 29.537999999999997
- type: recall_at_10
value: 71.5
- type: recall_at_100
value: 90.25
- type: recall_at_1000
value: 96.82600000000001
- type: recall_at_3
value: 51.108
- type: recall_at_5
value: 60.006
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval
type: quora
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 70.526
- type: map_at_10
value: 84.342
- type: map_at_100
value: 84.985
- type: map_at_1000
value: 85.003
- type: map_at_3
value: 81.472
- type: map_at_5
value: 83.292
- type: mrr_at_1
value: 81.17
- type: mrr_at_10
value: 87.33999999999999
- type: mrr_at_100
value: 87.445
- type: mrr_at_1000
value: 87.446
- type: mrr_at_3
value: 86.387
- type: mrr_at_5
value: 87.042
- type: ndcg_at_1
value: 81.19
- type: ndcg_at_10
value: 88.088
- type: ndcg_at_100
value: 89.35
- type: ndcg_at_1000
value: 89.462
- type: ndcg_at_3
value: 85.319
- type: ndcg_at_5
value: 86.858
- type: precision_at_1
value: 81.19
- type: precision_at_10
value: 13.33
- type: precision_at_100
value: 1.528
- type: precision_at_1000
value: 0.157
- type: precision_at_3
value: 37.31
- type: precision_at_5
value: 24.512
- type: recall_at_1
value: 70.526
- type: recall_at_10
value: 95.166
- type: recall_at_100
value: 99.479
- type: recall_at_1000
value: 99.984
- type: recall_at_3
value: 87.124
- type: recall_at_5
value: 91.53
- task:
type: Clustering
dataset:
name: MTEB RedditClustering
type: mteb/reddit-clustering
config: default
split: test
revision: 24640382cdbf8abc73003fb0fa6d111a705499eb
metrics:
- type: v_measure
value: 45.049073872893494
- task:
type: Clustering
dataset:
name: MTEB RedditClusteringP2P
type: mteb/reddit-clustering-p2p
config: default
split: test
revision: 282350215ef01743dc01b456c7f5241fa8937f16
metrics:
- type: v_measure
value: 55.13810914528368
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS
type: scidocs
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 4.593
- type: map_at_10
value: 10.907
- type: map_at_100
value: 12.888
- type: map_at_1000
value: 13.167000000000002
- type: map_at_3
value: 7.936
- type: map_at_5
value: 9.31
- type: mrr_at_1
value: 22.7
- type: mrr_at_10
value: 32.509
- type: mrr_at_100
value: 33.69
- type: mrr_at_1000
value: 33.747
- type: mrr_at_3
value: 29.599999999999998
- type: mrr_at_5
value: 31.155
- type: ndcg_at_1
value: 22.7
- type: ndcg_at_10
value: 18.445
- type: ndcg_at_100
value: 26.241999999999997
- type: ndcg_at_1000
value: 31.409
- type: ndcg_at_3
value: 17.864
- type: ndcg_at_5
value: 15.232999999999999
- type: precision_at_1
value: 22.7
- type: precision_at_10
value: 9.43
- type: precision_at_100
value: 2.061
- type: precision_at_1000
value: 0.331
- type: precision_at_3
value: 16.467000000000002
- type: precision_at_5
value: 13.08
- type: recall_at_1
value: 4.593
- type: recall_at_10
value: 19.115
- type: recall_at_100
value: 41.82
- type: recall_at_1000
value: 67.167
- type: recall_at_3
value: 9.983
- type: recall_at_5
value: 13.218
- task:
type: STS
dataset:
name: MTEB SICK-R
type: mteb/sickr-sts
config: default
split: test
revision: a6ea5a8cab320b040a23452cc28066d9beae2cee
metrics:
- type: cos_sim_pearson
value: 82.94432059816452
- type: cos_sim_spearman
value: 79.19993315048852
- type: euclidean_pearson
value: 72.43261099671753
- type: euclidean_spearman
value: 71.51531114998619
- type: manhattan_pearson
value: 71.83604124130447
- type: manhattan_spearman
value: 71.24460392842295
- task:
type: STS
dataset:
name: MTEB STS12
type: mteb/sts12-sts
config: default
split: test
revision: a0d554a64d88156834ff5ae9920b964011b16384
metrics:
- type: cos_sim_pearson
value: 84.25401068481673
- type: cos_sim_spearman
value: 74.5249604699309
- type: euclidean_pearson
value: 71.1324859629043
- type: euclidean_spearman
value: 58.77041705276752
- type: manhattan_pearson
value: 71.01471521586141
- type: manhattan_spearman
value: 58.69949381017865
- task:
type: STS
dataset:
name: MTEB STS13
type: mteb/sts13-sts
config: default
split: test
revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
metrics:
- type: cos_sim_pearson
value: 82.85731544223766
- type: cos_sim_spearman
value: 83.15607264736185
- type: euclidean_pearson
value: 75.8803249521361
- type: euclidean_spearman
value: 76.4862168799065
- type: manhattan_pearson
value: 75.80451454386811
- type: manhattan_spearman
value: 76.35986831074699
- task:
type: STS
dataset:
name: MTEB STS14
type: mteb/sts14-sts
config: default
split: test
revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
metrics:
- type: cos_sim_pearson
value: 82.40669043798857
- type: cos_sim_spearman
value: 78.08686090667834
- type: euclidean_pearson
value: 74.48574712193803
- type: euclidean_spearman
value: 70.79423012045118
- type: manhattan_pearson
value: 74.39099211477354
- type: manhattan_spearman
value: 70.73135427277684
- task:
type: STS
dataset:
name: MTEB STS15
type: mteb/sts15-sts
config: default
split: test
revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
metrics:
- type: cos_sim_pearson
value: 86.03027014209859
- type: cos_sim_spearman
value: 86.91082847840946
- type: euclidean_pearson
value: 69.13187603971996
- type: euclidean_spearman
value: 70.0370035340552
- type: manhattan_pearson
value: 69.2586635812031
- type: manhattan_spearman
value: 70.18638387118486
- task:
type: STS
dataset:
name: MTEB STS16
type: mteb/sts16-sts
config: default
split: test
revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
metrics:
- type: cos_sim_pearson
value: 82.41190748361883
- type: cos_sim_spearman
value: 83.64850851235231
- type: euclidean_pearson
value: 71.60523243575282
- type: euclidean_spearman
value: 72.26134033805099
- type: manhattan_pearson
value: 71.50771482066683
- type: manhattan_spearman
value: 72.13707967973161
- task:
type: STS
dataset:
name: MTEB STS17 (en-en)
type: mteb/sts17-crosslingual-sts
config: en-en
split: test
revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d
metrics:
- type: cos_sim_pearson
value: 90.42838477648627
- type: cos_sim_spearman
value: 90.15798155439076
- type: euclidean_pearson
value: 77.09619972244516
- type: euclidean_spearman
value: 75.5953488548861
- type: manhattan_pearson
value: 77.36892406451771
- type: manhattan_spearman
value: 75.76625156149356
- task:
type: STS
dataset:
name: MTEB STS22 (en)
type: mteb/sts22-crosslingual-sts
config: en
split: test
revision: 6d1ba47164174a496b7fa5d3569dae26a6813b80
metrics:
- type: cos_sim_pearson
value: 65.76151154879307
- type: cos_sim_spearman
value: 64.8846800918359
- type: euclidean_pearson
value: 50.23302700257155
- type: euclidean_spearman
value: 58.89455187289583
- type: manhattan_pearson
value: 50.05498582284945
- type: manhattan_spearman
value: 58.75893793871576
- task:
type: STS
dataset:
name: MTEB STSBenchmark
type: mteb/stsbenchmark-sts
config: default
split: test
revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831
metrics:
- type: cos_sim_pearson
value: 84.72381109169437
- type: cos_sim_spearman
value: 84.59820928231167
- type: euclidean_pearson
value: 74.85450857429493
- type: euclidean_spearman
value: 73.83634052565915
- type: manhattan_pearson
value: 74.97349743979106
- type: manhattan_spearman
value: 73.9636470375881
- task:
type: Reranking
dataset:
name: MTEB SciDocsRR
type: mteb/scidocs-reranking
config: default
split: test
revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab
metrics:
- type: map
value: 80.96736259172798
- type: mrr
value: 94.48378781712114
- task:
type: Retrieval
dataset:
name: MTEB SciFact
type: scifact
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 46.344
- type: map_at_10
value: 54.962
- type: map_at_100
value: 55.772
- type: map_at_1000
value: 55.81700000000001
- type: map_at_3
value: 51.832
- type: map_at_5
value: 53.718999999999994
- type: mrr_at_1
value: 49.0
- type: mrr_at_10
value: 56.721
- type: mrr_at_100
value: 57.287
- type: mrr_at_1000
value: 57.330000000000005
- type: mrr_at_3
value: 54.056000000000004
- type: mrr_at_5
value: 55.822
- type: ndcg_at_1
value: 49.0
- type: ndcg_at_10
value: 59.757000000000005
- type: ndcg_at_100
value: 63.149
- type: ndcg_at_1000
value: 64.43100000000001
- type: ndcg_at_3
value: 54.105000000000004
- type: ndcg_at_5
value: 57.196999999999996
- type: precision_at_1
value: 49.0
- type: precision_at_10
value: 8.200000000000001
- type: precision_at_100
value: 1.0070000000000001
- type: precision_at_1000
value: 0.11100000000000002
- type: precision_at_3
value: 20.889
- type: precision_at_5
value: 14.399999999999999
- type: recall_at_1
value: 46.344
- type: recall_at_10
value: 72.722
- type: recall_at_100
value: 88.167
- type: recall_at_1000
value: 98.333
- type: recall_at_3
value: 57.994
- type: recall_at_5
value: 65.506
- task:
type: PairClassification
dataset:
name: MTEB SprintDuplicateQuestions
type: mteb/sprintduplicatequestions-pairclassification
config: default
split: test
revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
metrics:
- type: cos_sim_accuracy
value: 99.83366336633664
- type: cos_sim_ap
value: 96.09329747251944
- type: cos_sim_f1
value: 91.66255550074001
- type: cos_sim_precision
value: 90.45764362220059
- type: cos_sim_recall
value: 92.9
- type: dot_accuracy
value: 99.32871287128712
- type: dot_ap
value: 63.95436644147969
- type: dot_f1
value: 60.61814556331008
- type: dot_precision
value: 60.437375745526836
- type: dot_recall
value: 60.8
- type: euclidean_accuracy
value: 99.66534653465347
- type: euclidean_ap
value: 85.85143979761818
- type: euclidean_f1
value: 81.57033805888769
- type: euclidean_precision
value: 89.68824940047962
- type: euclidean_recall
value: 74.8
- type: manhattan_accuracy
value: 99.65742574257426
- type: manhattan_ap
value: 85.55693926348405
- type: manhattan_f1
value: 81.13804004214963
- type: manhattan_precision
value: 85.74610244988864
- type: manhattan_recall
value: 77.0
- type: max_accuracy
value: 99.83366336633664
- type: max_ap
value: 96.09329747251944
- type: max_f1
value: 91.66255550074001
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClustering
type: mteb/stackexchange-clustering
config: default
split: test
revision: 6cbc1f7b2bc0622f2e39d2c77fa502909748c259
metrics:
- type: v_measure
value: 45.23573510003245
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClusteringP2P
type: mteb/stackexchange-clustering-p2p
config: default
split: test
revision: 815ca46b2622cec33ccafc3735d572c266efdb44
metrics:
- type: v_measure
value: 33.37478638401161
- task:
type: Reranking
dataset:
name: MTEB StackOverflowDupQuestions
type: mteb/stackoverflowdupquestions-reranking
config: default
split: test
revision: e185fbe320c72810689fc5848eb6114e1ef5ec69
metrics:
- type: map
value: 50.375920467392476
- type: mrr
value: 51.17302223919871
- task:
type: Summarization
dataset:
name: MTEB SummEval
type: mteb/summeval
config: default
split: test
revision: cda12ad7615edc362dbf25a00fdd61d3b1eaf93c
metrics:
- type: cos_sim_pearson
value: 29.768864092288343
- type: cos_sim_spearman
value: 29.854278347043266
- type: dot_pearson
value: 20.51281723837505
- type: dot_spearman
value: 21.799102540913665
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID
type: trec-covid
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 0.2
- type: map_at_10
value: 1.202
- type: map_at_100
value: 6.729
- type: map_at_1000
value: 15.928
- type: map_at_3
value: 0.492
- type: map_at_5
value: 0.712
- type: mrr_at_1
value: 76.0
- type: mrr_at_10
value: 84.75
- type: mrr_at_100
value: 84.75
- type: mrr_at_1000
value: 84.75
- type: mrr_at_3
value: 83.0
- type: mrr_at_5
value: 84.5
- type: ndcg_at_1
value: 71.0
- type: ndcg_at_10
value: 57.253
- type: ndcg_at_100
value: 44.383
- type: ndcg_at_1000
value: 38.666
- type: ndcg_at_3
value: 64.324
- type: ndcg_at_5
value: 60.791
- type: precision_at_1
value: 76.0
- type: precision_at_10
value: 59.599999999999994
- type: precision_at_100
value: 45.440000000000005
- type: precision_at_1000
value: 17.458000000000002
- type: precision_at_3
value: 69.333
- type: precision_at_5
value: 63.2
- type: recall_at_1
value: 0.2
- type: recall_at_10
value: 1.4949999999999999
- type: recall_at_100
value: 10.266
- type: recall_at_1000
value: 35.853
- type: recall_at_3
value: 0.5349999999999999
- type: recall_at_5
value: 0.8109999999999999
- task:
type: Retrieval
dataset:
name: MTEB Touche2020
type: webis-touche2020
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 2.0140000000000002
- type: map_at_10
value: 8.474
- type: map_at_100
value: 14.058000000000002
- type: map_at_1000
value: 15.381
- type: map_at_3
value: 4.508
- type: map_at_5
value: 5.87
- type: mrr_at_1
value: 22.448999999999998
- type: mrr_at_10
value: 37.242
- type: mrr_at_100
value: 38.291
- type: mrr_at_1000
value: 38.311
- type: mrr_at_3
value: 32.312999999999995
- type: mrr_at_5
value: 34.762
- type: ndcg_at_1
value: 20.408
- type: ndcg_at_10
value: 20.729
- type: ndcg_at_100
value: 33.064
- type: ndcg_at_1000
value: 44.324999999999996
- type: ndcg_at_3
value: 21.251
- type: ndcg_at_5
value: 20.28
- type: precision_at_1
value: 22.448999999999998
- type: precision_at_10
value: 18.98
- type: precision_at_100
value: 7.224
- type: precision_at_1000
value: 1.471
- type: precision_at_3
value: 22.448999999999998
- type: precision_at_5
value: 20.816000000000003
- type: recall_at_1
value: 2.0140000000000002
- type: recall_at_10
value: 13.96
- type: recall_at_100
value: 44.187
- type: recall_at_1000
value: 79.328
- type: recall_at_3
value: 5.345
- type: recall_at_5
value: 7.979
- task:
type: Classification
dataset:
name: MTEB ToxicConversationsClassification
type: mteb/toxic_conversations_50k
config: default
split: test
revision: d7c0de2777da35d6aae2200a62c6e0e5af397c4c
metrics:
- type: accuracy
value: 69.1312
- type: ap
value: 12.606776505497608
- type: f1
value: 52.4112415600534
- task:
type: Classification
dataset:
name: MTEB TweetSentimentExtractionClassification
type: mteb/tweet_sentiment_extraction
config: default
split: test
revision: d604517c81ca91fe16a244d1248fc021f9ecee7a
metrics:
- type: accuracy
value: 58.16072439162422
- type: f1
value: 58.29152785435414
- task:
type: Clustering
dataset:
name: MTEB TwentyNewsgroupsClustering
type: mteb/twentynewsgroups-clustering
config: default
split: test
revision: 6125ec4e24fa026cec8a478383ee943acfbd5449
metrics:
- type: v_measure
value: 40.421119289825924
- task:
type: PairClassification
dataset:
name: MTEB TwitterSemEval2015
type: mteb/twittersemeval2015-pairclassification
config: default
split: test
revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1
metrics:
- type: cos_sim_accuracy
value: 85.48012159504083
- type: cos_sim_ap
value: 72.31974877212102
- type: cos_sim_f1
value: 67.96846573681019
- type: cos_sim_precision
value: 62.89562289562289
- type: cos_sim_recall
value: 73.93139841688654
- type: dot_accuracy
value: 78.52416999463551
- type: dot_ap
value: 43.65271285411479
- type: dot_f1
value: 46.94641449960599
- type: dot_precision
value: 37.456774599182644
- type: dot_recall
value: 62.875989445910285
- type: euclidean_accuracy
value: 83.90057817249806
- type: euclidean_ap
value: 65.96278727778665
- type: euclidean_f1
value: 63.35733232284957
- type: euclidean_precision
value: 60.770535497940394
- type: euclidean_recall
value: 66.17414248021109
- type: manhattan_accuracy
value: 83.96614412588663
- type: manhattan_ap
value: 66.03670273156699
- type: manhattan_f1
value: 63.49128406579917
- type: manhattan_precision
value: 59.366391184573
- type: manhattan_recall
value: 68.23218997361478
- type: max_accuracy
value: 85.48012159504083
- type: max_ap
value: 72.31974877212102
- type: max_f1
value: 67.96846573681019
- task:
type: PairClassification
dataset:
name: MTEB TwitterURLCorpus
type: mteb/twitterurlcorpus-pairclassification
config: default
split: test
revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf
metrics:
- type: cos_sim_accuracy
value: 88.97038848139093
- type: cos_sim_ap
value: 85.982764495556
- type: cos_sim_f1
value: 78.73283281450284
- type: cos_sim_precision
value: 75.07857791436754
- type: cos_sim_recall
value: 82.7610101632276
- type: dot_accuracy
value: 83.21108394458028
- type: dot_ap
value: 70.97956937273386
- type: dot_f1
value: 66.53083038279111
- type: dot_precision
value: 58.7551622418879
- type: dot_recall
value: 76.67847243609486
- type: euclidean_accuracy
value: 84.31520937633407
- type: euclidean_ap
value: 74.67323411319909
- type: euclidean_f1
value: 67.21935410935676
- type: euclidean_precision
value: 65.82773636430733
- type: euclidean_recall
value: 68.67108099784416
- type: manhattan_accuracy
value: 84.35013777312066
- type: manhattan_ap
value: 74.66508905354597
- type: manhattan_f1
value: 67.28264162375038
- type: manhattan_precision
value: 66.19970193740686
- type: manhattan_recall
value: 68.40160147828766
- type: max_accuracy
value: 88.97038848139093
- type: max_ap
value: 85.982764495556
- type: max_f1
value: 78.73283281450284
---
<br><br>
<p align="center">
<img src="https://huggingface.co/datasets/jinaai/documentation-images/resolve/main/logo.webp" alt="Jina AI: Your Search Foundation, Supercharged!" width="150px">
</p>
<p align="center">
<b>The text embedding set trained by <a href="https://jina.ai/"><b>Jina AI</b></a></b>
</p>
## Intented Usage & Model Info
`jina-embedding-l-en-v1` is a language model that has been trained using Jina AI's Linnaeus-Clean dataset.
This dataset consists of 380 million pairs of sentences, which include both query-document pairs.
These pairs were obtained from various domains and were carefully selected through a thorough cleaning process.
The Linnaeus-Full dataset, from which the Linnaeus-Clean dataset is derived, originally contained 1.6 billion sentence pairs.
The model has a range of use cases, including information retrieval, semantic textual similarity, text reranking, and more.
With a size of 330 million parameters,
the model enables single-gpu inference while delivering better performance than our small and base model.
Additionally, we provide the following options:
- [`jina-embedding-t-en-v1`](https://huggingface.co/jinaai/jina-embedding-t-en-v1): 14 million parameters.
- [`jina-embedding-s-en-v1`](https://huggingface.co/jinaai/jina-embedding-s-en-v1): 35 million parameters
- [`jina-embedding-b-en-v1`](https://huggingface.co/jinaai/jina-embedding-b-en-v1): 110 million parameters.
- [`jina-embedding-l-en-v1`](https://huggingface.co/jinaai/jina-embedding-l-en-v1): 330 million parameters **(you are here)**.
- `jina-embedding-1b-en-v1`: 1.2 billion parameters, 10 times bert-base (soon).
- `jina-embedding-6b-en-v1`: 6 billion parameters, 30 times bert-base (soon).
## Data & Parameters
Please checkout our [technical blog](https://arxiv.org/abs/2307.11224).
## Metrics
We compared the model against `all-minilm-l6-v2`/`all-mpnet-base-v2` from sbert and `text-embeddings-ada-002` from OpenAI:
|Name|param |dimension|
|------------------------------|-----|------|
|all-minilm-l6-v2|23m |384|
|all-mpnet-base-v2 |110m |768|
|ada-embedding-002|Unknown/OpenAI API |1536|
|jina-embedding-t-en-v1|14m |312|
|jina-embedding-s-en-v1|35m |512|
|jina-embedding-b-en-v1|110m |768|
|jina-embedding-l-en-v1|330m |1024|
|Name|STS12|STS13|STS14|STS15|STS16|STS17|TRECOVID|Quora|SciFact|
|------------------------------|-----|-----|-----|-----|-----|-----|--------|-----|-----|
|all-minilm-l6-v2|0.724|0.806|0.756|0.854|0.79 |0.876|0.473 |0.876|0.645 |
|all-mpnet-base-v2|0.726|**0.835**|0.78 |0.857|0.8 |**0.906**|0.513 |0.875|0.656 |
|ada-embedding-002|0.698|0.833|0.761|0.861|**0.86** |0.903|**0.685** |0.876|**0.726** |
|jina-embedding-t-en-v1|0.717|0.773|0.731|0.829|0.777|0.860|0.482 |0.840|0.522 |
|jina-embedding-s-en-v1|0.743|0.786|0.738|0.837|0.80|0.875|0.523 |0.857|0.524 |
|jina-embedding-b-en-v1|**0.751**|0.809|0.761|0.856|0.812|0.890|0.606 |0.876|0.594 |
|jina-embedding-l-en-v1|0.745|0.832|**0.781**|**0.869**|0.837|0.902|0.573 |**0.881**|0.598 |
## Usage
Use with Jina AI Finetuner
```python
!pip install finetuner
import finetuner
model = finetuner.build_model('jinaai/jina-embedding-l-en-v1')
embeddings = finetuner.encode(
model=model,
data=['how is the weather today', 'What is the current weather like today?']
)
print(finetuner.cos_sim(embeddings[0], embeddings[1]))
```
Use with sentence-transformers:
```python
from sentence_transformers import SentenceTransformer
from sentence_transformers.util import cos_sim
sentences = ['how is the weather today', 'What is the current weather like today?']
model = SentenceTransformer('jinaai/jina-embedding-b-en-v1')
embeddings = model.encode(sentences)
print(cos_sim(embeddings[0], embeddings[1]))
```
## Fine-tuning
Please consider [Finetuner](https://github.com/jina-ai/finetuner).
## Plans
1. The development of `jina-embedding-s-en-v2` is currently underway with two main objectives: improving performance and increasing the maximum sequence length.
2. We are currently working on a bilingual embedding model that combines English and X language. The upcoming model will be called `jina-embedding-s/b/l-de-v1`.
## Contact
Join our [Discord community](https://discord.jina.ai) and chat with other community members about ideas.
## Citation
If you find Jina Embeddings useful in your research, please cite the following paper:
``` latex
@misc{günther2023jina,
title={Jina Embeddings: A Novel Set of High-Performance Sentence Embedding Models},
author={Michael Günther and Louis Milliken and Jonathan Geuter and Georgios Mastrapas and Bo Wang and Han Xiao},
year={2023},
eprint={2307.11224},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` | [
"BIOSSES",
"LINNAEUS",
"SCIFACT"
] |
Model-SafeTensors/Llama3-OpenBioLLM-70B | Model-SafeTensors | null | [
"pytorch",
"safetensors",
"llama",
"llama-3",
"Mixtral",
"instruct",
"finetune",
"chatml",
"DPO",
"RLHF",
"gpt4",
"distillation",
"heathcare",
"medical",
"clinical",
"med",
"lifescience",
"Pharmaceutical",
"Pharma",
"en",
"arxiv:2305.18290",
"arxiv:2303.13375",
"arxiv:2212.13138",
"arxiv:2305.09617",
"arxiv:2402.07023",
"base_model:meta-llama/Meta-Llama-3-70B-Instruct",
"base_model:finetune:meta-llama/Meta-Llama-3-70B-Instruct",
"license:llama3",
"region:us"
] | "2024-12-24T21:17:35Z" | 2024-12-24T22:47:05+00:00 | 561 | 0 | ---
base_model: meta-llama/Meta-Llama-3-70B-Instruct
language:
- en
license: llama3
tags:
- llama-3
- llama
- Mixtral
- instruct
- finetune
- chatml
- DPO
- RLHF
- gpt4
- distillation
- heathcare
- medical
- clinical
- med
- lifescience
- Pharmaceutical
- Pharma
widget:
- example_title: OpenBioLLM-70B
messages:
- role: system
content: You are an expert and experienced from the healthcare and biomedical
domain with extensive medical knowledge and practical experience.
- role: user
content: How long does it take for newborn jaundice to go away?
output:
text: 'Newborn jaundice, also known as neonatal jaundice, is a common condition
in newborns where the yellowing of the skin and eyes occurs due to an elevated
level of bilirubin in the blood. Bilirubin is a yellow pigment that forms when
red blood cells break down. In most cases, newborn jaundice resolves on its
own without any specific treatment.
The duration of newborn jaundice can vary depending on several factors such
as the underlying cause, gestational age at birth, and individual variations
in bilirubin metabolism. Here are some general guidelines:
1. Physiological jaundice: This is the most common type of newborn jaundice
and usually appears within 24-72 hours after birth. It tends to peak between
the second and fifth day of life and gradually improves over the next week or
two. By the time the baby is one week old, the jaundice should have mostly resolved.
2. Breast milk jaundice: This type of jaundice occurs in breastfed babies and
may appear later than physiological jaundice, typically between the fifth and
fourteenth day of life. It tends to persist for a longer duration but usually
resolves within six weeks after birth. 3. Pathological jaundice: This type of
jaundice is less common and occurs due to an underlying medical condition that
affects bilirubin metabolism or liver function. The duration of pathological
jaundice depends on the specific cause and may require treatment.
It''s important for parents to monitor their newborn''s jaundice closely and
seek medical advice if the jaundice progresses rapidly, becomes severe, or is
accompanied by other symptoms such as poor feeding, lethargy, or excessive sleepiness.
In these cases, further evaluation and management may be necessary. Remember
that each baby is unique, and the timing of jaundice resolution can vary. If
you have concerns about your newborn''s jaundice, it''s always best to consult
with a healthcare professional for personalized advice and guidance.'
model-index:
- name: OpenBioLLM-70B
results: []
---
<div align="center">
<img width="260px" src="https://cdn-uploads.huggingface.co/production/uploads/5f3fe13d79c1ba4c353d0c19/BrQCb95lmEIFz79QAmoNA.png"></div>
<div align="center">
<h1>Advancing Open-source Large Language Models in Medical Domain</h1>
</div>
<p align="center" style="margin-top: 0px;">
<a href="https://colab.research.google.com/drive/1F5oV20InEYeAJGmBwYF9NM_QhLmjBkKJ?usp=sharing">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="OpenChat Logo" style="width:20px; vertical-align: middle; display: inline-block; margin-right: 5px; margin-left: 10px; margin-top: 0px; margin-bottom: 0px;"/>
<span class="link-text" style=" margin-right: 5px;">Online Demo</span>
</a> |
<a href="https://github.com/openlifescience-ai">
<img src="https://github.githubassets.com/assets/GitHub-Mark-ea2971cee799.png" alt="GitHub Logo" style="width:20px; vertical-align: middle; display: inline-block; margin-right: 5px; margin-left: 5px; margin-top: 0px; margin-bottom: 0px;"/>
<span class="link-text" style=" margin-right: 5px;">GitHub</span>
</a> |
<a href="#">
<img src="https://github.com/alpayariyak/openchat/blob/master/assets/arxiv-logomark-small-square-border.png?raw=true" alt="ArXiv Logo" style="width:20px; vertical-align: middle; display: inline-block; margin-right: 5px; margin-left: 5px; margin-top: 0px; margin-bottom: 0px;"/>
<span class="link-text" style="margin-right: 5px;">Paper</span>
</a> |
<a href="https://discord.gg/A5Fjf5zC69">
<img src="https://cloud.githubusercontent.com/assets/6291467/26705903/96c2d66e-477c-11e7-9f4e-f3c0efe96c9a.png" alt="Discord Logo" style="width:20px; vertical-align: middle; display: inline-block; margin-right: 5px; margin-left: 5px; margin-top: 0px; margin-bottom: 0px;"/>
<span class="link-text">Discord</span>
</a>
</p>
Introducing OpenBioLLM-70B: A State-of-the-Art Open Source Biomedical Large Language Model
OpenBioLLM-70B is an advanced open source language model designed specifically for the biomedical domain. Developed by Saama AI Labs, this model leverages cutting-edge techniques to achieve state-of-the-art performance on a wide range of biomedical tasks.
🏥 **Biomedical Specialization**: OpenBioLLM-70B is tailored for the unique language and knowledge requirements of the medical and life sciences fields. It was fine-tuned on a vast corpus of high-quality biomedical data, enabling it to understand and generate text with domain-specific accuracy and fluency.
🎓 **Superior Performance**: With 70 billion parameters, OpenBioLLM-70B outperforms other open source biomedical language models of similar scale. It has also demonstrated better results compared to larger proprietary & open-source models like GPT-4, Gemini, Meditron-70B, Med-PaLM-1 & Med-PaLM-2 on biomedical benchmarks.
🧠 **Advanced Training Techniques**: OpenBioLLM-70B builds upon the powerful foundations of the **Meta-Llama-3-70B-Instruct** and [Meta-Llama-3-70B-Instruct](meta-llama/Meta-Llama-3-70B-Instruct) models. It incorporates the DPO dataset and fine-tuning recipe along with a custom diverse medical instruction dataset. Key components of the training pipeline include:
<div align="center">
<img width="1200px" src="https://cdn-uploads.huggingface.co/production/uploads/5f3fe13d79c1ba4c353d0c19/oPchsJsEpQoGcGXVbh7YS.png">
</div>
- **Policy Optimization**: [Direct Preference Optimization: Your Language Model is Secretly a Reward Model (DPO)](https://arxiv.org/abs/2305.18290)
- **Fine-tuning dataset**: Custom Medical Instruct dataset (We plan to release a sample training dataset in our upcoming paper; please stay updated)
This combination of cutting-edge techniques enables OpenBioLLM-70B to align with key capabilities and preferences for biomedical applications.
⚙️ **Release Details**:
- **Model Size**: 70 billion parameters
- **Quantization**: Optimized quantized versions available [Here](https://huggingface.co/aaditya/OpenBioLLM-70B-GGUF)
- **Language(s) (NLP):** en
- **Developed By**: [Ankit Pal (Aaditya Ura)](https://aadityaura.github.io/) from Saama AI Labs
- **License:** Meta-Llama License
- **Fine-tuned from models:** [Meta-Llama-3-70B-Instruct](meta-llama/Meta-Llama-3-70B-Instruct)
- **Resources for more information:**
- Paper: Coming soon
The model can be fine-tuned for more specialized tasks and datasets as needed.
OpenBioLLM-70B represents an important step forward in democratizing advanced language AI for the biomedical community. By leveraging state-of-the-art architectures and training techniques from leading open source efforts like Llama-3, we have created a powerful tool to accelerate innovation and discovery in healthcare and the life sciences.
We are excited to share OpenBioLLM-70B with researchers and developers around the world.
### Use with transformers
**Important: Please use the exact chat template provided by Llama-3 instruct version. Otherwise there will be a degradation in the performance. The model output can be verbose in rare cases. Please consider setting temperature = 0 to make this happen less.**
See the snippet below for usage with Transformers:
```python
import transformers
import torch
model_id = "aaditya/OpenBioLLM-Llama3-70B"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device="auto",
)
messages = [
{"role": "system", "content": "You are an expert and experienced from the healthcare and biomedical domain with extensive medical knowledge and practical experience. Your name is OpenBioLLM, and you were developed by Saama AI Labs. who's willing to help answer the user's query with explanation. In your explanation, leverage your deep medical expertise such as relevant anatomical structures, physiological processes, diagnostic criteria, treatment guidelines, or other pertinent medical concepts. Use precise medical terminology while still aiming to make the explanation clear and accessible to a general audience."},
{"role": "user", "content": "How can i split a 3mg or 4mg waefin pill so i can get a 2.5mg pill?"},
]
prompt = pipeline.tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
terminators = [
pipeline.tokenizer.eos_token_id,
pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = pipeline(
prompt,
max_new_tokens=256,
eos_token_id=terminators,
do_sample=True,
temperature=0.0,
top_p=0.9,
)
print(outputs[0]["generated_text"][len(prompt):])
```
## **Training procedure**
### **Training hyperparameters**
<details>
<summary>Click to see details</summary>
- learning_rate: 0.0002
- lr_scheduler: cosine
- train_batch_size: 12
- eval_batch_size: 8
- GPU: H100 80GB SXM5
- num_devices: 8
- optimizer: adamw_bnb_8bit
- lr_scheduler_warmup_steps: 100
- num_epochs: 4
</details>
### **Peft hyperparameters**
<details>
<summary>Click to see details</summary>
- adapter: qlora
- lora_r: 128
- lora_alpha: 256
- lora_dropout: 0.05
- lora_target_linear: true
-lora_target_modules:
- q_proj
- v_proj
- k_proj
- o_proj
- gate_proj
- down_proj
- up_proj
</details>
### **Training results**
### **Framework versions**
- Transformers 4.39.3
- Pytorch 2.1.2+cu121
- Datasets 2.18.0
- Tokenizers 0.15.1
- Axolotl
- Lm harness for evaluation
# Benchmark Results
🔥 OpenBioLLM-70B demonstrates superior performance compared to larger models, such as GPT-4, Gemini, Meditron-70B, Med-PaLM-1 & Med-PaLM-2 across 9 diverse biomedical datasets, achieving state-of-the-art results with an average score of 86.06%, despite having a significantly smaller parameter count. The model's strong performance in domain-specific tasks, such as Clinical KG, Medical Genetics, and PubMedQA, highlights its ability to effectively capture and apply biomedical knowledge.
🚨 The GPT-4, Med-PaLM-1, and Med-PaLM-2 results are taken from their official papers. Since Med-PaLM doesn't provide zero-shot accuracy, we are using 5-shot accuracy from their paper for comparison. All results presented are in the zero-shot setting, except for Med-PaLM-2 and Med-PaLM-1, which use 5-shot accuracy.
| | Clinical KG | Medical Genetics | Anatomy | Pro Medicine | College Biology | College Medicine | MedQA 4 opts | PubMedQA | MedMCQA | Avg |
|--------------------|-------------|------------------|---------|--------------|-----------------|------------------|--------------|----------|---------|-------|
| **OpenBioLLM-70B** | **92.93** | **93.197** | **83.904** | 93.75 | 93.827 | **85.749** | 78.162 | 78.97 | **74.014** | **86.05588** |
| Med-PaLM-2 (5-shot) | 88.3 | 90 | 77.8 | **95.2** | 94.4 | 80.9 | **79.7** | **79.2** | 71.3 | 84.08 |
| **GPT-4** | 86.04 | 91 | 80 | 93.01 | **95.14** | 76.88 | 78.87 | 75.2 | 69.52 | 82.85 |
| Med-PaLM-1 (Flan-PaLM, 5-shot) | 80.4 | 75 | 63.7 | 83.8 | 88.9 | 76.3 | 67.6 | 79 | 57.6 | 74.7 |
| **OpenBioLLM-8B** | 76.101 | 86.1 | 69.829 | 78.21 | 84.213 | 68.042 | 58.993 | 74.12 | 56.913 | 72.502 |
| Gemini-1.0 | 76.7 | 75.8 | 66.7 | 77.7 | 88 | 69.2 | 58 | 70.7 | 54.3 | 70.79 |
| GPT-3.5 Turbo 1106 | 74.71 | 74 | 72.79 | 72.79 | 72.91 | 64.73 | 57.71 | 72.66 | 53.79 | 66 |
| Meditron-70B | 66.79 | 69 | 53.33 | 71.69 | 76.38 | 63 | 57.1 | 76.6 | 46.85 | 64.52 |
| gemma-7b | 69.81 | 70 | 59.26 | 66.18 | 79.86 | 60.12 | 47.21 | 76.2 | 48.96 | 64.18 |
| Mistral-7B-v0.1 | 68.68 | 71 | 55.56 | 68.38 | 68.06 | 59.54 | 50.82 | 75.4 | 48.2 | 62.85 |
| Apollo-7B | 62.26 | 72 | 61.48 | 69.12 | 70.83 | 55.49 | 55.22 | 39.8 | 53.77 | 60 |
| MedAlpaca-7b | 57.36 | 69 | 57.04 | 67.28 | 65.28 | 54.34 | 41.71 | 72.8 | 37.51 | 58.03 |
| BioMistral-7B | 59.9 | 64 | 56.5 | 60.4 | 59 | 54.7 | 50.6 | 77.5 | 48.1 | 57.3 |
| AlpaCare-llama2-7b | 49.81 | 49 | 45.92 | 33.82 | 50 | 43.35 | 29.77 | 72.2 | 34.42 | 45.36 |
| ClinicalGPT | 30.56 | 27 | 30.37 | 19.48 | 25 | 24.27 | 26.08 | 63.8 | 28.18 | 30.52 |
<div align="center">
<img width="1600px" src="https://cdn-uploads.huggingface.co/production/uploads/5f3fe13d79c1ba4c353d0c19/_SzdcJSBjZyo8RS1bTEkP.png">
</div>
## Detailed Medical Subjectwise accuracy
# Use Cases & Examples
🚨 **Below results are from the quantized version of OpenBioLLM-70B
# Summarize Clinical Notes
OpenBioLLM-70B can efficiently analyze and summarize complex clinical notes, EHR data, and discharge summaries, extracting key information and generating concise, structured summaries
# Answer Medical Questions
OpenBioLLM-70B can provide answers to a wide range of medical questions.
<details>
<summary>Click to see details</summary>
</details>
# Clinical Entity Recognition
OpenBioLLM-70B can perform advanced clinical entity recognition by identifying and extracting key medical concepts, such as diseases, symptoms, medications, procedures, and anatomical structures, from unstructured clinical text. By leveraging its deep understanding of medical terminology and context, the model can accurately annotate and categorize clinical entities, enabling more efficient information retrieval, data analysis, and knowledge discovery from electronic health records, research articles, and other biomedical text sources. This capability can support various downstream applications, such as clinical decision support, pharmacovigilance, and medical research.
# Biomarkers Extraction
# Classification
OpenBioLLM-70B can perform various biomedical classification tasks, such as disease prediction, sentiment analysis, medical document categorization
# De-Identification
OpenBioLLM-70B can detect and remove personally identifiable information (PII) from medical records, ensuring patient privacy and compliance with data protection regulations like HIPAA.
**Advisory Notice!**
While OpenBioLLM-70B leverages high-quality data sources, its outputs may still contain inaccuracies, biases, or misalignments that could pose risks if relied upon for medical decision-making without further testing and refinement. The model's performance has not yet been rigorously evaluated in randomized controlled trials or real-world healthcare environments.
Therefore, we strongly advise against using OpenBioLLM-70B for any direct patient care, clinical decision support, or other professional medical purposes at this time. Its use should be limited to research, development, and exploratory applications by qualified individuals who understand its limitations.
OpenBioLLM-70B is intended solely as a research tool to assist healthcare professionals and should never be considered a replacement for the professional judgment and expertise of a qualified medical doctor.
Appropriately adapting and validating OpenBioLLM-70B for specific medical use cases would require significant additional work, potentially including:
- Thorough testing and evaluation in relevant clinical scenarios
- Alignment with evidence-based guidelines and best practices
- Mitigation of potential biases and failure modes
- Integration with human oversight and interpretation
- Compliance with regulatory and ethical standards
Always consult a qualified healthcare provider for personal medical needs.
# Citation
If you find OpenBioLLM-70B & 8B useful in your work, please cite the model as follows:
```
@misc{OpenBioLLMs,
author = {Ankit Pal, Malaikannan Sankarasubbu},
title = {OpenBioLLMs: Advancing Open-Source Large Language Models for Healthcare and Life Sciences},
year = {2024},
publisher = {Hugging Face},
journal = {Hugging Face repository},
howpublished = {\url{https://huggingface.co/aaditya/OpenBioLLM-Llama3-70B}}
}
```
The accompanying paper is currently in progress and will be released soon.
<div align="center">
<h2> 💌 Contact </h2>
</div>
We look forward to hearing you and collaborating on this exciting project!
**Contributors:**
- [Ankit Pal (Aaditya Ura)](https://aadityaura.github.io/) [aadityaura at gmail dot com]
- Saama AI Labs
- Note: I am looking for a funded PhD opportunity, especially if it fits my Responsible Generative AI, Multimodal LLMs, Geometric Deep Learning, and Healthcare AI skillset.
# References
We thank the [Meta Team](meta-llama/Meta-Llama-3-70B-Instruct) for their amazing models!
Result sources
- [1] GPT-4 [Capabilities of GPT-4 on Medical Challenge Problems] (https://arxiv.org/abs/2303.13375)
- [2] Med-PaLM-1 [Large Language Models Encode Clinical Knowledge](https://arxiv.org/abs/2212.13138)
- [3] Med-PaLM-2 [Towards Expert-Level Medical Question Answering with Large Language Models](https://arxiv.org/abs/2305.09617)
- [4] Gemini-1.0 [Gemini Goes to Med School](https://arxiv.org/abs/2402.07023) | [
"MEDQA",
"PUBMEDQA"
] |
silma-ai/SILMA-Kashif-2B-Instruct-v1.0 | silma-ai | text-generation | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"rag",
"qa",
"conversational",
"ar",
"en",
"license:gemma",
"model-index",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2025-01-26T16:01:49Z" | 2025-02-04T16:34:05+00:00 | 560 | 14 | ---
language:
- ar
- en
library_name: transformers
license: gemma
metrics:
- bleu
- bertscore
- rouge
- exact_match
pipeline_tag: text-generation
tags:
- rag
- qa
extra_gated_button_content: Acknowledge license
model-index:
- name: SILMA-Kashif-2B-Instruct-v1.0
results:
- task:
type: text-generation
dataset:
name: SILMA RAGQA Benchmark Dataset V1.0
type: silma-ai/silma-rag-qa-benchmark-v1.0
metrics:
- type: Average of Exact Match, BLEU, ROUGE, and BERTScore.
value: 0.347
name: SILMA RAGQA Benchmark Score
source:
url: https://huggingface.co/datasets/silma-ai/silma-rag-qa-benchmark-v1.0
name: SILMA RAGQA Benchmark
- task:
type: text-generation
dataset:
name: OALL (All)
type: OALL/Open-Arabic-LLM-Leaderboard
metrics:
- type: loglikelihood_acc_norm
value: 44.61
name: acc_norm
source:
url: https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard
name: Open Arabic LLM Leaderboard
---
## SILMA Kashif v1.0 (The Arabic RAG Model)
* **SILMA Kashif 2B Instruct v1.0** is the premier release within the SILMA Kashif Family of models, specifically designed for **RAG** (Retrieval-Augmented Generation) tasks
* Kashif excels in a specific task, answering questions based on contextual pieces in both Arabic and English. In addition, the model is also capable of performing Entity Extraction tasks as a minor skill
* SILMA Kashif 2B v1.0 stands out as the top-performing open model in RAG within the 3-9 billion parameter range based on our evaluations using [SILMA RAGQA Benchmark](https://huggingface.co/datasets/silma-ai/silma-rag-qa-benchmark-v1.0)
* SILMA Kashif is built on the powerful foundational models of Google Gemma, merging their strengths to deliver unmatched performance for users
* Kashif is an open-weight model, free to use in accordance with our open license
* Finally, the model comes with a context length of 12k
**Important note:** Kashif is a specialized model which should ONLY be used in RAG setups. If you are looking for a general purpose model, please refer to [SILMA 9B Instruct v1.0](https://huggingface.co/silma-ai/SILMA-9B-Instruct-v1.0)
## Model Skills and Capabilities
The model underwent intensive training to master a wide range of tasks and excel in performance.
- The ability to answer questions in Arabic and English
- The ability to deal with short and long contexts
- The ability to provide short and long answers effectively
- The ability to answer complex numerical questions
- The ability to answer questions based on tabular data
- Answering multi-hop questions: The ability to answer a single question using pieces of data from multiple paragraphs
- Negative rejection: The ability to identify and exclude inaccurate answers, and provide a more accurate statement such as "The answer cannot be found in the given context"
- Multi-domains: The ability to answer questions based on texts from different fields such as finance, medical, legal, etc.
- The ability to deal with ambiguous contexts
- The ability to extract entities from text
- Ability to deal with diverse and complex prompts
## Model Evaluation
|Dataset | exact_match | rouge1 | bleu | bertscore|
|---|---|---|---|---|
|ragbench-finqa-en-test | 0.000 | 0.587 | 0.321 | 0.760|
|ragbench-tatqa-ar-test | 0.000 | 0.484 | 0.130 | 0.774|
|ragbench-tatqa-en-test | 0.059 | 0.646 | 0.423 | 0.808|
|rag-instruct-benchmark-tester-en | 0.370 | 0.683 | 0.196 | 0.791|
|ragbench-expertqa-en-test |0.000 | 0.465 | 0.151 | 0.677|
|ragbench-msmarco-ar-test |0.000 | 0.144 | 0.096 | 0.781|
|sciq-ar-test |0.170 | 0.000 | 0.000 | 0.753|
|ragbench-covidqa-en-test |0.020 | 0.521 | 0.242 | 0.734|
|ragbench-emanual-ar-test |0.000 | 0.237 | 0.159 | 0.806|
|ragbench-finqa-ar-test |0.000 | 0.377 | 0.109 | 0.780|
|xquad-r-validation-en |0.120 | 0.326 | 0.041 | 0.603|
|ragbench-emanual-en-test |0.000 | 0.565 | 0.288 | 0.722|
|xquad-r-ar-validation |0.070 | 0.130 | 0.042 | 0.698|
|boolq-ar-test |0.450 | 0.000 | 0.000 | 0.700|
|ragbench-hotpotqa-en-test |0.060 | 0.732 | 0.503 | 0.837|
|ragbench-covidqa-ar-test |0.000 | 0.179 | 0.104 | 0.783|
|ragbench-msmarco-en-test |0.020 | 0.491 | 0.207 | 0.729|
|### Benchmark Average Scores |0.079 | 0.386 | 0.177 | 0.749|
SILMA RAG QA Benchmark Score: 0.3478
## SILMA AI
[silma.ai](https://silma.ai) is a leading GenAI startup that excels in building and tailoring cutting-edge Large Language Models (LLMs) and AI technologies for the Arabic language
### Usage
Below we share some code snippets on how to get quickly started with running the model. First, install the Transformers library with:
```sh
pip install -U transformers
```
Then, copy the snippet from the section below
#### Running with the `pipeline` API
```python
import torch
from transformers import pipeline
pipe = pipeline(
"text-generation",
model="silma-ai/SILMA-Kashif-2B-Instruct-v1.0",
model_kwargs={"torch_dtype": torch.bfloat16},
device="cuda", # replace with "mps" to run on a Mac device
)
messages = [
{"role": "user", "content":
"""
أجب على السؤال بناءً على السياق أدناه
السياق:
تشمل الاتفاقيات رسوم حمل سنوية ثابت قدها 30 مليون جنيه إسترليني للقنوات نظراً لأن كلاً من مزوديها قادرين على تأمين دفعات إضافية إذا ما حققت هذه القنوات أهدافاً متعلقةً بالأداء.
لا يوجد حالياً ما يشير إلى ما إذا كان الاتفاق الجديد يشمل محتوىً إضافياً كالفيديو عند الطلب والدقة العالية ، كذلك الذي سبق أن قدمته بي سكاي بي.
وقد وافقت كل من بي سكاي بي و فيرجين ميديا على إنهاء الدعاوى القضائية بالمحكمة العليا ضد بعضهما بشأن معاليم الحمل التي تخص قنواتهما الأساسية.
السؤال: ماسم الشركة التي وافقت على إنهاء دعواها القضائية ضد بي سكاي بي بالمحكمة العليا؟
الإجابة:
"""},
]
outputs = pipe(messages, max_new_tokens=600)
assistant_response = outputs[0]["generated_text"][-1]["content"].strip()
print(assistant_response)
```
- Response:
```text
فيرجين ميديا
"وقد وافقت كل من بي سكاي بي و فيرجين ميديا على إنهاء الدعاوى القضائية بالمحكمة العليا ضد بعضهما بشأن معاليم الحمل التي تخص قنواتهما الأساسية."
```
Note: for advanced usage examples such as multi-gpu, quantization or chat templates, please refer to [SILMA v1.0](https://huggingface.co/silma-ai/SILMA-9B-Instruct-v1.0#running-the-model-on-a-single--multi-gpu) examples
### Runing with Ollama
```sh
ollama run hf.co/silma-ai/SILMA-Kashif-2B-Instruct-v1.0-GGUF
```
### Prompt Format
Here is a recommended way to prompt the model. You can modify the prompt based on your specific requirements, but if you encounter any challenges, following the format below in which we used to train the model may be helpful.
- Arabic
```text
أجب على السؤال بناءً على السياق أدناه
السياق:
.....
.....
السؤال: ...
الإجابة: ...
```
- English
```text
Answer the following question using the provided context below
Context:
.....
.....
Question: ...
Answer: ...
```
### GPU Requirements
The following are the minimum/recommended GPU requirements for running inference:
* Recommended
* At least one GPU with a minimum of 24 GB of GPU memory
* Examples: Nvidia RTX 4090
* Minimum
* At least one GPU with 8 GB of GPU memory
* Examples: Nvidia RTX 3070, RTX 3080 or T4
## Effect of Quantization
We have seen 2.6% drop in score (to 0.338) for the same model quantized 4bit
### Citation
```none
@article{silma_01_2025,
title={SILMA Kashif 2B Instruct v1.0},
url={https://huggingface.co/silma-ai/SILMA-Kashif-2B-Instruct-v1.0},
publisher={SILMA AI},
author={Silma Team},
year={2025}
}
```
### Intended Usage
* The model should only be used in question answering use-cases such as RAG
* The model can also be used to extract entities from text
### Limitations
* Because it has few parameters, we've noticed that the model isn't very effective for handling complex numerical and financial reasoning, such as solving tricky calculations
* The model has been trained specifically for text-based question answering, which may limit its ability to perform tasks beyond this scope, including simple tasks | [
"SCIQ"
] |
infly/inf-retriever-v1 | infly | sentence-similarity | [
"sentence-transformers",
"safetensors",
"qwen2",
"feature-extraction",
"transformers",
"sentence-similarity",
"mteb",
"custom_code",
"en",
"zh",
"base_model:Alibaba-NLP/gte-Qwen2-7B-instruct",
"base_model:finetune:Alibaba-NLP/gte-Qwen2-7B-instruct",
"doi:10.57967/hf/4262",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-generation-inference",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2024-12-23T09:48:31Z" | 2025-02-19T10:04:50+00:00 | 559 | 20 | ---
base_model:
- Alibaba-NLP/gte-Qwen2-7B-instruct
language:
- en
- zh
license: apache-2.0
tags:
- sentence-transformers
- transformers
- sentence-similarity
- mteb
model-index:
- name: infly/inf-retriever-v1
results:
- task:
type: Retrieval
dataset:
name: MTEB CmedqaRetrieval (default)
type: C-MTEB/CmedqaRetrieval
config: default
split: dev
revision: cd540c506dae1cf9e9a59c3e06f42030d54e7301
metrics:
- type: ndcg_at_1
value: 38.185
- type: ndcg_at_3
value: 38.438
- type: ndcg_at_5
value: 40.445
- type: ndcg_at_10
value: 43.308
- type: ndcg_at_20
value: 46.177
- type: ndcg_at_100
value: 50.644999999999996
- type: ndcg_at_1000
value: 52.819
- type: recall_at_1
value: 25.14
- type: recall_at_3
value: 38.253
- type: recall_at_5
value: 44.507999999999996
- type: recall_at_10
value: 53.025
- type: recall_at_20
value: 62.89
- type: recall_at_100
value: 83.487
- type: recall_at_1000
value: 98.059
- type: main_score
value: 43.308
- task:
type: Retrieval
dataset:
name: MTEB CovidRetrieval (default)
type: C-MTEB/CovidRetrieval
config: default
split: dev
revision: 1271c7809071a13532e05f25fb53511ffce77117
metrics:
- type: ndcg_at_1
value: 77.97699999999999
- type: ndcg_at_3
value: 85.24199999999999
- type: ndcg_at_5
value: 86.901
- type: ndcg_at_10
value: 87.77000000000001
- type: ndcg_at_20
value: 88.295
- type: ndcg_at_100
value: 88.479
- type: ndcg_at_1000
value: 88.527
- type: recall_at_1
value: 77.819
- type: recall_at_3
value: 89.96300000000001
- type: recall_at_5
value: 93.941
- type: recall_at_10
value: 96.575
- type: recall_at_20
value: 98.63
- type: recall_at_100
value: 99.579
- type: recall_at_1000
value: 100.0
- type: main_score
value: 87.77000000000001
- task:
type: Retrieval
dataset:
name: MTEB DuRetrieval (default)
type: C-MTEB/DuRetrieval
config: default
split: dev
revision: a1a333e290fe30b10f3f56498e3a0d911a693ced
metrics:
- type: ndcg_at_1
value: 91.45
- type: ndcg_at_3
value: 89.249
- type: ndcg_at_5
value: 88.506
- type: ndcg_at_10
value: 90.66
- type: ndcg_at_20
value: 91.886
- type: ndcg_at_100
value: 92.78699999999999
- type: ndcg_at_1000
value: 92.944
- type: recall_at_1
value: 27.332
- type: recall_at_3
value: 61.07599999999999
- type: recall_at_5
value: 78.49199999999999
- type: recall_at_10
value: 92.002
- type: recall_at_20
value: 96.116
- type: recall_at_100
value: 99.009
- type: recall_at_1000
value: 99.844
- type: main_score
value: 90.66
- task:
type: Retrieval
dataset:
name: MTEB EcomRetrieval (default)
type: C-MTEB/EcomRetrieval
config: default
split: dev
revision: 687de13dc7294d6fd9be10c6945f9e8fec8166b9
metrics:
- type: ndcg_at_1
value: 55.900000000000006
- type: ndcg_at_3
value: 66.019
- type: ndcg_at_5
value: 68.47999999999999
- type: ndcg_at_10
value: 70.678
- type: ndcg_at_20
value: 72.024
- type: ndcg_at_100
value: 72.933
- type: ndcg_at_1000
value: 73.20400000000001
- type: recall_at_1
value: 55.900000000000006
- type: recall_at_3
value: 73.1
- type: recall_at_5
value: 79.10000000000001
- type: recall_at_10
value: 85.9
- type: recall_at_20
value: 91.2
- type: recall_at_100
value: 96.1
- type: recall_at_1000
value: 98.3
- type: main_score
value: 70.678
- task:
type: Retrieval
dataset:
name: MTEB MMarcoRetrieval (default)
type: C-MTEB/MMarcoRetrieval
config: default
split: dev
revision: 539bbde593d947e2a124ba72651aafc09eb33fc2
metrics:
- type: ndcg_at_1
value: 75.74499999999999
- type: ndcg_at_3
value: 82.188
- type: ndcg_at_5
value: 83.869
- type: ndcg_at_10
value: 85.119
- type: ndcg_at_20
value: 85.624
- type: ndcg_at_100
value: 86.051
- type: ndcg_at_1000
value: 86.177
- type: recall_at_1
value: 73.33
- type: recall_at_3
value: 86.823
- type: recall_at_5
value: 90.814
- type: recall_at_10
value: 94.509
- type: recall_at_20
value: 96.422
- type: recall_at_100
value: 98.6
- type: recall_at_1000
value: 99.599
- type: main_score
value: 85.119
- task:
type: Retrieval
dataset:
name: MTEB MedicalRetrieval (default)
type: C-MTEB/MedicalRetrieval
config: default
split: dev
revision: 2039188fb5800a9803ba5048df7b76e6fb151fc6
metrics:
- type: ndcg_at_1
value: 55.00000000000001
- type: ndcg_at_3
value: 61.334
- type: ndcg_at_5
value: 62.590999999999994
- type: ndcg_at_10
value: 63.913
- type: ndcg_at_20
value: 64.748
- type: ndcg_at_100
value: 66.675
- type: ndcg_at_1000
value: 67.894
- type: recall_at_1
value: 55.00000000000001
- type: recall_at_3
value: 65.60000000000001
- type: recall_at_5
value: 68.60000000000001
- type: recall_at_10
value: 72.7
- type: recall_at_20
value: 76.0
- type: recall_at_100
value: 86.6
- type: recall_at_1000
value: 96.3
- type: main_score
value: 63.913
- task:
type: Retrieval
dataset:
name: MTEB T2Retrieval (default)
type: C-MTEB/T2Retrieval
config: default
split: dev
revision: 8731a845f1bf500a4f111cf1070785c793d10e64
metrics:
- type: ndcg_at_1
value: 91.526
- type: ndcg_at_3
value: 88.35499999999999
- type: ndcg_at_5
value: 87.408
- type: ndcg_at_10
value: 87.641
- type: ndcg_at_20
value: 89.265
- type: ndcg_at_100
value: 90.693
- type: ndcg_at_1000
value: 91.105
- type: recall_at_1
value: 28.359
- type: recall_at_3
value: 58.101
- type: recall_at_5
value: 72.99
- type: recall_at_10
value: 86.921
- type: recall_at_20
value: 92.497
- type: recall_at_100
value: 96.978
- type: recall_at_1000
value: 99.075
- type: main_score
value: 87.641
- task:
type: Retrieval
dataset:
name: MTEB VideoRetrieval (default)
type: C-MTEB/VideoRetrieval
config: default
split: dev
revision: 58c2597a5943a2ba48f4668c3b90d796283c5639
metrics:
- type: ndcg_at_1
value: 66.0
- type: ndcg_at_3
value: 75.495
- type: ndcg_at_5
value: 77.027
- type: ndcg_at_10
value: 78.606
- type: ndcg_at_20
value: 79.54599999999999
- type: ndcg_at_100
value: 80.326
- type: ndcg_at_1000
value: 80.516
- type: recall_at_1
value: 66.0
- type: recall_at_3
value: 81.89999999999999
- type: recall_at_5
value: 85.6
- type: recall_at_10
value: 90.4
- type: recall_at_20
value: 94.1
- type: recall_at_100
value: 98.2
- type: recall_at_1000
value: 99.7
- type: main_score
value: 78.606
- task:
type: Retrieval
dataset:
name: MTEB AILACasedocs (default)
type: mteb/AILA_casedocs
config: default
split: test
revision: 4106e6bcc72e0698d714ea8b101355e3e238431a
metrics:
- type: ndcg_at_1
value: 40.0
- type: ndcg_at_3
value: 37.37
- type: ndcg_at_5
value: 37.913999999999994
- type: ndcg_at_10
value: 41.162
- type: ndcg_at_20
value: 45.72
- type: ndcg_at_100
value: 54.126
- type: ndcg_at_1000
value: 55.907
- type: recall_at_1
value: 15.406
- type: recall_at_3
value: 26.56
- type: recall_at_5
value: 33.084
- type: recall_at_10
value: 45.972
- type: recall_at_20
value: 60.775
- type: recall_at_100
value: 91.105
- type: recall_at_1000
value: 100.0
- type: main_score
value: 41.162
- task:
type: Retrieval
dataset:
name: MTEB AILAStatutes (default)
type: mteb/AILA_statutes
config: default
split: test
revision: ebfcd844eadd3d667efa3c57fc5c8c87f5c2867e
metrics:
- type: ndcg_at_1
value: 36.0
- type: ndcg_at_3
value: 32.427
- type: ndcg_at_5
value: 31.512
- type: ndcg_at_10
value: 37.727
- type: ndcg_at_20
value: 43.808
- type: ndcg_at_100
value: 56.445
- type: ndcg_at_1000
value: 56.445
- type: recall_at_1
value: 8.1
- type: recall_at_3
value: 20.599999999999998
- type: recall_at_5
value: 30.733
- type: recall_at_10
value: 42.733
- type: recall_at_20
value: 57.733000000000004
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 37.727
- task:
type: Retrieval
dataset:
name: MTEB AlloprofRetrieval (default)
type: lyon-nlp/alloprof
config: default
split: test
revision: fcf295ea64c750f41fadbaa37b9b861558e1bfbd
metrics:
- type: ndcg_at_1
value: 45.509
- type: ndcg_at_3
value: 57.912
- type: ndcg_at_5
value: 60.885
- type: ndcg_at_10
value: 63.611
- type: ndcg_at_20
value: 64.976
- type: ndcg_at_100
value: 66.507
- type: ndcg_at_1000
value: 66.998
- type: recall_at_1
value: 45.509
- type: recall_at_3
value: 66.537
- type: recall_at_5
value: 73.748
- type: recall_at_10
value: 82.16799999999999
- type: recall_at_20
value: 87.522
- type: recall_at_100
value: 95.72500000000001
- type: recall_at_1000
value: 99.655
- type: main_score
value: 63.611
- task:
type: Retrieval
dataset:
name: MTEB AppsRetrieval (default)
type: CoIR-Retrieval/apps
config: default
split: test
revision: f22508f96b7a36c2415181ed8bb76f76e04ae2d5
metrics:
- type: ndcg_at_1
value: 35.405
- type: ndcg_at_3
value: 42.945
- type: ndcg_at_5
value: 44.984
- type: ndcg_at_10
value: 47.369
- type: ndcg_at_20
value: 49.095
- type: ndcg_at_100
value: 51.821
- type: ndcg_at_1000
value: 53.581
- type: recall_at_1
value: 35.405
- type: recall_at_3
value: 48.287
- type: recall_at_5
value: 53.227000000000004
- type: recall_at_10
value: 60.611000000000004
- type: recall_at_20
value: 67.437
- type: recall_at_100
value: 82.231
- type: recall_at_1000
value: 96.38799999999999
- type: main_score
value: 47.369
- task:
type: Retrieval
dataset:
name: MTEB ArguAna (default)
type: mteb/arguana
config: default
split: test
revision: c22ab2a51041ffd869aaddef7af8d8215647e41a
metrics:
- type: ndcg_at_1
value: 69.132
- type: ndcg_at_3
value: 81.661
- type: ndcg_at_5
value: 83.773
- type: ndcg_at_10
value: 84.855
- type: ndcg_at_20
value: 85.073
- type: ndcg_at_100
value: 85.134
- type: ndcg_at_1000
value: 85.134
- type: recall_at_1
value: 69.132
- type: recall_at_3
value: 90.185
- type: recall_at_5
value: 95.235
- type: recall_at_10
value: 98.506
- type: recall_at_20
value: 99.36
- type: recall_at_100
value: 99.644
- type: recall_at_1000
value: 99.644
- type: main_score
value: 84.855
- task:
type: Retrieval
dataset:
name: MTEB ArguAna-PL (default)
type: clarin-knext/arguana-pl
config: default
split: test
revision: 63fc86750af76253e8c760fc9e534bbf24d260a2
metrics:
- type: ndcg_at_1
value: 46.657
- type: ndcg_at_3
value: 63.388999999999996
- type: ndcg_at_5
value: 67.931
- type: ndcg_at_10
value: 70.745
- type: ndcg_at_20
value: 71.60300000000001
- type: ndcg_at_100
value: 71.941
- type: ndcg_at_1000
value: 71.961
- type: recall_at_1
value: 46.657
- type: recall_at_3
value: 75.036
- type: recall_at_5
value: 85.989
- type: recall_at_10
value: 94.523
- type: recall_at_20
value: 97.795
- type: recall_at_100
value: 99.502
- type: recall_at_1000
value: 99.644
- type: main_score
value: 70.745
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval (default)
type: mteb/cqadupstack-android
config: default
split: test
revision: f46a197baaae43b4f621051089b82a364682dfeb
metrics:
- type: ndcg_at_1
value: 45.494
- type: ndcg_at_3
value: 51.53
- type: ndcg_at_5
value: 54.062
- type: ndcg_at_10
value: 56.599
- type: ndcg_at_20
value: 58.663
- type: ndcg_at_100
value: 61.36200000000001
- type: ndcg_at_1000
value: 62.824000000000005
- type: recall_at_1
value: 37.078
- type: recall_at_3
value: 53.529
- type: recall_at_5
value: 60.772999999999996
- type: recall_at_10
value: 68.65299999999999
- type: recall_at_20
value: 75.92999999999999
- type: recall_at_100
value: 88.127
- type: recall_at_1000
value: 97.059
- type: main_score
value: 56.599
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackEnglishRetrieval (default)
type: mteb/cqadupstack-english
config: default
split: test
revision: ad9991cb51e31e31e430383c75ffb2885547b5f0
metrics:
- type: ndcg_at_1
value: 47.134
- type: ndcg_at_3
value: 52.186
- type: ndcg_at_5
value: 53.94
- type: ndcg_at_10
value: 55.96
- type: ndcg_at_20
value: 57.521
- type: ndcg_at_100
value: 59.865
- type: ndcg_at_1000
value: 61.611000000000004
- type: recall_at_1
value: 37.405
- type: recall_at_3
value: 53.869
- type: recall_at_5
value: 59.18600000000001
- type: recall_at_10
value: 65.786
- type: recall_at_20
value: 71.56099999999999
- type: recall_at_100
value: 82.062
- type: recall_at_1000
value: 92.863
- type: main_score
value: 55.96
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGamingRetrieval (default)
type: mteb/cqadupstack-gaming
config: default
split: test
revision: 4885aa143210c98657558c04aaf3dc47cfb54340
metrics:
- type: ndcg_at_1
value: 52.22599999999999
- type: ndcg_at_3
value: 59.797999999999995
- type: ndcg_at_5
value: 62.260000000000005
- type: ndcg_at_10
value: 64.85300000000001
- type: ndcg_at_20
value: 66.398
- type: ndcg_at_100
value: 68.298
- type: ndcg_at_1000
value: 69.003
- type: recall_at_1
value: 45.789
- type: recall_at_3
value: 64.9
- type: recall_at_5
value: 70.902
- type: recall_at_10
value: 78.388
- type: recall_at_20
value: 84.086
- type: recall_at_100
value: 93.006
- type: recall_at_1000
value: 97.928
- type: main_score
value: 64.85300000000001
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGisRetrieval (default)
type: mteb/cqadupstack-gis
config: default
split: test
revision: 5003b3064772da1887988e05400cf3806fe491f2
metrics:
- type: ndcg_at_1
value: 32.09
- type: ndcg_at_3
value: 38.339
- type: ndcg_at_5
value: 41.427
- type: ndcg_at_10
value: 43.606
- type: ndcg_at_20
value: 45.784000000000006
- type: ndcg_at_100
value: 48.908
- type: ndcg_at_1000
value: 50.585
- type: recall_at_1
value: 29.146
- type: recall_at_3
value: 43.168
- type: recall_at_5
value: 50.717
- type: recall_at_10
value: 57.120000000000005
- type: recall_at_20
value: 65.254
- type: recall_at_100
value: 81.04599999999999
- type: recall_at_1000
value: 93.487
- type: main_score
value: 43.606
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackMathematicaRetrieval (default)
type: mteb/cqadupstack-mathematica
config: default
split: test
revision: 90fceea13679c63fe563ded68f3b6f06e50061de
metrics:
- type: ndcg_at_1
value: 24.876
- type: ndcg_at_3
value: 29.663
- type: ndcg_at_5
value: 32.193
- type: ndcg_at_10
value: 34.694
- type: ndcg_at_20
value: 37.075
- type: ndcg_at_100
value: 40.615
- type: ndcg_at_1000
value: 43.317
- type: recall_at_1
value: 20.395
- type: recall_at_3
value: 32.521
- type: recall_at_5
value: 38.887
- type: recall_at_10
value: 46.388
- type: recall_at_20
value: 54.885
- type: recall_at_100
value: 71.597
- type: recall_at_1000
value: 90.75
- type: main_score
value: 34.694
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackPhysicsRetrieval (default)
type: mteb/cqadupstack-physics
config: default
split: test
revision: 79531abbd1fb92d06c6d6315a0cbbbf5bb247ea4
metrics:
- type: ndcg_at_1
value: 42.733
- type: ndcg_at_3
value: 47.236
- type: ndcg_at_5
value: 49.327
- type: ndcg_at_10
value: 52.346000000000004
- type: ndcg_at_20
value: 54.446000000000005
- type: ndcg_at_100
value: 57.736
- type: ndcg_at_1000
value: 59.245000000000005
- type: recall_at_1
value: 34.414
- type: recall_at_3
value: 50.233000000000004
- type: recall_at_5
value: 55.967
- type: recall_at_10
value: 65.173
- type: recall_at_20
value: 72.27799999999999
- type: recall_at_100
value: 87.163
- type: recall_at_1000
value: 96.64
- type: main_score
value: 52.346000000000004
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackProgrammersRetrieval (default)
type: mteb/cqadupstack-programmers
config: default
split: test
revision: 6184bc1440d2dbc7612be22b50686b8826d22b32
metrics:
- type: ndcg_at_1
value: 37.329
- type: ndcg_at_3
value: 41.319
- type: ndcg_at_5
value: 43.444
- type: ndcg_at_10
value: 46.643
- type: ndcg_at_20
value: 49.257
- type: ndcg_at_100
value: 52.524
- type: ndcg_at_1000
value: 54.478
- type: recall_at_1
value: 30.278
- type: recall_at_3
value: 43.464999999999996
- type: recall_at_5
value: 49.419999999999995
- type: recall_at_10
value: 58.650999999999996
- type: recall_at_20
value: 67.90899999999999
- type: recall_at_100
value: 83.276
- type: recall_at_1000
value: 96.114
- type: main_score
value: 46.643
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackRetrieval (default)
type: CQADupstackRetrieval_is_a_combined_dataset
config: default
split: test
revision: CQADupstackRetrieval_is_a_combined_dataset
metrics:
- type: main_score
value: 46.644083333333334
- type: ndcg_at_10
value: 46.644083333333334
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackStatsRetrieval (default)
type: mteb/cqadupstack-stats
config: default
split: test
revision: 65ac3a16b8e91f9cee4c9828cc7c335575432a2a
metrics:
- type: ndcg_at_1
value: 30.368000000000002
- type: ndcg_at_3
value: 35.004000000000005
- type: ndcg_at_5
value: 37.125
- type: ndcg_at_10
value: 39.831
- type: ndcg_at_20
value: 42.099
- type: ndcg_at_100
value: 45.032
- type: ndcg_at_1000
value: 47.016999999999996
- type: recall_at_1
value: 27.151999999999997
- type: recall_at_3
value: 38.2
- type: recall_at_5
value: 43.349
- type: recall_at_10
value: 51.50599999999999
- type: recall_at_20
value: 60.035000000000004
- type: recall_at_100
value: 74.869
- type: recall_at_1000
value: 89.159
- type: main_score
value: 39.831
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackTexRetrieval (default)
type: mteb/cqadupstack-tex
config: default
split: test
revision: 46989137a86843e03a6195de44b09deda022eec7
metrics:
- type: ndcg_at_1
value: 26.222
- type: ndcg_at_3
value: 30.085
- type: ndcg_at_5
value: 31.977
- type: ndcg_at_10
value: 34.107
- type: ndcg_at_20
value: 35.939
- type: ndcg_at_100
value: 39.054
- type: ndcg_at_1000
value: 41.899
- type: recall_at_1
value: 21.552
- type: recall_at_3
value: 32.66
- type: recall_at_5
value: 37.785000000000004
- type: recall_at_10
value: 44.143
- type: recall_at_20
value: 50.968999999999994
- type: recall_at_100
value: 66.392
- type: recall_at_1000
value: 86.601
- type: main_score
value: 34.107
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackUnixRetrieval (default)
type: mteb/cqadupstack-unix
config: default
split: test
revision: 6c6430d3a6d36f8d2a829195bc5dc94d7e063e53
metrics:
- type: ndcg_at_1
value: 36.287000000000006
- type: ndcg_at_3
value: 41.15
- type: ndcg_at_5
value: 43.283
- type: ndcg_at_10
value: 45.698
- type: ndcg_at_20
value: 47.754000000000005
- type: ndcg_at_100
value: 50.800999999999995
- type: ndcg_at_1000
value: 53.024
- type: recall_at_1
value: 30.791
- type: recall_at_3
value: 44.802
- type: recall_at_5
value: 50.434999999999995
- type: recall_at_10
value: 57.424
- type: recall_at_20
value: 64.702
- type: recall_at_100
value: 79.216
- type: recall_at_1000
value: 94.602
- type: main_score
value: 45.698
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWebmastersRetrieval (default)
type: mteb/cqadupstack-webmasters
config: default
split: test
revision: 160c094312a0e1facb97e55eeddb698c0abe3571
metrics:
- type: ndcg_at_1
value: 37.352000000000004
- type: ndcg_at_3
value: 43.029
- type: ndcg_at_5
value: 44.811
- type: ndcg_at_10
value: 47.493
- type: ndcg_at_20
value: 49.76
- type: ndcg_at_100
value: 52.925
- type: ndcg_at_1000
value: 55.117000000000004
- type: recall_at_1
value: 31.719
- type: recall_at_3
value: 45.466
- type: recall_at_5
value: 50.087
- type: recall_at_10
value: 57.86
- type: recall_at_20
value: 66.27
- type: recall_at_100
value: 81.437
- type: recall_at_1000
value: 95.162
- type: main_score
value: 47.493
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWordpressRetrieval (default)
type: mteb/cqadupstack-wordpress
config: default
split: test
revision: 4ffe81d471b1924886b33c7567bfb200e9eec5c4
metrics:
- type: ndcg_at_1
value: 29.020000000000003
- type: ndcg_at_3
value: 33.715
- type: ndcg_at_5
value: 35.266
- type: ndcg_at_10
value: 37.899
- type: ndcg_at_20
value: 39.812999999999995
- type: ndcg_at_100
value: 42.998999999999995
- type: ndcg_at_1000
value: 45.257
- type: recall_at_1
value: 26.784000000000002
- type: recall_at_3
value: 37.049
- type: recall_at_5
value: 40.638000000000005
- type: recall_at_10
value: 48.204
- type: recall_at_20
value: 55.496
- type: recall_at_100
value: 71.749
- type: recall_at_1000
value: 88.22
- type: main_score
value: 37.899
- task:
type: Retrieval
dataset:
name: MTEB CodeFeedbackMT (default)
type: CoIR-Retrieval/codefeedback-mt
config: default
split: test
revision: b0f12fa0c0dd67f59c95a5c33d02aeeb4c398c5f
metrics:
- type: ndcg_at_1
value: 67.214
- type: ndcg_at_3
value: 74.774
- type: ndcg_at_5
value: 76.297
- type: ndcg_at_10
value: 77.644
- type: ndcg_at_20
value: 78.41
- type: ndcg_at_100
value: 79.374
- type: ndcg_at_1000
value: 79.77
- type: recall_at_1
value: 67.214
- type: recall_at_3
value: 79.95
- type: recall_at_5
value: 83.65599999999999
- type: recall_at_10
value: 87.776
- type: recall_at_20
value: 90.781
- type: recall_at_100
value: 95.993
- type: recall_at_1000
value: 99.104
- type: main_score
value: 77.644
- task:
type: Retrieval
dataset:
name: MTEB CodeFeedbackST (default)
type: CoIR-Retrieval/codefeedback-st
config: default
split: test
revision: d213819e87aab9010628da8b73ab4eb337c89340
metrics:
- type: ndcg_at_1
value: 74.05000000000001
- type: ndcg_at_3
value: 84.59
- type: ndcg_at_5
value: 85.949
- type: ndcg_at_10
value: 86.627
- type: ndcg_at_20
value: 86.907
- type: ndcg_at_100
value: 87.149
- type: ndcg_at_1000
value: 87.21799999999999
- type: recall_at_1
value: 74.05000000000001
- type: recall_at_3
value: 91.685
- type: recall_at_5
value: 94.959
- type: recall_at_10
value: 97.017
- type: recall_at_20
value: 98.10900000000001
- type: recall_at_100
value: 99.396
- type: recall_at_1000
value: 99.92699999999999
- type: main_score
value: 86.627
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetCCRetrieval (python)
type: CoIR-Retrieval/CodeSearchNet-ccr
config: python
split: test
revision: 6e1effa2c03723c5fde48ee912b5ee08d4f211e8
metrics:
- type: ndcg_at_1
value: 69.875
- type: ndcg_at_3
value: 79.45100000000001
- type: ndcg_at_5
value: 80.95400000000001
- type: ndcg_at_10
value: 82.025
- type: ndcg_at_20
value: 82.526
- type: ndcg_at_100
value: 83.07
- type: ndcg_at_1000
value: 83.28999999999999
- type: recall_at_1
value: 69.875
- type: recall_at_3
value: 85.957
- type: recall_at_5
value: 89.59
- type: recall_at_10
value: 92.874
- type: recall_at_20
value: 94.838
- type: recall_at_100
value: 97.748
- type: recall_at_1000
value: 99.47
- type: main_score
value: 82.025
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetCCRetrieval (javascript)
type: CoIR-Retrieval/CodeSearchNet-ccr
config: javascript
split: test
revision: 6e1effa2c03723c5fde48ee912b5ee08d4f211e8
metrics:
- type: ndcg_at_1
value: 66.18
- type: ndcg_at_3
value: 76.294
- type: ndcg_at_5
value: 77.849
- type: ndcg_at_10
value: 78.95400000000001
- type: ndcg_at_20
value: 79.71000000000001
- type: ndcg_at_100
value: 80.402
- type: ndcg_at_1000
value: 80.694
- type: recall_at_1
value: 66.18
- type: recall_at_3
value: 83.10499999999999
- type: recall_at_5
value: 86.873
- type: recall_at_10
value: 90.277
- type: recall_at_20
value: 93.22399999999999
- type: recall_at_100
value: 96.87
- type: recall_at_1000
value: 99.21
- type: main_score
value: 78.95400000000001
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetCCRetrieval (go)
type: CoIR-Retrieval/CodeSearchNet-ccr
config: go
split: test
revision: 6e1effa2c03723c5fde48ee912b5ee08d4f211e8
metrics:
- type: ndcg_at_1
value: 57.24
- type: ndcg_at_3
value: 67.84700000000001
- type: ndcg_at_5
value: 70.126
- type: ndcg_at_10
value: 71.839
- type: ndcg_at_20
value: 72.89
- type: ndcg_at_100
value: 73.904
- type: ndcg_at_1000
value: 74.343
- type: recall_at_1
value: 57.24
- type: recall_at_3
value: 75.179
- type: recall_at_5
value: 80.67
- type: recall_at_10
value: 85.939
- type: recall_at_20
value: 90.076
- type: recall_at_100
value: 95.48100000000001
- type: recall_at_1000
value: 98.929
- type: main_score
value: 71.839
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetCCRetrieval (ruby)
type: CoIR-Retrieval/CodeSearchNet-ccr
config: ruby
split: test
revision: 6e1effa2c03723c5fde48ee912b5ee08d4f211e8
metrics:
- type: ndcg_at_1
value: 64.235
- type: ndcg_at_3
value: 73.451
- type: ndcg_at_5
value: 75.233
- type: ndcg_at_10
value: 76.53
- type: ndcg_at_20
value: 77.35
- type: ndcg_at_100
value: 78.13799999999999
- type: ndcg_at_1000
value: 78.57
- type: recall_at_1
value: 64.235
- type: recall_at_3
value: 79.699
- type: recall_at_5
value: 83.981
- type: recall_at_10
value: 88.02499999999999
- type: recall_at_20
value: 91.277
- type: recall_at_100
value: 95.638
- type: recall_at_1000
value: 99.048
- type: main_score
value: 76.53
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetCCRetrieval (java)
type: CoIR-Retrieval/CodeSearchNet-ccr
config: java
split: test
revision: 6e1effa2c03723c5fde48ee912b5ee08d4f211e8
metrics:
- type: ndcg_at_1
value: 65.468
- type: ndcg_at_3
value: 75.064
- type: ndcg_at_5
value: 76.786
- type: ndcg_at_10
value: 77.929
- type: ndcg_at_20
value: 78.596
- type: ndcg_at_100
value: 79.28699999999999
- type: ndcg_at_1000
value: 79.625
- type: recall_at_1
value: 65.468
- type: recall_at_3
value: 81.56099999999999
- type: recall_at_5
value: 85.714
- type: recall_at_10
value: 89.229
- type: recall_at_20
value: 91.83
- type: recall_at_100
value: 95.509
- type: recall_at_1000
value: 98.17399999999999
- type: main_score
value: 77.929
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetCCRetrieval (php)
type: CoIR-Retrieval/CodeSearchNet-ccr
config: php
split: test
revision: 6e1effa2c03723c5fde48ee912b5ee08d4f211e8
metrics:
- type: ndcg_at_1
value: 52.71900000000001
- type: ndcg_at_3
value: 63.025
- type: ndcg_at_5
value: 65.17399999999999
- type: ndcg_at_10
value: 66.982
- type: ndcg_at_20
value: 68.113
- type: ndcg_at_100
value: 69.443
- type: ndcg_at_1000
value: 70.111
- type: recall_at_1
value: 52.71900000000001
- type: recall_at_3
value: 70.158
- type: recall_at_5
value: 75.35300000000001
- type: recall_at_10
value: 80.919
- type: recall_at_20
value: 85.36500000000001
- type: recall_at_100
value: 92.486
- type: recall_at_1000
value: 97.788
- type: main_score
value: 66.982
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetRetrieval (python)
type: code-search-net/code_search_net
config: python
split: test
revision: fdc6a9e39575768c27eb8a2a5f702bf846eb4759
metrics:
- type: ndcg_at_1
value: 86.9
- type: ndcg_at_3
value: 92.012
- type: ndcg_at_5
value: 93.002
- type: ndcg_at_10
value: 93.304
- type: ndcg_at_20
value: 93.432
- type: ndcg_at_100
value: 93.50500000000001
- type: ndcg_at_1000
value: 93.54
- type: recall_at_1
value: 86.9
- type: recall_at_3
value: 95.5
- type: recall_at_5
value: 97.89999999999999
- type: recall_at_10
value: 98.8
- type: recall_at_20
value: 99.3
- type: recall_at_100
value: 99.7
- type: recall_at_1000
value: 100.0
- type: main_score
value: 93.304
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetRetrieval (javascript)
type: code-search-net/code_search_net
config: javascript
split: test
revision: fdc6a9e39575768c27eb8a2a5f702bf846eb4759
metrics:
- type: ndcg_at_1
value: 73.9
- type: ndcg_at_3
value: 80.297
- type: ndcg_at_5
value: 81.162
- type: ndcg_at_10
value: 82.075
- type: ndcg_at_20
value: 82.432
- type: ndcg_at_100
value: 82.948
- type: ndcg_at_1000
value: 83.722
- type: recall_at_1
value: 73.9
- type: recall_at_3
value: 84.6
- type: recall_at_5
value: 86.7
- type: recall_at_10
value: 89.5
- type: recall_at_20
value: 90.9
- type: recall_at_100
value: 93.7
- type: recall_at_1000
value: 100.0
- type: main_score
value: 82.075
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetRetrieval (go)
type: code-search-net/code_search_net
config: go
split: test
revision: fdc6a9e39575768c27eb8a2a5f702bf846eb4759
metrics:
- type: ndcg_at_1
value: 86.9
- type: ndcg_at_3
value: 92.961
- type: ndcg_at_5
value: 93.632
- type: ndcg_at_10
value: 93.865
- type: ndcg_at_20
value: 93.917
- type: ndcg_at_100
value: 93.994
- type: ndcg_at_1000
value: 94.02199999999999
- type: recall_at_1
value: 86.9
- type: recall_at_3
value: 96.89999999999999
- type: recall_at_5
value: 98.5
- type: recall_at_10
value: 99.2
- type: recall_at_20
value: 99.4
- type: recall_at_100
value: 99.8
- type: recall_at_1000
value: 100.0
- type: main_score
value: 93.865
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetRetrieval (ruby)
type: code-search-net/code_search_net
config: ruby
split: test
revision: fdc6a9e39575768c27eb8a2a5f702bf846eb4759
metrics:
- type: ndcg_at_1
value: 79.10000000000001
- type: ndcg_at_3
value: 85.626
- type: ndcg_at_5
value: 86.629
- type: ndcg_at_10
value: 87.16000000000001
- type: ndcg_at_20
value: 87.414
- type: ndcg_at_100
value: 87.7
- type: ndcg_at_1000
value: 88.115
- type: recall_at_1
value: 79.10000000000001
- type: recall_at_3
value: 89.9
- type: recall_at_5
value: 92.30000000000001
- type: recall_at_10
value: 93.89999999999999
- type: recall_at_20
value: 94.89999999999999
- type: recall_at_100
value: 96.39999999999999
- type: recall_at_1000
value: 100.0
- type: main_score
value: 87.16000000000001
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetRetrieval (java)
type: code-search-net/code_search_net
config: java
split: test
revision: fdc6a9e39575768c27eb8a2a5f702bf846eb4759
metrics:
- type: ndcg_at_1
value: 82.0
- type: ndcg_at_3
value: 89.205
- type: ndcg_at_5
value: 89.86699999999999
- type: ndcg_at_10
value: 90.269
- type: ndcg_at_20
value: 90.32
- type: ndcg_at_100
value: 90.36999999999999
- type: ndcg_at_1000
value: 90.691
- type: recall_at_1
value: 82.0
- type: recall_at_3
value: 94.0
- type: recall_at_5
value: 95.6
- type: recall_at_10
value: 96.8
- type: recall_at_20
value: 97.0
- type: recall_at_100
value: 97.3
- type: recall_at_1000
value: 100.0
- type: main_score
value: 90.269
- task:
type: Retrieval
dataset:
name: MTEB CodeSearchNetRetrieval (php)
type: code-search-net/code_search_net
config: php
split: test
revision: fdc6a9e39575768c27eb8a2a5f702bf846eb4759
metrics:
- type: ndcg_at_1
value: 76.1
- type: ndcg_at_3
value: 83.97
- type: ndcg_at_5
value: 85.128
- type: ndcg_at_10
value: 85.922
- type: ndcg_at_20
value: 86.279
- type: ndcg_at_100
value: 86.53
- type: ndcg_at_1000
value: 86.846
- type: recall_at_1
value: 76.1
- type: recall_at_3
value: 89.3
- type: recall_at_5
value: 92.10000000000001
- type: recall_at_10
value: 94.5
- type: recall_at_20
value: 95.89999999999999
- type: recall_at_100
value: 97.3
- type: recall_at_1000
value: 100.0
- type: main_score
value: 85.922
- task:
type: Retrieval
dataset:
name: MTEB CodeTransOceanContest (default)
type: CoIR-Retrieval/codetrans-contest
config: default
split: test
revision: 20da4eb20a4b17300c0986ee148c90867a7f2a4d
metrics:
- type: ndcg_at_1
value: 82.353
- type: ndcg_at_3
value: 86.792
- type: ndcg_at_5
value: 88.116
- type: ndcg_at_10
value: 89.164
- type: ndcg_at_20
value: 89.627
- type: ndcg_at_100
value: 89.816
- type: ndcg_at_1000
value: 89.929
- type: recall_at_1
value: 82.353
- type: recall_at_3
value: 90.045
- type: recall_at_5
value: 93.21300000000001
- type: recall_at_10
value: 96.38
- type: recall_at_20
value: 98.19
- type: recall_at_100
value: 99.095
- type: recall_at_1000
value: 100.0
- type: main_score
value: 89.164
- task:
type: Retrieval
dataset:
name: MTEB CodeTransOceanDL (default)
type: CoIR-Retrieval/codetrans-dl
config: default
split: test
revision: 281562cb8a1265ab5c0824bfa6ddcd9b0a15618f
metrics:
- type: ndcg_at_1
value: 9.443999999999999
- type: ndcg_at_3
value: 13.141
- type: ndcg_at_5
value: 20.149
- type: ndcg_at_10
value: 35.181000000000004
- type: ndcg_at_20
value: 39.898
- type: ndcg_at_100
value: 40.337
- type: ndcg_at_1000
value: 40.337
- type: recall_at_1
value: 9.443999999999999
- type: recall_at_3
value: 16.111
- type: recall_at_5
value: 32.778
- type: recall_at_10
value: 80.55600000000001
- type: recall_at_20
value: 97.77799999999999
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 35.181000000000004
- task:
type: Retrieval
dataset:
name: MTEB CosQA (default)
type: CoIR-Retrieval/cosqa
config: default
split: test
revision: bc5efb7e9d437246ce393ed19d772e08e4a79535
metrics:
- type: ndcg_at_1
value: 14.2
- type: ndcg_at_3
value: 23.647000000000002
- type: ndcg_at_5
value: 28.655
- type: ndcg_at_10
value: 34.175
- type: ndcg_at_20
value: 37.04
- type: ndcg_at_100
value: 41.074
- type: ndcg_at_1000
value: 41.917
- type: recall_at_1
value: 14.2
- type: recall_at_3
value: 31.0
- type: recall_at_5
value: 43.4
- type: recall_at_10
value: 60.4
- type: recall_at_20
value: 71.8
- type: recall_at_100
value: 93.0
- type: recall_at_1000
value: 99.2
- type: main_score
value: 34.175
- task:
type: Retrieval
dataset:
name: MTEB DBPedia (default)
type: mteb/dbpedia
config: default
split: test
revision: c0f706b76e590d620bd6618b3ca8efdd34e2d659
metrics:
- type: ndcg_at_1
value: 61.5
- type: ndcg_at_3
value: 53.476
- type: ndcg_at_5
value: 51.601
- type: ndcg_at_10
value: 50.391
- type: ndcg_at_20
value: 49.342000000000006
- type: ndcg_at_100
value: 55.37800000000001
- type: ndcg_at_1000
value: 62.470000000000006
- type: recall_at_1
value: 9.757
- type: recall_at_3
value: 17.203
- type: recall_at_5
value: 21.878
- type: recall_at_10
value: 30.425
- type: recall_at_20
value: 39.137
- type: recall_at_100
value: 62.885000000000005
- type: recall_at_1000
value: 85.795
- type: main_score
value: 50.391
- task:
type: Retrieval
dataset:
name: MTEB FiQA-PL (default)
type: clarin-knext/fiqa-pl
config: default
split: test
revision: 2e535829717f8bf9dc829b7f911cc5bbd4e6608e
metrics:
- type: ndcg_at_1
value: 46.296
- type: ndcg_at_3
value: 43.682
- type: ndcg_at_5
value: 44.818999999999996
- type: ndcg_at_10
value: 47.137
- type: ndcg_at_20
value: 49.957
- type: ndcg_at_100
value: 53.998999999999995
- type: ndcg_at_1000
value: 56.547000000000004
- type: recall_at_1
value: 23.116999999999997
- type: recall_at_3
value: 39.967000000000006
- type: recall_at_5
value: 46.745
- type: recall_at_10
value: 54.202
- type: recall_at_20
value: 62.61600000000001
- type: recall_at_100
value: 79.322
- type: recall_at_1000
value: 94.114
- type: main_score
value: 47.137
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018 (default)
type: mteb/fiqa
config: default
split: test
revision: 27a168819829fe9bcd655c2df245fb19452e8e06
metrics:
- type: ndcg_at_1
value: 63.117000000000004
- type: ndcg_at_3
value: 58.538999999999994
- type: ndcg_at_5
value: 59.147000000000006
- type: ndcg_at_10
value: 62.35000000000001
- type: ndcg_at_20
value: 65.36800000000001
- type: ndcg_at_100
value: 68.801
- type: ndcg_at_1000
value: 70.06599999999999
- type: recall_at_1
value: 33.377
- type: recall_at_3
value: 52.817
- type: recall_at_5
value: 59.03699999999999
- type: recall_at_10
value: 69.116
- type: recall_at_20
value: 78.30799999999999
- type: recall_at_100
value: 91.715
- type: recall_at_1000
value: 98.783
- type: main_score
value: 62.35000000000001
- task:
type: Retrieval
dataset:
name: MTEB GerDaLIRSmall (default)
type: mteb/GerDaLIRSmall
config: default
split: test
revision: 48327de6ee192e9610f3069789719788957c7abd
metrics:
- type: ndcg_at_1
value: 30.047
- type: ndcg_at_3
value: 36.635
- type: ndcg_at_5
value: 39.237
- type: ndcg_at_10
value: 41.752
- type: ndcg_at_20
value: 43.467
- type: ndcg_at_100
value: 45.793
- type: ndcg_at_1000
value: 47.404
- type: recall_at_1
value: 27.272999999999996
- type: recall_at_3
value: 41.534
- type: recall_at_5
value: 47.678
- type: recall_at_10
value: 55.131
- type: recall_at_20
value: 61.592
- type: recall_at_100
value: 73.604
- type: recall_at_1000
value: 86.146
- type: main_score
value: 41.752
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA (default)
type: mteb/hotpotqa
config: default
split: test
revision: ab518f4d6fcca38d87c25209f94beba119d02014
metrics:
- type: ndcg_at_1
value: 88.062
- type: ndcg_at_3
value: 77.443
- type: ndcg_at_5
value: 80.05600000000001
- type: ndcg_at_10
value: 81.979
- type: ndcg_at_20
value: 83.033
- type: ndcg_at_100
value: 84.232
- type: ndcg_at_1000
value: 84.827
- type: recall_at_1
value: 44.031
- type: recall_at_3
value: 75.71900000000001
- type: recall_at_5
value: 80.851
- type: recall_at_10
value: 85.652
- type: recall_at_20
value: 89.021
- type: recall_at_100
value: 94.267
- type: recall_at_1000
value: 98.136
- type: main_score
value: 81.979
- task:
type: Retrieval
dataset:
name: MTEB LEMBNarrativeQARetrieval (default)
type: dwzhu/LongEmbed
config: default
split: test
revision: 6e346642246bfb4928c560ee08640dc84d074e8c
metrics:
- type: ndcg_at_1
value: 49.009
- type: ndcg_at_3
value: 56.69
- type: ndcg_at_5
value: 58.572
- type: ndcg_at_10
value: 60.702
- type: ndcg_at_20
value: 62.160000000000004
- type: ndcg_at_100
value: 64.461
- type: ndcg_at_1000
value: 65.604
- type: recall_at_1
value: 49.009
- type: recall_at_3
value: 62.073
- type: recall_at_5
value: 66.648
- type: recall_at_10
value: 73.222
- type: recall_at_20
value: 78.974
- type: recall_at_100
value: 91.444
- type: recall_at_1000
value: 100.0
- type: main_score
value: 60.702
- type: ndcg_at_1
value: 37.263000000000005
- type: ndcg_at_3
value: 48.207
- type: ndcg_at_5
value: 51.464
- type: ndcg_at_10
value: 55.071999999999996
- type: ndcg_at_20
value: 57.364000000000004
- type: ndcg_at_100
value: 60.236999999999995
- type: ndcg_at_1000
value: 60.352
- type: recall_at_1
value: 37.263000000000005
- type: recall_at_3
value: 55.92700000000001
- type: recall_at_5
value: 63.851
- type: recall_at_10
value: 74.91799999999999
- type: recall_at_20
value: 83.955
- type: recall_at_100
value: 99.214
- type: recall_at_1000
value: 100.0
- type: main_score
value: 55.071999999999996
- type: ndcg_at_1
value: 80.0
- type: ndcg_at_3
value: 84.024
- type: ndcg_at_5
value: 84.985
- type: ndcg_at_10
value: 85.751
- type: ndcg_at_20
value: 86.634
- type: ndcg_at_100
value: 87.348
- type: ndcg_at_1000
value: 87.48599999999999
- type: recall_at_1
value: 80.0
- type: recall_at_3
value: 87.0
- type: recall_at_5
value: 89.333
- type: recall_at_10
value: 91.667
- type: recall_at_20
value: 95.0
- type: recall_at_100
value: 99.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 85.751
- task:
type: Retrieval
dataset:
name: MTEB LEMBNeedleRetrieval (default)
type: dwzhu/LongEmbed
config: default
split: test_256
revision: 6e346642246bfb4928c560ee08640dc84d074e8c
metrics:
- type: ndcg_at_1
value: 8.0
- type: ndcg_at_3
value: 12.786
- type: ndcg_at_5
value: 15.282000000000002
- type: ndcg_at_10
value: 20.096
- type: ndcg_at_20
value: 22.631
- type: ndcg_at_100
value: 32.174
- type: ndcg_at_1000
value: 32.174
- type: recall_at_1
value: 8.0
- type: recall_at_3
value: 16.0
- type: recall_at_5
value: 22.0
- type: recall_at_10
value: 36.0
- type: recall_at_20
value: 46.0
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 8.0
- type: ndcg_at_1
value: 10.0
- type: ndcg_at_3
value: 12.0
- type: ndcg_at_5
value: 12.0
- type: ndcg_at_10
value: 12.631
- type: ndcg_at_20
value: 14.982000000000001
- type: ndcg_at_100
value: 28.534
- type: ndcg_at_1000
value: 28.534
- type: recall_at_1
value: 10.0
- type: recall_at_3
value: 14.000000000000002
- type: recall_at_5
value: 14.000000000000002
- type: recall_at_10
value: 16.0
- type: recall_at_20
value: 26.0
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 10.0
- task:
type: Retrieval
dataset:
name: MTEB LEMBSummScreenFDRetrieval (default)
type: dwzhu/LongEmbed
config: default
split: validation
revision: 6e346642246bfb4928c560ee08640dc84d074e8c
metrics:
- type: ndcg_at_1
value: 94.345
- type: ndcg_at_3
value: 96.66900000000001
- type: ndcg_at_5
value: 97.297
- type: ndcg_at_10
value: 97.387
- type: ndcg_at_20
value: 97.387
- type: ndcg_at_100
value: 97.387
- type: ndcg_at_1000
value: 97.387
- type: recall_at_1
value: 94.345
- type: recall_at_3
value: 98.214
- type: recall_at_5
value: 99.702
- type: recall_at_10
value: 100.0
- type: recall_at_20
value: 100.0
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 97.387
- task:
type: Retrieval
dataset:
name: MTEB LeCaRDv2 (default)
type: mteb/LeCaRDv2
config: default
split: test
revision: b78e18688c3d012a33dc3676597c1d1b2243ce1c
metrics:
- type: ndcg_at_1
value: 87.421
- type: ndcg_at_3
value: 83.159
- type: ndcg_at_5
value: 79.818
- type: ndcg_at_10
value: 74.168
- type: ndcg_at_20
value: 67.81
- type: ndcg_at_100
value: 80.432
- type: ndcg_at_1000
value: 84.423
- type: recall_at_1
value: 4.1450000000000005
- type: recall_at_3
value: 10.988000000000001
- type: recall_at_5
value: 16.808999999999997
- type: recall_at_10
value: 29.329
- type: recall_at_20
value: 48.425000000000004
- type: recall_at_100
value: 89.63600000000001
- type: recall_at_1000
value: 99.823
- type: main_score
value: 74.168
- task:
type: Retrieval
dataset:
name: MTEB LegalBenchConsumerContractsQA (default)
type: mteb/legalbench_consumer_contracts_qa
config: default
split: test
revision: b23590301ec94e8087e2850b21d43d4956b1cca9
metrics:
- type: ndcg_at_1
value: 73.485
- type: ndcg_at_3
value: 81.977
- type: ndcg_at_5
value: 84.63000000000001
- type: ndcg_at_10
value: 85.444
- type: ndcg_at_20
value: 86.008
- type: ndcg_at_100
value: 86.262
- type: ndcg_at_1000
value: 86.262
- type: recall_at_1
value: 73.485
- type: recall_at_3
value: 87.626
- type: recall_at_5
value: 93.939
- type: recall_at_10
value: 96.465
- type: recall_at_20
value: 98.737
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 85.444
- task:
type: Retrieval
dataset:
name: MTEB LegalBenchCorporateLobbying (default)
type: mteb/legalbench_corporate_lobbying
config: default
split: test
revision: f69691c650464e62546d7f2a4536f8f87c891e38
metrics:
- type: ndcg_at_1
value: 91.471
- type: ndcg_at_3
value: 95.84700000000001
- type: ndcg_at_5
value: 96.088
- type: ndcg_at_10
value: 96.17999999999999
- type: ndcg_at_20
value: 96.17999999999999
- type: ndcg_at_100
value: 96.17999999999999
- type: ndcg_at_1000
value: 96.259
- type: recall_at_1
value: 91.471
- type: recall_at_3
value: 98.529
- type: recall_at_5
value: 99.118
- type: recall_at_10
value: 99.412
- type: recall_at_20
value: 99.412
- type: recall_at_100
value: 99.412
- type: recall_at_1000
value: 100.0
- type: main_score
value: 96.17999999999999
- task:
type: Retrieval
dataset:
name: MTEB LegalQuAD (default)
type: mteb/LegalQuAD
config: default
split: test
revision: 37aa6cfb01d48960b0f8e3f17d6e3d99bf1ebc3e
metrics:
- type: ndcg_at_1
value: 48.0
- type: ndcg_at_3
value: 59.397999999999996
- type: ndcg_at_5
value: 61.05500000000001
- type: ndcg_at_10
value: 63.219
- type: ndcg_at_20
value: 65.102
- type: ndcg_at_100
value: 67.254
- type: ndcg_at_1000
value: 67.746
- type: recall_at_1
value: 48.0
- type: recall_at_3
value: 67.0
- type: recall_at_5
value: 71.0
- type: recall_at_10
value: 77.5
- type: recall_at_20
value: 85.0
- type: recall_at_100
value: 96.5
- type: recall_at_1000
value: 100.0
- type: main_score
value: 63.219
- task:
type: Retrieval
dataset:
name: MTEB LegalSummarization (default)
type: mteb/legal_summarization
config: default
split: test
revision: 3bb1a05c66872889662af04c5691c14489cebd72
metrics:
- type: ndcg_at_1
value: 58.451
- type: ndcg_at_3
value: 63.70099999999999
- type: ndcg_at_5
value: 66.792
- type: ndcg_at_10
value: 69.76
- type: ndcg_at_20
value: 71.487
- type: ndcg_at_100
value: 73.6
- type: ndcg_at_1000
value: 74.05000000000001
- type: recall_at_1
value: 52.028
- type: recall_at_3
value: 66.7
- type: recall_at_5
value: 74.119
- type: recall_at_10
value: 82.595
- type: recall_at_20
value: 88.209
- type: recall_at_100
value: 97.24000000000001
- type: recall_at_1000
value: 100.0
- type: main_score
value: 69.76
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (ar)
type: jinaai/mintakaqa
config: ar
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 19.791
- type: ndcg_at_3
value: 29.751
- type: ndcg_at_5
value: 32.83
- type: ndcg_at_10
value: 35.553000000000004
- type: ndcg_at_20
value: 37.528
- type: ndcg_at_100
value: 40.025
- type: ndcg_at_1000
value: 42.693
- type: recall_at_1
value: 19.791
- type: recall_at_3
value: 36.632
- type: recall_at_5
value: 44.076
- type: recall_at_10
value: 52.474
- type: recall_at_20
value: 60.281
- type: recall_at_100
value: 73.94500000000001
- type: recall_at_1000
value: 96.096
- type: main_score
value: 35.553000000000004
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (de)
type: jinaai/mintakaqa
config: de
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 34.288000000000004
- type: ndcg_at_3
value: 47.29
- type: ndcg_at_5
value: 50.622
- type: ndcg_at_10
value: 53.291999999999994
- type: ndcg_at_20
value: 55.062999999999995
- type: ndcg_at_100
value: 56.987
- type: ndcg_at_1000
value: 58.084
- type: recall_at_1
value: 34.288000000000004
- type: recall_at_3
value: 56.486999999999995
- type: recall_at_5
value: 64.532
- type: recall_at_10
value: 72.746
- type: recall_at_20
value: 79.697
- type: recall_at_100
value: 90.185
- type: recall_at_1000
value: 98.989
- type: main_score
value: 53.291999999999994
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (es)
type: jinaai/mintakaqa
config: es
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 31.889
- type: ndcg_at_3
value: 45.182
- type: ndcg_at_5
value: 48.475
- type: ndcg_at_10
value: 51.402
- type: ndcg_at_20
value: 53.089
- type: ndcg_at_100
value: 55.116
- type: ndcg_at_1000
value: 56.333999999999996
- type: recall_at_1
value: 31.889
- type: recall_at_3
value: 54.455
- type: recall_at_5
value: 62.417
- type: recall_at_10
value: 71.328
- type: recall_at_20
value: 77.97
- type: recall_at_100
value: 88.944
- type: recall_at_1000
value: 98.639
- type: main_score
value: 51.402
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (fr)
type: jinaai/mintakaqa
config: fr
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 32.555
- type: ndcg_at_3
value: 45.278
- type: ndcg_at_5
value: 48.559000000000005
- type: ndcg_at_10
value: 51.485
- type: ndcg_at_20
value: 53.263000000000005
- type: ndcg_at_100
value: 55.221
- type: ndcg_at_1000
value: 56.501999999999995
- type: recall_at_1
value: 32.555
- type: recall_at_3
value: 54.054
- type: recall_at_5
value: 62.039
- type: recall_at_10
value: 70.966
- type: recall_at_20
value: 77.969
- type: recall_at_100
value: 88.411
- type: recall_at_1000
value: 98.69
- type: main_score
value: 51.485
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (hi)
type: jinaai/mintakaqa
config: hi
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 24.757
- type: ndcg_at_3
value: 35.427
- type: ndcg_at_5
value: 38.431
- type: ndcg_at_10
value: 41.459
- type: ndcg_at_20
value: 44.137
- type: ndcg_at_100
value: 47.174
- type: ndcg_at_1000
value: 48.907000000000004
- type: recall_at_1
value: 24.757
- type: recall_at_3
value: 43.082
- type: recall_at_5
value: 50.336999999999996
- type: recall_at_10
value: 59.611000000000004
- type: recall_at_20
value: 70.157
- type: recall_at_100
value: 86.387
- type: recall_at_1000
value: 100.0
- type: main_score
value: 41.459
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (it)
type: jinaai/mintakaqa
config: it
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 32.818000000000005
- type: ndcg_at_3
value: 46.503
- type: ndcg_at_5
value: 49.68
- type: ndcg_at_10
value: 52.510999999999996
- type: ndcg_at_20
value: 54.269999999999996
- type: ndcg_at_100
value: 56.17100000000001
- type: ndcg_at_1000
value: 57.38100000000001
- type: recall_at_1
value: 32.818000000000005
- type: recall_at_3
value: 56.033
- type: recall_at_5
value: 63.715999999999994
- type: recall_at_10
value: 72.48400000000001
- type: recall_at_20
value: 79.374
- type: recall_at_100
value: 89.436
- type: recall_at_1000
value: 98.914
- type: main_score
value: 52.510999999999996
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (ja)
type: jinaai/mintakaqa
config: ja
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 25.216
- type: ndcg_at_3
value: 35.982
- type: ndcg_at_5
value: 38.694
- type: ndcg_at_10
value: 41.585
- type: ndcg_at_20
value: 43.334
- type: ndcg_at_100
value: 45.831
- type: ndcg_at_1000
value: 48.06
- type: recall_at_1
value: 25.216
- type: recall_at_3
value: 43.599
- type: recall_at_5
value: 50.173
- type: recall_at_10
value: 59.083
- type: recall_at_20
value: 65.96
- type: recall_at_100
value: 79.542
- type: recall_at_1000
value: 97.794
- type: main_score
value: 41.585
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (pt)
type: jinaai/mintakaqa
config: pt
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 33.517
- type: ndcg_at_3
value: 46.955999999999996
- type: ndcg_at_5
value: 50.441
- type: ndcg_at_10
value: 53.256
- type: ndcg_at_20
value: 55.086
- type: ndcg_at_100
value: 57.104
- type: ndcg_at_1000
value: 58.07600000000001
- type: recall_at_1
value: 33.517
- type: recall_at_3
value: 56.245
- type: recall_at_5
value: 64.63499999999999
- type: recall_at_10
value: 73.258
- type: recall_at_20
value: 80.47999999999999
- type: recall_at_100
value: 91.27
- type: recall_at_1000
value: 99.10799999999999
- type: main_score
value: 53.256
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus (default)
type: mteb/nfcorpus
config: default
split: test
revision: ec0fa4fe99da2ff19ca1214b7966684033a58814
metrics:
- type: ndcg_at_1
value: 51.702999999999996
- type: ndcg_at_3
value: 48.064
- type: ndcg_at_5
value: 46.379
- type: ndcg_at_10
value: 43.663999999999994
- type: ndcg_at_20
value: 41.407
- type: ndcg_at_100
value: 42.083
- type: ndcg_at_1000
value: 52.335
- type: recall_at_1
value: 6.241
- type: recall_at_3
value: 12.214
- type: recall_at_5
value: 16.473
- type: recall_at_10
value: 21.84
- type: recall_at_20
value: 27.474999999999998
- type: recall_at_100
value: 45.01
- type: recall_at_1000
value: 80.71300000000001
- type: main_score
value: 43.663999999999994
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus-PL (default)
type: clarin-knext/nfcorpus-pl
config: default
split: test
revision: 9a6f9567fda928260afed2de480d79c98bf0bec0
metrics:
- type: ndcg_at_1
value: 41.641
- type: ndcg_at_3
value: 37.617
- type: ndcg_at_5
value: 36.024
- type: ndcg_at_10
value: 33.51
- type: ndcg_at_20
value: 31.575999999999997
- type: ndcg_at_100
value: 31.601000000000003
- type: ndcg_at_1000
value: 41.099000000000004
- type: recall_at_1
value: 4.61
- type: recall_at_3
value: 9.366
- type: recall_at_5
value: 11.793
- type: recall_at_10
value: 16.255
- type: recall_at_20
value: 20.713
- type: recall_at_100
value: 33.396
- type: recall_at_1000
value: 65.532
- type: main_score
value: 33.51
- task:
type: Retrieval
dataset:
name: MTEB NQ (default)
type: mteb/nq
config: default
split: test
revision: b774495ed302d8c44a3a7ea25c90dbce03968f31
metrics:
- type: ndcg_at_1
value: 50.753
- type: ndcg_at_3
value: 62.541000000000004
- type: ndcg_at_5
value: 66.46600000000001
- type: ndcg_at_10
value: 69.65400000000001
- type: ndcg_at_20
value: 70.91499999999999
- type: ndcg_at_100
value: 71.908
- type: ndcg_at_1000
value: 72.08200000000001
- type: recall_at_1
value: 45.293
- type: recall_at_3
value: 71.089
- type: recall_at_5
value: 79.93
- type: recall_at_10
value: 89.01599999999999
- type: recall_at_20
value: 93.60300000000001
- type: recall_at_100
value: 98.501
- type: recall_at_1000
value: 99.768
- type: main_score
value: 69.65400000000001
- task:
type: Retrieval
dataset:
name: MTEB NQ-PL (default)
type: clarin-knext/nq-pl
config: default
split: test
revision: f171245712cf85dd4700b06bef18001578d0ca8d
metrics:
- type: ndcg_at_1
value: 34.791
- type: ndcg_at_3
value: 45.418
- type: ndcg_at_5
value: 49.486000000000004
- type: ndcg_at_10
value: 53.141000000000005
- type: ndcg_at_20
value: 55.230999999999995
- type: ndcg_at_100
value: 57.358
- type: ndcg_at_1000
value: 58.166
- type: recall_at_1
value: 31.04
- type: recall_at_3
value: 53.179
- type: recall_at_5
value: 62.539
- type: recall_at_10
value: 73.08099999999999
- type: recall_at_20
value: 80.83500000000001
- type: recall_at_100
value: 91.503
- type: recall_at_1000
value: 97.429
- type: main_score
value: 53.141000000000005
- task:
type: Retrieval
dataset:
name: MTEB Quora-PL (default)
type: clarin-knext/quora-pl
config: default
split: validation
revision: 0be27e93455051e531182b85e85e425aba12e9d4
metrics:
- type: ndcg_at_1
value: 76.99000000000001
- type: ndcg_at_3
value: 81.781
- type: ndcg_at_5
value: 83.627
- type: ndcg_at_10
value: 85.146
- type: ndcg_at_20
value: 86.015
- type: ndcg_at_100
value: 86.745
- type: ndcg_at_1000
value: 86.882
- type: recall_at_1
value: 66.806
- type: recall_at_3
value: 84.09400000000001
- type: recall_at_5
value: 89.09899999999999
- type: recall_at_10
value: 93.512
- type: recall_at_20
value: 96.365
- type: recall_at_100
value: 99.22
- type: recall_at_1000
value: 99.937
- type: main_score
value: 85.146
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval (default)
type: mteb/quora
config: default
split: test
revision: e4e08e0b7dbe3c8700f0daef558ff32256715259
metrics:
- type: ndcg_at_1
value: 83.66
- type: ndcg_at_3
value: 87.863
- type: ndcg_at_5
value: 89.279
- type: ndcg_at_10
value: 90.372
- type: ndcg_at_20
value: 90.955
- type: ndcg_at_100
value: 91.352
- type: ndcg_at_1000
value: 91.39500000000001
- type: recall_at_1
value: 72.75399999999999
- type: recall_at_3
value: 89.41799999999999
- type: recall_at_5
value: 93.509
- type: recall_at_10
value: 96.679
- type: recall_at_20
value: 98.519
- type: recall_at_100
value: 99.845
- type: recall_at_1000
value: 99.998
- type: main_score
value: 90.372
- task:
type: Retrieval
dataset:
name: MTEB RiaNewsRetrieval (default)
type: ai-forever/ria-news-retrieval
config: default
split: test
revision: 82374b0bbacda6114f39ff9c5b925fa1512ca5d7
metrics:
- type: ndcg_at_1
value: 75.41
- type: ndcg_at_3
value: 83.13000000000001
- type: ndcg_at_5
value: 84.313
- type: ndcg_at_10
value: 85.009
- type: ndcg_at_20
value: 85.436
- type: ndcg_at_100
value: 85.875
- type: ndcg_at_1000
value: 86.048
- type: recall_at_1
value: 75.41
- type: recall_at_3
value: 88.38000000000001
- type: recall_at_5
value: 91.23
- type: recall_at_10
value: 93.34
- type: recall_at_20
value: 95.02000000000001
- type: recall_at_100
value: 97.37
- type: recall_at_1000
value: 98.78
- type: main_score
value: 85.009
- task:
type: Retrieval
dataset:
name: MTEB RuBQRetrieval (default)
type: ai-forever/rubq-retrieval
config: default
split: test
revision: e19b6ffa60b3bc248e0b41f4cc37c26a55c2a67b
metrics:
- type: ndcg_at_1
value: 63.652
- type: ndcg_at_3
value: 67.829
- type: ndcg_at_5
value: 72.141
- type: ndcg_at_10
value: 75.551
- type: ndcg_at_20
value: 76.925
- type: ndcg_at_100
value: 77.813
- type: ndcg_at_1000
value: 77.994
- type: recall_at_1
value: 45.09
- type: recall_at_3
value: 71.562
- type: recall_at_5
value: 81.474
- type: recall_at_10
value: 90.237
- type: recall_at_20
value: 94.679
- type: recall_at_100
value: 98.752
- type: recall_at_1000
value: 99.83999999999999
- type: main_score
value: 75.551
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS (default)
type: mteb/scidocs
config: default
split: test
revision: f8c2fcf00f625baaa80f62ec5bd9e1fff3b8ae88
metrics:
- type: ndcg_at_1
value: 33.7
- type: ndcg_at_3
value: 28.360999999999997
- type: ndcg_at_5
value: 25.259999999999998
- type: ndcg_at_10
value: 30.775999999999996
- type: ndcg_at_20
value: 34.782000000000004
- type: ndcg_at_100
value: 41.753
- type: ndcg_at_1000
value: 46.887
- type: recall_at_1
value: 6.843000000000001
- type: recall_at_3
value: 16.228
- type: recall_at_5
value: 22.828
- type: recall_at_10
value: 33.007
- type: recall_at_20
value: 42.433
- type: recall_at_100
value: 64.967
- type: recall_at_1000
value: 89.587
- type: main_score
value: 30.775999999999996
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS-PL (default)
type: clarin-knext/scidocs-pl
config: default
split: test
revision: 45452b03f05560207ef19149545f168e596c9337
metrics:
- type: ndcg_at_1
value: 26.5
- type: ndcg_at_3
value: 21.079
- type: ndcg_at_5
value: 18.63
- type: ndcg_at_10
value: 22.483
- type: ndcg_at_20
value: 25.552999999999997
- type: ndcg_at_100
value: 31.572
- type: ndcg_at_1000
value: 37.147000000000006
- type: recall_at_1
value: 5.367999999999999
- type: recall_at_3
value: 11.907
- type: recall_at_5
value: 16.631999999999998
- type: recall_at_10
value: 23.647000000000002
- type: recall_at_20
value: 30.857
- type: recall_at_100
value: 50.236999999999995
- type: recall_at_1000
value: 77.445
- type: main_score
value: 22.483
- task:
type: Retrieval
dataset:
name: MTEB SciFact (default)
type: mteb/scifact
config: default
split: test
revision: 0228b52cf27578f30900b9e5271d331663a030d7
metrics:
- type: ndcg_at_1
value: 74.333
- type: ndcg_at_3
value: 82.071
- type: ndcg_at_5
value: 83.83800000000001
- type: ndcg_at_10
value: 85.399
- type: ndcg_at_20
value: 85.57900000000001
- type: ndcg_at_100
value: 86.075
- type: ndcg_at_1000
value: 86.164
- type: recall_at_1
value: 70.994
- type: recall_at_3
value: 87.417
- type: recall_at_5
value: 91.89399999999999
- type: recall_at_10
value: 96.167
- type: recall_at_20
value: 96.833
- type: recall_at_100
value: 99.333
- type: recall_at_1000
value: 100.0
- type: main_score
value: 85.399
- task:
type: Retrieval
dataset:
name: MTEB SciFact-PL (default)
type: clarin-knext/scifact-pl
config: default
split: test
revision: 47932a35f045ef8ed01ba82bf9ff67f6e109207e
metrics:
- type: ndcg_at_1
value: 65.333
- type: ndcg_at_3
value: 73.291
- type: ndcg_at_5
value: 75.149
- type: ndcg_at_10
value: 77.633
- type: ndcg_at_20
value: 78.236
- type: ndcg_at_100
value: 79.182
- type: ndcg_at_1000
value: 79.431
- type: recall_at_1
value: 61.99400000000001
- type: recall_at_3
value: 79.01700000000001
- type: recall_at_5
value: 83.72800000000001
- type: recall_at_10
value: 90.72200000000001
- type: recall_at_20
value: 93.0
- type: recall_at_100
value: 98.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 77.633
- task:
type: Retrieval
dataset:
name: MTEB StackOverflowQA (default)
type: CoIR-Retrieval/stackoverflow-qa
config: default
split: test
revision: db8f169f3894c14a00251061f957b2063eef2bd5
metrics:
- type: ndcg_at_1
value: 90.07
- type: ndcg_at_3
value: 93.30199999999999
- type: ndcg_at_5
value: 93.812
- type: ndcg_at_10
value: 94.219
- type: ndcg_at_20
value: 94.46799999999999
- type: ndcg_at_100
value: 94.581
- type: ndcg_at_1000
value: 94.626
- type: recall_at_1
value: 90.07
- type: recall_at_3
value: 95.537
- type: recall_at_5
value: 96.78999999999999
- type: recall_at_10
value: 98.044
- type: recall_at_20
value: 99.047
- type: recall_at_100
value: 99.649
- type: recall_at_1000
value: 100.0
- type: main_score
value: 94.219
- task:
type: Retrieval
dataset:
name: MTEB SyntecRetrieval (default)
type: lyon-nlp/mteb-fr-retrieval-syntec-s2p
config: default
split: test
revision: 19661ccdca4dfc2d15122d776b61685f48c68ca9
metrics:
- type: ndcg_at_1
value: 83.0
- type: ndcg_at_3
value: 90.809
- type: ndcg_at_5
value: 91.583
- type: ndcg_at_10
value: 92.199
- type: ndcg_at_20
value: 92.199
- type: ndcg_at_100
value: 92.199
- type: ndcg_at_1000
value: 92.199
- type: recall_at_1
value: 83.0
- type: recall_at_3
value: 96.0
- type: recall_at_5
value: 98.0
- type: recall_at_10
value: 100.0
- type: recall_at_20
value: 100.0
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 92.199
- task:
type: Retrieval
dataset:
name: MTEB SyntheticText2SQL (default)
type: CoIR-Retrieval/synthetic-text2sql
config: default
split: test
revision: 686b87296c3a0191b5d9415a00526c62db9fce09
metrics:
- type: ndcg_at_1
value: 20.526
- type: ndcg_at_3
value: 60.12
- type: ndcg_at_5
value: 62.134
- type: ndcg_at_10
value: 63.50599999999999
- type: ndcg_at_20
value: 64.167
- type: ndcg_at_100
value: 64.687
- type: ndcg_at_1000
value: 64.801
- type: recall_at_1
value: 20.526
- type: recall_at_3
value: 84.721
- type: recall_at_5
value: 89.574
- type: recall_at_10
value: 93.762
- type: recall_at_20
value: 96.36
- type: recall_at_100
value: 99.09400000000001
- type: recall_at_1000
value: 99.966
- type: main_score
value: 63.50599999999999
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID (default)
type: mteb/trec-covid
config: default
split: test
revision: bb9466bac8153a0349341eb1b22e06409e78ef4e
metrics:
- type: ndcg_at_1
value: 76.0
- type: ndcg_at_3
value: 78.899
- type: ndcg_at_5
value: 78.212
- type: ndcg_at_10
value: 75.09700000000001
- type: ndcg_at_20
value: 72.158
- type: ndcg_at_100
value: 58.465999999999994
- type: ndcg_at_1000
value: 53.702000000000005
- type: recall_at_1
value: 0.231
- type: recall_at_3
value: 0.7000000000000001
- type: recall_at_5
value: 1.146
- type: recall_at_10
value: 2.174
- type: recall_at_20
value: 4.031
- type: recall_at_100
value: 14.713999999999999
- type: recall_at_1000
value: 50.8
- type: main_score
value: 75.09700000000001
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID-PL (default)
type: clarin-knext/trec-covid-pl
config: default
split: test
revision: 81bcb408f33366c2a20ac54adafad1ae7e877fdd
metrics:
- type: ndcg_at_1
value: 75.0
- type: ndcg_at_3
value: 75.531
- type: ndcg_at_5
value: 75.327
- type: ndcg_at_10
value: 74.28
- type: ndcg_at_20
value: 71.5
- type: ndcg_at_100
value: 58.412
- type: ndcg_at_1000
value: 52.580000000000005
- type: recall_at_1
value: 0.214
- type: recall_at_3
value: 0.647
- type: recall_at_5
value: 1.083
- type: recall_at_10
value: 2.141
- type: recall_at_20
value: 3.9309999999999996
- type: recall_at_100
value: 14.738999999999999
- type: recall_at_1000
value: 49.494
- type: main_score
value: 74.28
- task:
type: Retrieval
dataset:
name: MTEB Touche2020 (default)
type: mteb/touche2020
config: default
split: test
revision: a34f9a33db75fa0cbb21bb5cfc3dae8dc8bec93f
metrics:
- type: ndcg_at_1
value: 20.408
- type: ndcg_at_3
value: 23.368
- type: ndcg_at_5
value: 24.795
- type: ndcg_at_10
value: 24.442
- type: ndcg_at_20
value: 26.712000000000003
- type: ndcg_at_100
value: 38.218999999999994
- type: ndcg_at_1000
value: 50.395
- type: recall_at_1
value: 2.414
- type: recall_at_3
value: 6.3549999999999995
- type: recall_at_5
value: 9.888
- type: recall_at_10
value: 16.31
- type: recall_at_20
value: 25.369000000000003
- type: recall_at_100
value: 51.449999999999996
- type: recall_at_1000
value: 88.532
- type: main_score
value: 24.442
- task:
type: Retrieval
dataset:
name: MTEB ARCChallenge (default)
type: RAR-b/ARC-Challenge
config: default
split: test
revision: c481e0da3dcbbad8bce7721dea9085b74320a0a3
metrics:
- type: ndcg_at_1
value: 8.959
- type: ndcg_at_3
value: 16.238
- type: ndcg_at_5
value: 18.841
- type: ndcg_at_10
value: 21.606
- type: ndcg_at_20
value: 24.326
- type: ndcg_at_100
value: 28.410999999999998
- type: ndcg_at_1000
value: 31.279
- type: recall_at_1
value: 8.959
- type: recall_at_3
value: 21.416
- type: recall_at_5
value: 27.73
- type: recall_at_10
value: 36.348
- type: recall_at_20
value: 47.184
- type: recall_at_100
value: 69.539
- type: recall_at_1000
value: 92.747
- type: main_score
value: 21.606
- task:
type: Retrieval
dataset:
name: MTEB AlphaNLI (default)
type: RAR-b/alphanli
config: default
split: test
revision: 303f40ef3d50918d3dc43577d33f2f7344ad72c1
metrics:
- type: ndcg_at_1
value: 29.047
- type: ndcg_at_3
value: 37.782
- type: ndcg_at_5
value: 39.989999999999995
- type: ndcg_at_10
value: 41.926
- type: ndcg_at_20
value: 43.573
- type: ndcg_at_100
value: 45.957
- type: ndcg_at_1000
value: 47.799
- type: recall_at_1
value: 29.047
- type: recall_at_3
value: 43.799
- type: recall_at_5
value: 49.151
- type: recall_at_10
value: 55.222
- type: recall_at_20
value: 61.748999999999995
- type: recall_at_100
value: 74.543
- type: recall_at_1000
value: 89.491
- type: main_score
value: 41.926
- task:
type: Retrieval
dataset:
name: MTEB BSARDRetrieval (default)
type: maastrichtlawtech/bsard
config: default
split: test
revision: 5effa1b9b5fa3b0f9e12523e6e43e5f86a6e6d59
metrics:
- type: ndcg_at_1
value: 15.315000000000001
- type: ndcg_at_3
value: 22.742
- type: ndcg_at_5
value: 25.146
- type: ndcg_at_10
value: 28.993000000000002
- type: ndcg_at_20
value: 30.797
- type: ndcg_at_100
value: 34.189
- type: ndcg_at_1000
value: 36.507
- type: recall_at_1
value: 15.315000000000001
- type: recall_at_3
value: 27.927999999999997
- type: recall_at_5
value: 33.784
- type: recall_at_10
value: 45.495000000000005
- type: recall_at_20
value: 52.703
- type: recall_at_100
value: 71.622
- type: recall_at_1000
value: 90.54100000000001
- type: main_score
value: 71.622
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER (default)
type: mteb/climate-fever
config: default
split: test
revision: 47f2ac6acb640fc46020b02a5b59fdda04d39380
metrics:
- type: ndcg_at_1
value: 38.111
- type: ndcg_at_3
value: 34.489999999999995
- type: ndcg_at_5
value: 36.986999999999995
- type: ndcg_at_10
value: 41.825
- type: ndcg_at_20
value: 45.326
- type: ndcg_at_100
value: 50.207
- type: ndcg_at_1000
value: 52.686
- type: recall_at_1
value: 16.898
- type: recall_at_3
value: 31.636999999999997
- type: recall_at_5
value: 39.147
- type: recall_at_10
value: 49.787
- type: recall_at_20
value: 59.41499999999999
- type: recall_at_100
value: 77.506
- type: recall_at_1000
value: 90.803
- type: main_score
value: 41.825
- task:
type: Retrieval
dataset:
name: MTEB DBPedia-PL (default)
type: clarin-knext/dbpedia-pl
config: default
split: test
revision: 76afe41d9af165cc40999fcaa92312b8b012064a
metrics:
- type: ndcg_at_1
value: 50.875
- type: ndcg_at_3
value: 43.745
- type: ndcg_at_5
value: 42.186
- type: ndcg_at_10
value: 40.506
- type: ndcg_at_20
value: 40.372
- type: ndcg_at_100
value: 45.967
- type: ndcg_at_1000
value: 53.247
- type: recall_at_1
value: 8.14
- type: recall_at_3
value: 14.038
- type: recall_at_5
value: 18.394
- type: recall_at_10
value: 24.476
- type: recall_at_20
value: 32.141999999999996
- type: recall_at_100
value: 53.027
- type: recall_at_1000
value: 76.108
- type: main_score
value: 40.506
- task:
type: Retrieval
dataset:
name: MTEB FEVER (default)
type: mteb/fever
config: default
split: test
revision: bea83ef9e8fb933d90a2f1d5515737465d613e12
metrics:
- type: ndcg_at_1
value: 91.899
- type: ndcg_at_3
value: 93.267
- type: ndcg_at_5
value: 93.757
- type: ndcg_at_10
value: 94.146
- type: ndcg_at_20
value: 94.42399999999999
- type: ndcg_at_100
value: 94.647
- type: ndcg_at_1000
value: 94.765
- type: recall_at_1
value: 85.329
- type: recall_at_3
value: 94.89
- type: recall_at_5
value: 96.185
- type: recall_at_10
value: 97.234
- type: recall_at_20
value: 98.059
- type: recall_at_100
value: 98.946
- type: recall_at_1000
value: 99.605
- type: main_score
value: 94.146
- task:
type: Retrieval
dataset:
name: MTEB GermanDPR (default)
type: deepset/germandpr
config: default
split: test
revision: 5129d02422a66be600ac89cd3e8531b4f97d347d
metrics:
- type: ndcg_at_1
value: 67.415
- type: ndcg_at_3
value: 81.684
- type: ndcg_at_5
value: 83.829
- type: ndcg_at_10
value: 84.624
- type: ndcg_at_20
value: 84.77900000000001
- type: ndcg_at_100
value: 84.832
- type: ndcg_at_1000
value: 84.832
- type: recall_at_1
value: 67.415
- type: recall_at_3
value: 91.61
- type: recall_at_5
value: 96.78
- type: recall_at_10
value: 99.122
- type: recall_at_20
value: 99.70700000000001
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 84.624
- task:
type: Retrieval
dataset:
name: MTEB GermanQuAD-Retrieval (default)
type: mteb/germanquad-retrieval
config: default
split: test
revision: f5c87ae5a2e7a5106606314eef45255f03151bb3
metrics:
- type: ndcg_at_1
value: 92.967
- type: ndcg_at_3
value: 96.289
- type: ndcg_at_5
value: 96.626
- type: ndcg_at_10
value: 96.68900000000001
- type: ndcg_at_20
value: 96.767
- type: ndcg_at_100
value: 96.812
- type: ndcg_at_1000
value: 96.812
- type: recall_at_1
value: 92.967
- type: recall_at_3
value: 98.457
- type: recall_at_5
value: 99.274
- type: recall_at_10
value: 99.456
- type: recall_at_20
value: 99.773
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 95.7191
- task:
type: Retrieval
dataset:
name: MTEB HellaSwag (default)
type: RAR-b/hellaswag
config: default
split: test
revision: a5c990205e017d10761197ccab3000936689c3ae
metrics:
- type: ndcg_at_1
value: 24.139
- type: ndcg_at_3
value: 34.455999999999996
- type: ndcg_at_5
value: 37.217
- type: ndcg_at_10
value: 39.655
- type: ndcg_at_20
value: 41.177
- type: ndcg_at_100
value: 43.695
- type: ndcg_at_1000
value: 45.528
- type: recall_at_1
value: 24.139
- type: recall_at_3
value: 41.894
- type: recall_at_5
value: 48.565999999999995
- type: recall_at_10
value: 56.065
- type: recall_at_20
value: 62.07899999999999
- type: recall_at_100
value: 75.812
- type: recall_at_1000
value: 90.5
- type: main_score
value: 39.655
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA-PL (default)
type: clarin-knext/hotpotqa-pl
config: default
split: test
revision: a0bd479ac97b4ccb5bd6ce320c415d0bb4beb907
metrics:
- type: ndcg_at_1
value: 81.796
- type: ndcg_at_3
value: 68.66499999999999
- type: ndcg_at_5
value: 71.364
- type: ndcg_at_10
value: 73.414
- type: ndcg_at_20
value: 74.634
- type: ndcg_at_100
value: 76.276
- type: ndcg_at_1000
value: 77.34299999999999
- type: recall_at_1
value: 40.898
- type: recall_at_3
value: 66.009
- type: recall_at_5
value: 71.317
- type: recall_at_10
value: 76.435
- type: recall_at_20
value: 80.35799999999999
- type: recall_at_100
value: 87.54899999999999
- type: recall_at_1000
value: 94.537
- type: main_score
value: 73.414
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO (default)
type: mteb/msmarco
config: default
split: dev
revision: c5a29a104738b98a9e76336939199e264163d4a0
metrics:
- type: ndcg_at_1
value: 23.854
- type: ndcg_at_3
value: 35.573
- type: ndcg_at_5
value: 39.96
- type: ndcg_at_10
value: 44.064
- type: ndcg_at_20
value: 46.572
- type: ndcg_at_100
value: 49.492000000000004
- type: ndcg_at_1000
value: 50.43
- type: recall_at_1
value: 23.202
- type: recall_at_3
value: 44.092999999999996
- type: recall_at_5
value: 54.6
- type: recall_at_10
value: 67.11399999999999
- type: recall_at_20
value: 76.79899999999999
- type: recall_at_100
value: 92.085
- type: recall_at_1000
value: 99.122
- type: main_score
value: 44.064
- task:
type: Retrieval
dataset:
name: MTEB PIQA (default)
type: RAR-b/piqa
config: default
split: test
revision: bb30be7e9184e6b6b1d99bbfe1bb90a3a81842e6
metrics:
- type: ndcg_at_1
value: 26.387
- type: ndcg_at_3
value: 36.972
- type: ndcg_at_5
value: 39.534000000000006
- type: ndcg_at_10
value: 42.443
- type: ndcg_at_20
value: 44.36
- type: ndcg_at_100
value: 46.575
- type: ndcg_at_1000
value: 48.024
- type: recall_at_1
value: 26.387
- type: recall_at_3
value: 44.45
- type: recall_at_5
value: 50.598
- type: recall_at_10
value: 59.57599999999999
- type: recall_at_20
value: 67.13799999999999
- type: recall_at_100
value: 79.217
- type: recall_at_1000
value: 91.023
- type: main_score
value: 42.443
- task:
type: Retrieval
dataset:
name: MTEB Quail (default)
type: RAR-b/quail
config: default
split: test
revision: 1851bc536f8bdab29e03e29191c4586b1d8d7c5a
metrics:
- type: ndcg_at_1
value: 7.242999999999999
- type: ndcg_at_3
value: 11.727
- type: ndcg_at_5
value: 13.69
- type: ndcg_at_10
value: 16.186
- type: ndcg_at_20
value: 17.988
- type: ndcg_at_100
value: 20.926000000000002
- type: ndcg_at_1000
value: 23.980999999999998
- type: recall_at_1
value: 7.242999999999999
- type: recall_at_3
value: 15.037
- type: recall_at_5
value: 19.853
- type: recall_at_10
value: 27.573999999999998
- type: recall_at_20
value: 34.669
- type: recall_at_100
value: 50.662
- type: recall_at_1000
value: 75.735
- type: main_score
value: 16.186
- task:
type: Retrieval
dataset:
name: MTEB RARbCode (default)
type: RAR-b/humanevalpack-mbpp-pooled
config: default
split: test
revision: 25f7d11a7ac12dcbb8d3836eb2de682b98c825e4
metrics:
- type: ndcg_at_1
value: 75.40400000000001
- type: ndcg_at_3
value: 84.796
- type: ndcg_at_5
value: 86.68599999999999
- type: ndcg_at_10
value: 87.63499999999999
- type: ndcg_at_20
value: 87.813
- type: ndcg_at_100
value: 87.912
- type: ndcg_at_1000
value: 87.938
- type: recall_at_1
value: 75.40400000000001
- type: recall_at_3
value: 91.24
- type: recall_at_5
value: 95.822
- type: recall_at_10
value: 98.585
- type: recall_at_20
value: 99.259
- type: recall_at_100
value: 99.798
- type: recall_at_1000
value: 100.0
- type: main_score
value: 87.63499999999999
- task:
type: Retrieval
dataset:
name: MTEB RARbMath (default)
type: RAR-b/math-pooled
config: default
split: test
revision: 2393603c0221ff52f448d12dd75f0856103c6cca
metrics:
- type: ndcg_at_1
value: 90.869
- type: ndcg_at_3
value: 92.971
- type: ndcg_at_5
value: 93.365
- type: ndcg_at_10
value: 93.75099999999999
- type: ndcg_at_20
value: 94.05799999999999
- type: ndcg_at_100
value: 94.426
- type: ndcg_at_1000
value: 94.46600000000001
- type: recall_at_1
value: 90.869
- type: recall_at_3
value: 94.414
- type: recall_at_5
value: 95.363
- type: recall_at_10
value: 96.55
- type: recall_at_20
value: 97.753
- type: recall_at_100
value: 99.699
- type: recall_at_1000
value: 100.0
- type: main_score
value: 93.75099999999999
- task:
type: Retrieval
dataset:
name: MTEB SIQA (default)
type: RAR-b/siqa
config: default
split: test
revision: 4ed8415e9dc24060deefc84be59e2db0aacbadcc
metrics:
- type: ndcg_at_1
value: 2.661
- type: ndcg_at_3
value: 4.207000000000001
- type: ndcg_at_5
value: 4.577
- type: ndcg_at_10
value: 5.219
- type: ndcg_at_20
value: 5.917
- type: ndcg_at_100
value: 7.9670000000000005
- type: ndcg_at_1000
value: 11.527999999999999
- type: recall_at_1
value: 2.661
- type: recall_at_3
value: 5.271
- type: recall_at_5
value: 6.192
- type: recall_at_10
value: 8.187999999999999
- type: recall_at_20
value: 10.952
- type: recall_at_100
value: 22.262
- type: recall_at_1000
value: 52.098
- type: main_score
value: 5.219
- task:
type: Retrieval
dataset:
name: MTEB SpartQA (default)
type: RAR-b/spartqa
config: default
split: test
revision: 9ab3ca3ccdd0d43f9cd6d346a363935d127f4f45
metrics:
- type: ndcg_at_1
value: 1.252
- type: ndcg_at_3
value: 3.644
- type: ndcg_at_5
value: 5.27
- type: ndcg_at_10
value: 7.768
- type: ndcg_at_20
value: 10.181
- type: ndcg_at_100
value: 14.29
- type: ndcg_at_1000
value: 18.417
- type: recall_at_1
value: 0.788
- type: recall_at_3
value: 5.157
- type: recall_at_5
value: 8.728
- type: recall_at_10
value: 15.786
- type: recall_at_20
value: 24.365000000000002
- type: recall_at_100
value: 43.553999999999995
- type: recall_at_1000
value: 73.66
- type: main_score
value: 7.768
- task:
type: Retrieval
dataset:
name: MTEB TempReasonL1 (default)
type: RAR-b/TempReason-l1
config: default
split: test
revision: 9097e99aa8c9d827189c65f2e11bfe756af439f6
metrics:
- type: ndcg_at_1
value: 0.1
- type: ndcg_at_3
value: 0.716
- type: ndcg_at_5
value: 1.095
- type: ndcg_at_10
value: 1.6889999999999998
- type: ndcg_at_20
value: 2.374
- type: ndcg_at_100
value: 4.125
- type: ndcg_at_1000
value: 9.126
- type: recall_at_1
value: 0.1
- type: recall_at_3
value: 1.175
- type: recall_at_5
value: 2.1
- type: recall_at_10
value: 3.975
- type: recall_at_20
value: 6.675000000000001
- type: recall_at_100
value: 16.575
- type: recall_at_1000
value: 59.275
- type: main_score
value: 1.6889999999999998
- task:
type: Retrieval
dataset:
name: MTEB TempReasonL2Fact (default)
type: RAR-b/TempReason-l2-fact
config: default
split: test
revision: 13758bcf978613b249d0de4d0840f57815122bdf
metrics:
- type: ndcg_at_1
value: 28.942
- type: ndcg_at_3
value: 45.412
- type: ndcg_at_5
value: 50.43299999999999
- type: ndcg_at_10
value: 53.976
- type: ndcg_at_20
value: 55.703
- type: ndcg_at_100
value: 57.445
- type: ndcg_at_1000
value: 57.838
- type: recall_at_1
value: 28.942
- type: recall_at_3
value: 57.495
- type: recall_at_5
value: 69.631
- type: recall_at_10
value: 80.452
- type: recall_at_20
value: 87.252
- type: recall_at_100
value: 96.44200000000001
- type: recall_at_1000
value: 99.518
- type: main_score
value: 53.976
- task:
type: Retrieval
dataset:
name: MTEB TempReasonL2Pure (default)
type: RAR-b/TempReason-l2-pure
config: default
split: test
revision: 27668949b97bfb178901e0cf047cbee805305dc1
metrics:
- type: ndcg_at_1
value: 2.001
- type: ndcg_at_3
value: 3.746
- type: ndcg_at_5
value: 4.665
- type: ndcg_at_10
value: 5.972
- type: ndcg_at_20
value: 7.321999999999999
- type: ndcg_at_100
value: 11.068
- type: ndcg_at_1000
value: 15.675
- type: recall_at_1
value: 2.001
- type: recall_at_3
value: 5.04
- type: recall_at_5
value: 7.3
- type: recall_at_10
value: 11.34
- type: recall_at_20
value: 16.713
- type: recall_at_100
value: 37.576
- type: recall_at_1000
value: 75.394
- type: main_score
value: 5.972
- task:
type: Retrieval
dataset:
name: MTEB TempReasonL3Fact (default)
type: RAR-b/TempReason-l3-fact
config: default
split: test
revision: 4b70e90197901da24f3cfcd51d27111292878680
metrics:
- type: ndcg_at_1
value: 19.114
- type: ndcg_at_3
value: 34.72
- type: ndcg_at_5
value: 40.509
- type: ndcg_at_10
value: 44.894
- type: ndcg_at_20
value: 47.021
- type: ndcg_at_100
value: 49.162
- type: ndcg_at_1000
value: 49.833
- type: recall_at_1
value: 19.114
- type: recall_at_3
value: 46.385
- type: recall_at_5
value: 60.438
- type: recall_at_10
value: 73.882
- type: recall_at_20
value: 82.219
- type: recall_at_100
value: 93.47
- type: recall_at_1000
value: 98.735
- type: main_score
value: 44.894
- task:
type: Retrieval
dataset:
name: MTEB TempReasonL3Pure (default)
type: RAR-b/TempReason-l3-pure
config: default
split: test
revision: 68fba138e7e63daccecfbdad0a9d2714e56e34ff
metrics:
- type: ndcg_at_1
value: 0.836
- type: ndcg_at_3
value: 5.319
- type: ndcg_at_5
value: 7.468
- type: ndcg_at_10
value: 10.282
- type: ndcg_at_20
value: 12.457
- type: ndcg_at_100
value: 16.384
- type: ndcg_at_1000
value: 20.081
- type: recall_at_1
value: 0.836
- type: recall_at_3
value: 8.744
- type: recall_at_5
value: 13.963000000000001
- type: recall_at_10
value: 22.729
- type: recall_at_20
value: 31.338
- type: recall_at_100
value: 52.824000000000005
- type: recall_at_1000
value: 82.784
- type: main_score
value: 10.282
- task:
type: Retrieval
dataset:
name: MTEB WinoGrande (default)
type: RAR-b/winogrande
config: default
split: test
revision: f74c094f321077cf909ddfb8bccc1b5912a4ac28
metrics:
- type: ndcg_at_1
value: 47.908
- type: ndcg_at_3
value: 71.58200000000001
- type: ndcg_at_5
value: 74.265
- type: ndcg_at_10
value: 75.61099999999999
- type: ndcg_at_20
value: 76.07300000000001
- type: ndcg_at_100
value: 76.249
- type: ndcg_at_1000
value: 76.249
- type: recall_at_1
value: 47.908
- type: recall_at_3
value: 86.74
- type: recall_at_5
value: 93.21199999999999
- type: recall_at_10
value: 97.316
- type: recall_at_20
value: 99.132
- type: recall_at_100
value: 100.0
- type: recall_at_1000
value: 100.0
- type: main_score
value: 75.61099999999999
- task:
type: Retrieval
dataset:
name: MTEB XMarket (de)
type: jinaai/xmarket_ml
config: de
split: test
revision: dfe57acff5b62c23732a7b7d3e3fb84ff501708b
metrics:
- type: ndcg_at_1
value: 30.394
- type: ndcg_at_3
value: 30.701
- type: ndcg_at_5
value: 31.574
- type: ndcg_at_10
value: 32.961
- type: ndcg_at_20
value: 34.765
- type: ndcg_at_100
value: 38.772
- type: ndcg_at_1000
value: 43.317
- type: recall_at_1
value: 10.193000000000001
- type: recall_at_3
value: 19.141
- type: recall_at_5
value: 24.362000000000002
- type: recall_at_10
value: 31.995
- type: recall_at_20
value: 40.047
- type: recall_at_100
value: 56.769000000000005
- type: recall_at_1000
value: 76.318
- type: main_score
value: 32.961
- task:
type: Retrieval
dataset:
name: MTEB XMarket (en)
type: jinaai/xmarket_ml
config: en
split: test
revision: dfe57acff5b62c23732a7b7d3e3fb84ff501708b
metrics:
- type: ndcg_at_1
value: 37.652
- type: ndcg_at_3
value: 38.444
- type: ndcg_at_5
value: 39.163
- type: ndcg_at_10
value: 40.557
- type: ndcg_at_20
value: 42.224000000000004
- type: ndcg_at_100
value: 46.817
- type: ndcg_at_1000
value: 51.939
- type: recall_at_1
value: 8.909
- type: recall_at_3
value: 18.673000000000002
- type: recall_at_5
value: 24.364
- type: recall_at_10
value: 32.919
- type: recall_at_20
value: 41.908
- type: recall_at_100
value: 61.663999999999994
- type: recall_at_1000
value: 80.619
- type: main_score
value: 40.557
- task:
type: Retrieval
dataset:
name: MTEB XMarket (es)
type: jinaai/xmarket_ml
config: es
split: test
revision: dfe57acff5b62c23732a7b7d3e3fb84ff501708b
metrics:
- type: ndcg_at_1
value: 32.168
- type: ndcg_at_3
value: 32.389
- type: ndcg_at_5
value: 33.054
- type: ndcg_at_10
value: 34.549
- type: ndcg_at_20
value: 36.34
- type: ndcg_at_100
value: 40.324
- type: ndcg_at_1000
value: 44.784
- type: recall_at_1
value: 10.845
- type: recall_at_3
value: 21.058
- type: recall_at_5
value: 26.327
- type: recall_at_10
value: 34.306
- type: recall_at_20
value: 42.46
- type: recall_at_100
value: 59.156
- type: recall_at_1000
value: 78.249
- type: main_score
value: 34.549
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (ara-ara)
type: jinaai/xpqa
config: ara-ara
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 42.0
- type: ndcg_at_3
value: 43.802
- type: ndcg_at_5
value: 46.1
- type: ndcg_at_10
value: 50.858000000000004
- type: ndcg_at_20
value: 54.303999999999995
- type: ndcg_at_100
value: 57.692
- type: ndcg_at_1000
value: 58.97599999999999
- type: recall_at_1
value: 23.989
- type: recall_at_3
value: 42.753
- type: recall_at_5
value: 51.56699999999999
- type: recall_at_10
value: 63.92400000000001
- type: recall_at_20
value: 75.249
- type: recall_at_100
value: 90.851
- type: recall_at_1000
value: 99.733
- type: main_score
value: 50.858000000000004
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-ara)
type: jinaai/xpqa
config: eng-ara
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 31.2
- type: ndcg_at_3
value: 33.296
- type: ndcg_at_5
value: 35.727
- type: ndcg_at_10
value: 39.837
- type: ndcg_at_20
value: 43.354
- type: ndcg_at_100
value: 47.908
- type: ndcg_at_1000
value: 50.187000000000005
- type: recall_at_1
value: 18.007
- type: recall_at_3
value: 32.5
- type: recall_at_5
value: 41.422
- type: recall_at_10
value: 51.673
- type: recall_at_20
value: 63.144
- type: recall_at_100
value: 83.733
- type: recall_at_1000
value: 99.10900000000001
- type: main_score
value: 39.837
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (ara-eng)
type: jinaai/xpqa
config: ara-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 40.431
- type: ndcg_at_3
value: 41.419
- type: ndcg_at_5
value: 44.051
- type: ndcg_at_10
value: 48.94
- type: ndcg_at_20
value: 52.532999999999994
- type: ndcg_at_100
value: 56.203
- type: ndcg_at_1000
value: 57.467999999999996
- type: recall_at_1
value: 22.534000000000002
- type: recall_at_3
value: 40.119
- type: recall_at_5
value: 49.569
- type: recall_at_10
value: 62.156
- type: recall_at_20
value: 74.191
- type: recall_at_100
value: 90.973
- type: recall_at_1000
value: 99.72999999999999
- type: main_score
value: 48.94
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (deu-deu)
type: jinaai/xpqa
config: deu-deu
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 76.50099999999999
- type: ndcg_at_3
value: 79.38199999999999
- type: ndcg_at_5
value: 81.00500000000001
- type: ndcg_at_10
value: 82.786
- type: ndcg_at_20
value: 83.844
- type: ndcg_at_100
value: 84.708
- type: ndcg_at_1000
value: 84.956
- type: recall_at_1
value: 58.464000000000006
- type: recall_at_3
value: 79.963
- type: recall_at_5
value: 85.757
- type: recall_at_10
value: 90.372
- type: recall_at_20
value: 94.13
- type: recall_at_100
value: 98.24000000000001
- type: recall_at_1000
value: 100.0
- type: main_score
value: 82.786
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-deu)
type: jinaai/xpqa
config: eng-deu
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 52.611
- type: ndcg_at_3
value: 55.35099999999999
- type: ndcg_at_5
value: 57.452999999999996
- type: ndcg_at_10
value: 61.553999999999995
- type: ndcg_at_20
value: 63.919000000000004
- type: ndcg_at_100
value: 66.90700000000001
- type: ndcg_at_1000
value: 67.685
- type: recall_at_1
value: 33.47
- type: recall_at_3
value: 55.174
- type: recall_at_5
value: 63.512
- type: recall_at_10
value: 73.934
- type: recall_at_20
value: 81.26400000000001
- type: recall_at_100
value: 94.606
- type: recall_at_1000
value: 100.0
- type: main_score
value: 61.553999999999995
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (deu-eng)
type: jinaai/xpqa
config: deu-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 70.235
- type: ndcg_at_3
value: 74.824
- type: ndcg_at_5
value: 76.47699999999999
- type: ndcg_at_10
value: 78.803
- type: ndcg_at_20
value: 80.19
- type: ndcg_at_100
value: 81.07799999999999
- type: ndcg_at_1000
value: 81.40899999999999
- type: recall_at_1
value: 52.818
- type: recall_at_3
value: 76.754
- type: recall_at_5
value: 82.637
- type: recall_at_10
value: 88.655
- type: recall_at_20
value: 93.61
- type: recall_at_100
value: 97.731
- type: recall_at_1000
value: 100.0
- type: main_score
value: 78.803
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (spa-spa)
type: jinaai/xpqa
config: spa-spa
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 64.18700000000001
- type: ndcg_at_3
value: 62.714999999999996
- type: ndcg_at_5
value: 64.134
- type: ndcg_at_10
value: 68.143
- type: ndcg_at_20
value: 70.625
- type: ndcg_at_100
value: 73.333
- type: ndcg_at_1000
value: 74.02300000000001
- type: recall_at_1
value: 34.400999999999996
- type: recall_at_3
value: 57.654
- type: recall_at_5
value: 67.167
- type: recall_at_10
value: 76.31599999999999
- type: recall_at_20
value: 83.731
- type: recall_at_100
value: 95.502
- type: recall_at_1000
value: 99.58
- type: main_score
value: 68.143
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-spa)
type: jinaai/xpqa
config: eng-spa
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 47.667
- type: ndcg_at_3
value: 46.35
- type: ndcg_at_5
value: 47.879
- type: ndcg_at_10
value: 52.733
- type: ndcg_at_20
value: 55.620000000000005
- type: ndcg_at_100
value: 59.70100000000001
- type: ndcg_at_1000
value: 61.417
- type: recall_at_1
value: 23.394000000000002
- type: recall_at_3
value: 42.264
- type: recall_at_5
value: 51.144999999999996
- type: recall_at_10
value: 62.556
- type: recall_at_20
value: 71.269
- type: recall_at_100
value: 88.668
- type: recall_at_1000
value: 99.466
- type: main_score
value: 52.733
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (spa-eng)
type: jinaai/xpqa
config: spa-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 61.285999999999994
- type: ndcg_at_3
value: 60.303
- type: ndcg_at_5
value: 62.062
- type: ndcg_at_10
value: 66.042
- type: ndcg_at_20
value: 68.509
- type: ndcg_at_100
value: 71.539
- type: ndcg_at_1000
value: 72.258
- type: recall_at_1
value: 32.224000000000004
- type: recall_at_3
value: 55.443
- type: recall_at_5
value: 65.67699999999999
- type: recall_at_10
value: 74.607
- type: recall_at_20
value: 82.234
- type: recall_at_100
value: 95.275
- type: recall_at_1000
value: 99.723
- type: main_score
value: 66.042
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (fr)
type: jinaai/xpqa
config: fra-fra
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 71.429
- type: ndcg_at_3
value: 71.13000000000001
- type: ndcg_at_5
value: 72.709
- type: ndcg_at_10
value: 76.236
- type: ndcg_at_20
value: 77.78500000000001
- type: ndcg_at_100
value: 79.634
- type: ndcg_at_1000
value: 79.953
- type: recall_at_1
value: 45.943
- type: recall_at_3
value: 68.293
- type: recall_at_5
value: 76.5
- type: recall_at_10
value: 85.11999999999999
- type: recall_at_20
value: 90.069
- type: recall_at_100
value: 97.82600000000001
- type: recall_at_1000
value: 99.866
- type: main_score
value: 76.236
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-fra)
type: jinaai/xpqa
config: eng-fra
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 47.797
- type: ndcg_at_3
value: 49.514
- type: ndcg_at_5
value: 51.782
- type: ndcg_at_10
value: 55.891000000000005
- type: ndcg_at_20
value: 59.226
- type: ndcg_at_100
value: 62.612
- type: ndcg_at_1000
value: 63.749
- type: recall_at_1
value: 26.689
- type: recall_at_3
value: 47.408
- type: recall_at_5
value: 57.399
- type: recall_at_10
value: 67.147
- type: recall_at_20
value: 77.837
- type: recall_at_100
value: 92.494
- type: recall_at_1000
value: 99.74
- type: main_score
value: 55.891000000000005
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (fra-eng)
type: jinaai/xpqa
config: fra-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 68.625
- type: ndcg_at_3
value: 68.239
- type: ndcg_at_5
value: 70.175
- type: ndcg_at_10
value: 73.452
- type: ndcg_at_20
value: 75.66000000000001
- type: ndcg_at_100
value: 77.506
- type: ndcg_at_1000
value: 77.936
- type: recall_at_1
value: 44.035999999999994
- type: recall_at_3
value: 65.291
- type: recall_at_5
value: 74.37899999999999
- type: recall_at_10
value: 82.15
- type: recall_at_20
value: 89.457
- type: recall_at_100
value: 97.194
- type: recall_at_1000
value: 99.933
- type: main_score
value: 73.452
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (hin-hin)
type: jinaai/xpqa
config: hin-hin
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 66.703
- type: ndcg_at_3
value: 72.993
- type: ndcg_at_5
value: 75.138
- type: ndcg_at_10
value: 77.371
- type: ndcg_at_20
value: 78.389
- type: ndcg_at_100
value: 79.623
- type: ndcg_at_1000
value: 79.975
- type: recall_at_1
value: 57.094
- type: recall_at_3
value: 77.2
- type: recall_at_5
value: 82.50800000000001
- type: recall_at_10
value: 88.486
- type: recall_at_20
value: 91.863
- type: recall_at_100
value: 97.359
- type: recall_at_1000
value: 99.892
- type: main_score
value: 77.371
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-hin)
type: jinaai/xpqa
config: eng-hin
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 29.837999999999997
- type: ndcg_at_3
value: 34.187
- type: ndcg_at_5
value: 37.132
- type: ndcg_at_10
value: 41.357
- type: ndcg_at_20
value: 44.522
- type: ndcg_at_100
value: 49.486999999999995
- type: ndcg_at_1000
value: 51.458000000000006
- type: recall_at_1
value: 24.959999999999997
- type: recall_at_3
value: 36.472
- type: recall_at_5
value: 44.175
- type: recall_at_10
value: 55.371
- type: recall_at_20
value: 65.506
- type: recall_at_100
value: 87.252
- type: recall_at_1000
value: 99.78399999999999
- type: main_score
value: 41.357
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (hin-eng)
type: jinaai/xpqa
config: hin-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 62.829
- type: ndcg_at_3
value: 68.886
- type: ndcg_at_5
value: 71.812
- type: ndcg_at_10
value: 74.405
- type: ndcg_at_20
value: 75.702
- type: ndcg_at_100
value: 77.08500000000001
- type: ndcg_at_1000
value: 77.377
- type: recall_at_1
value: 53.568000000000005
- type: recall_at_3
value: 73.095
- type: recall_at_5
value: 80.211
- type: recall_at_10
value: 87.229
- type: recall_at_20
value: 91.625
- type: recall_at_100
value: 97.844
- type: recall_at_1000
value: 100.0
- type: main_score
value: 74.405
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (ita-ita)
type: jinaai/xpqa
config: ita-ita
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 73.303
- type: ndcg_at_3
value: 74.51299999999999
- type: ndcg_at_5
value: 76.383
- type: ndcg_at_10
value: 78.968
- type: ndcg_at_20
value: 80.331
- type: ndcg_at_100
value: 81.65599999999999
- type: ndcg_at_1000
value: 82.075
- type: recall_at_1
value: 50.68899999999999
- type: recall_at_3
value: 72.763
- type: recall_at_5
value: 80.85
- type: recall_at_10
value: 87.071
- type: recall_at_20
value: 91.62599999999999
- type: recall_at_100
value: 97.333
- type: recall_at_1000
value: 100.0
- type: main_score
value: 78.968
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-ita)
type: jinaai/xpqa
config: eng-ita
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 42.232
- type: ndcg_at_3
value: 46.231
- type: ndcg_at_5
value: 48.197
- type: ndcg_at_10
value: 52.217
- type: ndcg_at_20
value: 55.472
- type: ndcg_at_100
value: 58.803000000000004
- type: ndcg_at_1000
value: 60.321000000000005
- type: recall_at_1
value: 26.368000000000002
- type: recall_at_3
value: 46.709
- type: recall_at_5
value: 54.721
- type: recall_at_10
value: 64.46
- type: recall_at_20
value: 74.997
- type: recall_at_100
value: 89.527
- type: recall_at_1000
value: 99.698
- type: main_score
value: 52.217
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (ita-eng)
type: jinaai/xpqa
config: ita-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 68.326
- type: ndcg_at_3
value: 70.71499999999999
- type: ndcg_at_5
value: 72.748
- type: ndcg_at_10
value: 75.31
- type: ndcg_at_20
value: 76.958
- type: ndcg_at_100
value: 78.66300000000001
- type: ndcg_at_1000
value: 79.089
- type: recall_at_1
value: 46.583999999999996
- type: recall_at_3
value: 69.887
- type: recall_at_5
value: 78.10000000000001
- type: recall_at_10
value: 84.329
- type: recall_at_20
value: 89.51
- type: recall_at_100
value: 97.235
- type: recall_at_1000
value: 100.0
- type: main_score
value: 75.31
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (jpn-jpn)
type: jinaai/xpqa
config: jpn-jpn
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 72.0
- type: ndcg_at_3
value: 74.005
- type: ndcg_at_5
value: 75.411
- type: ndcg_at_10
value: 77.12
- type: ndcg_at_20
value: 78.625
- type: ndcg_at_100
value: 80.281
- type: ndcg_at_1000
value: 80.682
- type: recall_at_1
value: 46.988
- type: recall_at_3
value: 72.36200000000001
- type: recall_at_5
value: 79.501
- type: recall_at_10
value: 83.83
- type: recall_at_20
value: 88.907
- type: recall_at_100
value: 96.739
- type: recall_at_1000
value: 99.636
- type: main_score
value: 77.12
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-jpn)
type: jinaai/xpqa
config: eng-jpn
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 43.758
- type: ndcg_at_3
value: 45.513999999999996
- type: ndcg_at_5
value: 47.93
- type: ndcg_at_10
value: 51.983999999999995
- type: ndcg_at_20
value: 54.544000000000004
- type: ndcg_at_100
value: 58.022
- type: ndcg_at_1000
value: 59.843
- type: recall_at_1
value: 25.543
- type: recall_at_3
value: 44.374
- type: recall_at_5
value: 53.86300000000001
- type: recall_at_10
value: 63.756
- type: recall_at_20
value: 72.14699999999999
- type: recall_at_100
value: 87.58200000000001
- type: recall_at_1000
value: 99.295
- type: main_score
value: 51.983999999999995
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (jpn-eng)
type: jinaai/xpqa
config: jpn-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 68.978
- type: ndcg_at_3
value: 71.019
- type: ndcg_at_5
value: 72.697
- type: ndcg_at_10
value: 75.267
- type: ndcg_at_20
value: 76.655
- type: ndcg_at_100
value: 78.388
- type: ndcg_at_1000
value: 78.899
- type: recall_at_1
value: 44.958999999999996
- type: recall_at_3
value: 69.56400000000001
- type: recall_at_5
value: 77.082
- type: recall_at_10
value: 83.646
- type: recall_at_20
value: 88.238
- type: recall_at_100
value: 96.194
- type: recall_at_1000
value: 99.818
- type: main_score
value: 75.267
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (kor-kor)
type: jinaai/xpqa
config: kor-kor
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 33.18
- type: ndcg_at_3
value: 35.311
- type: ndcg_at_5
value: 38.366
- type: ndcg_at_10
value: 41.654
- type: ndcg_at_20
value: 44.244
- type: ndcg_at_100
value: 49.001
- type: ndcg_at_1000
value: 51.01
- type: recall_at_1
value: 23.201
- type: recall_at_3
value: 37.011
- type: recall_at_5
value: 44.493
- type: recall_at_10
value: 53.489
- type: recall_at_20
value: 62.548
- type: recall_at_100
value: 85.55
- type: recall_at_1000
value: 100.0
- type: main_score
value: 41.654
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-kor)
type: jinaai/xpqa
config: eng-kor
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 34.404
- type: ndcg_at_3
value: 35.821
- type: ndcg_at_5
value: 37.268
- type: ndcg_at_10
value: 40.967
- type: ndcg_at_20
value: 43.509
- type: ndcg_at_100
value: 49.326
- type: ndcg_at_1000
value: 51.410999999999994
- type: recall_at_1
value: 20.363999999999997
- type: recall_at_3
value: 35.293
- type: recall_at_5
value: 41.251
- type: recall_at_10
value: 50.766999999999996
- type: recall_at_20
value: 59.274
- type: recall_at_100
value: 86.669
- type: recall_at_1000
value: 100.0
- type: main_score
value: 40.967
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (kor-eng)
type: jinaai/xpqa
config: kor-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 33.062000000000005
- type: ndcg_at_3
value: 35.619
- type: ndcg_at_5
value: 37.684
- type: ndcg_at_10
value: 40.986
- type: ndcg_at_20
value: 43.736999999999995
- type: ndcg_at_100
value: 48.632999999999996
- type: ndcg_at_1000
value: 50.78
- type: recall_at_1
value: 23.18
- type: recall_at_3
value: 37.235
- type: recall_at_5
value: 42.448
- type: recall_at_10
value: 51.395
- type: recall_at_20
value: 61.01
- type: recall_at_100
value: 84.382
- type: recall_at_1000
value: 100.0
- type: main_score
value: 40.986
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (pol-pol)
type: jinaai/xpqa
config: pol-pol
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 46.115
- type: ndcg_at_3
value: 45.966
- type: ndcg_at_5
value: 48.119
- type: ndcg_at_10
value: 51.53
- type: ndcg_at_20
value: 54.447
- type: ndcg_at_100
value: 58.939
- type: ndcg_at_1000
value: 60.428000000000004
- type: recall_at_1
value: 27.641
- type: recall_at_3
value: 45.021
- type: recall_at_5
value: 52.580000000000005
- type: recall_at_10
value: 61.141999999999996
- type: recall_at_20
value: 70.588
- type: recall_at_100
value: 90.29700000000001
- type: recall_at_1000
value: 99.851
- type: main_score
value: 51.53
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-pol)
type: jinaai/xpqa
config: eng-pol
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 32.357
- type: ndcg_at_3
value: 31.573
- type: ndcg_at_5
value: 33.046
- type: ndcg_at_10
value: 37.364999999999995
- type: ndcg_at_20
value: 40.407
- type: ndcg_at_100
value: 45.965
- type: ndcg_at_1000
value: 48.982
- type: recall_at_1
value: 14.865999999999998
- type: recall_at_3
value: 28.51
- type: recall_at_5
value: 35.827999999999996
- type: recall_at_10
value: 46.11
- type: recall_at_20
value: 55.498999999999995
- type: recall_at_100
value: 79.73
- type: recall_at_1000
value: 99.236
- type: main_score
value: 37.364999999999995
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (pol-eng)
type: jinaai/xpqa
config: pol-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 43.114999999999995
- type: ndcg_at_3
value: 42.306
- type: ndcg_at_5
value: 44.484
- type: ndcg_at_10
value: 48.374
- type: ndcg_at_20
value: 51.347
- type: ndcg_at_100
value: 56.223
- type: ndcg_at_1000
value: 57.93899999999999
- type: recall_at_1
value: 25.746000000000002
- type: recall_at_3
value: 41.160000000000004
- type: recall_at_5
value: 48.256
- type: recall_at_10
value: 58.038999999999994
- type: recall_at_20
value: 67.499
- type: recall_at_100
value: 88.912
- type: recall_at_1000
value: 99.85000000000001
- type: main_score
value: 48.374
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (por-por)
type: jinaai/xpqa
config: por-por
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 47.25
- type: ndcg_at_3
value: 46.225
- type: ndcg_at_5
value: 47.813
- type: ndcg_at_10
value: 51.383
- type: ndcg_at_20
value: 54.291
- type: ndcg_at_100
value: 58.434
- type: ndcg_at_1000
value: 60.07
- type: recall_at_1
value: 25.394
- type: recall_at_3
value: 43.446
- type: recall_at_5
value: 51.037
- type: recall_at_10
value: 59.61
- type: recall_at_20
value: 68.925
- type: recall_at_100
value: 88.277
- type: recall_at_1000
value: 99.44800000000001
- type: main_score
value: 51.383
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-por)
type: jinaai/xpqa
config: eng-por
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 29.5
- type: ndcg_at_3
value: 29.971999999999998
- type: ndcg_at_5
value: 31.513999999999996
- type: ndcg_at_10
value: 35.449999999999996
- type: ndcg_at_20
value: 38.912
- type: ndcg_at_100
value: 44.695
- type: ndcg_at_1000
value: 47.309
- type: recall_at_1
value: 14.335
- type: recall_at_3
value: 27.839999999999996
- type: recall_at_5
value: 34.737
- type: recall_at_10
value: 44.358
- type: recall_at_20
value: 55.65
- type: recall_at_100
value: 82.077
- type: recall_at_1000
value: 99.44800000000001
- type: main_score
value: 35.449999999999996
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (por-eng)
type: jinaai/xpqa
config: por-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 46.048
- type: ndcg_at_3
value: 45.519
- type: ndcg_at_5
value: 47.693999999999996
- type: ndcg_at_10
value: 51.535
- type: ndcg_at_20
value: 54.179
- type: ndcg_at_100
value: 58.416999999999994
- type: ndcg_at_1000
value: 59.955000000000005
- type: recall_at_1
value: 25.325999999999997
- type: recall_at_3
value: 42.779
- type: recall_at_5
value: 51.453
- type: recall_at_10
value: 60.876
- type: recall_at_20
value: 69.184
- type: recall_at_100
value: 88.97699999999999
- type: recall_at_1000
value: 99.58200000000001
- type: main_score
value: 51.535
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (tam-tam)
type: jinaai/xpqa
config: tam-tam
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 31.968999999999998
- type: ndcg_at_3
value: 34.555
- type: ndcg_at_5
value: 36.504999999999995
- type: ndcg_at_10
value: 38.958
- type: ndcg_at_20
value: 40.77
- type: ndcg_at_100
value: 43.779
- type: ndcg_at_1000
value: 47.388999999999996
- type: recall_at_1
value: 21.13
- type: recall_at_3
value: 35.838
- type: recall_at_5
value: 41.535
- type: recall_at_10
value: 48.075
- type: recall_at_20
value: 54.290000000000006
- type: recall_at_100
value: 68.325
- type: recall_at_1000
value: 95.62
- type: main_score
value: 38.958
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-tam)
type: jinaai/xpqa
config: eng-tam
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 12.531999999999998
- type: ndcg_at_3
value: 12.849
- type: ndcg_at_5
value: 13.979
- type: ndcg_at_10
value: 16.573
- type: ndcg_at_20
value: 18.861
- type: ndcg_at_100
value: 23.779
- type: ndcg_at_1000
value: 29.859
- type: recall_at_1
value: 7.388999999999999
- type: recall_at_3
value: 12.531999999999998
- type: recall_at_5
value: 16.279
- type: recall_at_10
value: 23.099
- type: recall_at_20
value: 30.697000000000003
- type: recall_at_100
value: 53.608
- type: recall_at_1000
value: 94.719
- type: main_score
value: 16.573
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (tam-eng)
type: jinaai/xpqa
config: tam-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 21.066
- type: ndcg_at_3
value: 23.677999999999997
- type: ndcg_at_5
value: 25.851000000000003
- type: ndcg_at_10
value: 28.615000000000002
- type: ndcg_at_20
value: 30.817
- type: ndcg_at_100
value: 34.874
- type: ndcg_at_1000
value: 39.24
- type: recall_at_1
value: 15.037
- type: recall_at_3
value: 25.285999999999998
- type: recall_at_5
value: 30.717
- type: recall_at_10
value: 37.722
- type: recall_at_20
value: 44.927
- type: recall_at_100
value: 63.917
- type: recall_at_1000
value: 96.145
- type: main_score
value: 28.615000000000002
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (cmn-cmn)
type: jinaai/xpqa
config: cmn-cmn
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 65.049
- type: ndcg_at_3
value: 65.534
- type: ndcg_at_5
value: 67.498
- type: ndcg_at_10
value: 70.812
- type: ndcg_at_20
value: 73.026
- type: ndcg_at_100
value: 75.316
- type: ndcg_at_1000
value: 75.882
- type: recall_at_1
value: 41.357
- type: recall_at_3
value: 63.176
- type: recall_at_5
value: 71.381
- type: recall_at_10
value: 79.47
- type: recall_at_20
value: 86.616
- type: recall_at_100
value: 96.36099999999999
- type: recall_at_1000
value: 100.0
- type: main_score
value: 70.812
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-cmn)
type: jinaai/xpqa
config: eng-cmn
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 35.073
- type: ndcg_at_3
value: 35.782000000000004
- type: ndcg_at_5
value: 36.99
- type: ndcg_at_10
value: 40.974
- type: ndcg_at_20
value: 43.971
- type: ndcg_at_100
value: 49.165
- type: ndcg_at_1000
value: 51.93
- type: recall_at_1
value: 20.057
- type: recall_at_3
value: 34.064
- type: recall_at_5
value: 40.831
- type: recall_at_10
value: 50.33
- type: recall_at_20
value: 59.306000000000004
- type: recall_at_100
value: 82.231
- type: recall_at_1000
value: 99.759
- type: main_score
value: 40.974
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (cmn-eng)
type: jinaai/xpqa
config: cmn-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 57.68299999999999
- type: ndcg_at_3
value: 60.089000000000006
- type: ndcg_at_5
value: 62.217999999999996
- type: ndcg_at_10
value: 65.81700000000001
- type: ndcg_at_20
value: 67.886
- type: ndcg_at_100
value: 70.804
- type: ndcg_at_1000
value: 71.54
- type: recall_at_1
value: 36.146
- type: recall_at_3
value: 59.035000000000004
- type: recall_at_5
value: 67.376
- type: recall_at_10
value: 76.213
- type: recall_at_20
value: 82.756
- type: recall_at_100
value: 95.341
- type: recall_at_1000
value: 100.0
- type: main_score
value: 65.81700000000001
---
# INF-Retriever-v1
## Model Overview
- **INF-Retriever-v1** is an LLM-based dense retrieval model developed by [INF TECH](https://www.infly.cn/en).
It is built upon the [gte-Qwen2-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) model and specifically fine-tuned to excel in retrieval tasks, particularly for Chinese and English data.
- As of January 23, 2025, **INF-Retriever-v1** ranks both **No.1** on the Automated Heterogeneous Information Retrieval Benchmark of version 24.04 & 24.05([AIR-Bench](https://huggingface.co/spaces/AIR-Bench/leaderboard)), showcasing its cutting-edge performance in heterogeneous information retrieval tasks.
## Key Features
- **Optimized for Chinese and English retrieval**: The model has been specifically fine-tuned with retrieval-focused datasets in both languages, significantly improving its accuracy and efficiency for a variety of retrieval scenarios.
- **Top-tier performance**: **INF-Retriever-v1** has achieved outstanding results on the AIR-Bench leaderboard, making it a top choice for heterogeneous information retrieval tasks across various domains.
## Model Details
- Model Size: 7B
- Embedding Dimension: 3584
- Max Input Tokens: 32768
- Language Support: Chinese & English (also effective in other languages)
## Usage
### Sentence Transformers
```python
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("infly/inf-retriever-v1", trust_remote_code=True)
# In case you want to reduce the maximum length:
model.max_seq_length = 8192
queries = [
"how much protein should a female eat",
"summit define",
]
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments.",
]
query_embeddings = model.encode(queries, prompt_name="query")
document_embeddings = model.encode(documents)
scores = (query_embeddings @ document_embeddings.T) * 100
print(scores.tolist())
# [[86.8702392578125, 67.82364654541016], [59.51014709472656, 82.33668518066406]]
```
### Transformers
```python
import torch
import torch.nn.functional as F
from torch import Tensor
from transformers import AutoTokenizer, AutoModel
def last_token_pool(last_hidden_states: Tensor,
attention_mask: Tensor) -> Tensor:
left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])
if left_padding:
return last_hidden_states[:, -1]
else:
sequence_lengths = attention_mask.sum(dim=1) - 1
batch_size = last_hidden_states.shape[0]
return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'Instruct: {task_description}\nQuery: {query}'
# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = [
get_detailed_instruct(task, 'how much protein should a female eat'),
get_detailed_instruct(task, 'summit define')
]
# No need to add instruction for retrieval documents
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."
]
input_texts = queries + documents
tokenizer = AutoTokenizer.from_pretrained('infly/inf-retriever-v1', trust_remote_code=True)
model = AutoModel.from_pretrained('infly/inf-retriever-v1', trust_remote_code=True)
max_length = 8192
# Tokenize the input texts
batch_dict = tokenizer(input_texts, max_length=max_length, padding=True, truncation=True, return_tensors='pt')
outputs = model(**batch_dict)
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
# normalize embeddings
embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:2] @ embeddings[2:].T) * 100
print(scores.tolist())
# [[86.87025451660156, 67.82366180419922], [59.510135650634766, 82.33667755126953]]
```
## Evaluation
### AIR-Bench
**INF-Retriever-v1** has demonstrated superior retrieval capabilities across multiple domains and languages. The results from the Automated Heterogeneous Information Retrieval Benchmark ([AIR-Bench](https://huggingface.co/spaces/AIR-Bench/leaderboard)) as of January 23, 2025, are as follows:
#### AIR-Bench_24.04 (Bilingual, EN & ZH)
| Model Name | Average⬆️ | wiki_en | wiki_zh | web_en | web_zh | healthcare_en | healthcare_zh | law_en | arxiv_en | news_en | news_zh | finance_en | finance_zh | msmarco_en |
|-----------------------------------------------------------------------------------|-----------|-----------|-----------|-----------|----------|---------------|---------------|-----------|-----------|-----------|-----------|------------|------------|------------|
| [E5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct) | 45.26 | 61.67 | 55.97 | 44.41 | 45.96 | 56.32 | 35.79 | 19.32 | 44.78 | 48.18 | 35.99 | 54.79 | 26.11 | 59.03 |
| [BGE-M3](https://huggingface.co/BAAI/bge-m3) | 46.65 | 60.49 | 62.36 | 47.35 | 50.38 | 49.1 | **42.38** | 26.68 | 40.76 | 48.04 | 40.75 | 51.52 | 32.18 | 54.4 |
| [BGE-Multilingual-Gemma2](https://huggingface.co/BAAI/bge-multilingual-gemma2) | 46.83 | 63.71 | 67.3 | 50.38 | 53.24 | 47.24 | 42.13 | 22.58 | 23.28 | 50.91 | 44.02 | 49.3 | 31.6 | **63.14** |
| [GTE-Qwen2-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) | 48.38 | 63.46 | 66.44 | 51.2 | 51.98 | 54.2 | 38.82 | 22.31 | 40.27 | **54.07** | 43.03 | 58.2 | 26.63 | 58.39 |
| **INF-Retriever-v1** | **52.56** | **65.25** | **68.44** | **52.13** | **56.6** | **56.96** | 42.03 | **34.51** | **50.62** | 53.32 | **50.02** | **58.34** | **35.42** | 59.64 |
#### AIR-Bench_24.05 (Multilingual, 13 languages)
Although INF-Retriever-v1 has been fine-tuned exclusively on English and Chinese, it continues to perform exceptionally well across other languages, securing the No. 1 position on this multilingual benchmark.
| Model Name | Average⬆️ | wiki_en | wiki_zh | wiki_ar | wiki_bn | wiki_de | wiki_es | wiki_fa | wiki_fr | wiki_hi | wiki_id | wiki_ja | wiki_ko | wiki_ru | web_en | web_zh | web_ar | web_bn | web_de | web_es | web_fa | web_fr | web_hi | web_id | web_ja | web_ko | web_ru | healthcare_en | healthcare_zh | healthcare_de | healthcare_es | healthcare_fr | law_en | law_de | law_fr | arxiv_en | science_ru | news_en | news_zh | news_ar | news_bn | news_de | news_es | news_fa | news_fr | news_hi | news_id | news_ja | news_ko | news_ru | finance_en | finance_zh | finance_ar | finance_fr |
|--------------------------------------------------------------------------------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|----------|-----------|-----------|----------|--------|-----------|-----------|-----------|---------------|---------------|---------------|---------------|---------------|-----------|-----------|-----------|-----------|------------|-----------|-----------|-----------|-----------|-----------|----------|-----------|----------|-----------|-----------|-----------|-----------|-----------|------------|------------|------------|------------|
| [GTE-Qwen2-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) | 50.05 | **73.59** | 67.5 | 59.44 | 58.17 | 63.96 | 67.62 | 57.05 | 70.32 | 60.54 | 61.81 | 62.88 | 59.17 | 62.95 | **58.99** | 51.66 | 55.56 | 51.45 | 48.62 | 54.11 | 49.54 | 55.16 | 53.06 | 55.51 | 57.27 | 57.54 | 55.88 | 54.46 | 38.66 | 53.92 | 53.78 | 30.29 | 22.75 | 13.18 | 13.15 | 41.32 | 45.21 | **52.74** | 43.17 | 37.63 | **61.31** | 44.89 | 45.21 | 30.1 | 49.76 | 30.28 | 46.44 | 44.13 | 47.19 | 46.55 | 59.23 | 34.61 | 43.56 | 39.57 |
| [Multilingual-E5-large-instruct](https://huggingface.co/intfloat/multilingual-e5-large-instruct) | 51.11 | 68.62 | 62.82 | 63.21 | 64.45 | 65.81 | 68.1 | 64.2 | 69.72 | 71.81 | 66.36 | 64.12 | 64.79 | 62.57 | 41.58 | 47.06 | 56.4 | 56.17 | 50.87 | 52.24 | 58.68 | 50.2 | 56.32 | 54.49 | 54.89 | 55.81 | 54.97 | 54.02 | 39.76 | 52.06 | 51.74 | 36.64 | 16.9 | 15.59 | 15.12 | 39.52 | 56.86 | 44.28 | 35.46 | 48.2 | 49.31 | 47.84 | 45.99 | **45.59** | 50.58 | 39.66 | 48.59 | 47.6 | 50.52 | 48.81 | 52.79 | 37.72 | 48.95 | 42.74 |
| [BGE-M3](https://huggingface.co/BAAI/bge-m3) | 51.31 | 69.7 | 63.52 | 59.65 | 64.33 | 64.68 | 65.4 | 61.14 | 66.04 | 69.02 | 66.3 | 60.86 | 62.36 | 60.18 | 53.88 | 50.2 | 52.53 | 55.53 | 51.89 | 51.78 | 55.81 | 51.46 | 57.06 | 53.14 | 54.75 | 55.28 | 54.53 | 49.05 | 42.31 | 49 | 53.05 | 39.29 | 26.95 | 20.11 | 20.2 | 41.64 | 55.18 | 47.34 | 41 | 44.93 | 59.03 | 47.87 | 44.7 | 43.81 | 49.52 | 42.12 | 47.45 | 47.09 | 48.14 | 48.31 | 52.92 | 40.23 | 45.76 | 41.44 |
| [BGE-Multilingual-Gemma2](https://huggingface.co/BAAI/bge-multilingual-gemma2) | 54.46 | 72.8 | 68.64 | **63.42** | **69.48** | **67.91** | **71.79** | **67.57** | **71.28** | **75.39** | **68.91** | **68.29** | **66.78** | **64.15** | 56.48 | 53.04 | **59.97** | **59.68** | **57.72** | **58.2** | **62.43** | **59.54** | **64.5** | **60** | **60.26** | 59.64 | **60.12** | 47.48 | **42.35** | 55.4 | **63.13** | **45.13** | 22.6 | 15.75 | 14.29 | 24 | 44.13 | 50.29 | 43.42 | 48.41 | 58.77 | **52.05** | **49.9** | 43.4 | **56.8** | **44.89** | 50.65 | **51.51** | 51.64 | 51.48 | 50.08 | 39.23 | 50.25 | **51.1** |
| **INF-Retriever-v1** | **54.47** | 73.52 | **69.45** | 63.13 | 61.58 | 66.8 | 69.29 | 63.03 | 69.74 | 69.02 | 68.63 | 63.45 | 64.44 | 62.74 | 57.6 | **56.46** | 58.48 | 53.7 | 55.2 | 57.08 | 53.27 | 57.35 | 55.64 | 58.85 | 59.52 | **60.01** | 58.79 | **57.03** | 41.82 | **55.46** | 57.6 | 43.25 | **34.76** | **21.75** | **21.87** | **51.38** | **59.72** | 52.7 | **49.78** | **49.11** | 43.62 | 51.47 | 49.52 | 40.43 | 54.54 | 38.57 | **51.06** | 51.12 | **53.15** | **51.88** | **59.44** | **44.13** | **50.71** | 44.2 |
## Contributors
### Supervisors
Wei Chu • Yinghui Xu • Yuan Qi
### INF memory team
Junhan Yang (junhanyang@inftech.ai) • Jiahe Wan • Yichen Yao (eason.yyc@inftech.ai)
## Citation
If you find our model useful, please consider citing:
```
@misc {infly-ai_2025,
author = { Junhan Yang, Jiahe Wan, Yichen Yao, Wei Chu, Yinghui Xu, Yuan Qi },
title = { inf-retriever-v1 (Revision 5f469d7) },
year = 2025,
url = { https://huggingface.co/infly/inf-retriever-v1 },
doi = { 10.57967/hf/4262 },
publisher = { Hugging Face }
}
``` | [
"SCIFACT"
] |
EleutherAI/pythia-12b-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:the_pile",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-16T19:03:14Z" | 2023-03-29T18:46:38+00:00 | 558 | 21 | ---
datasets:
- the_pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-12B
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change over the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-12B for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-12B as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-12B has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-12B will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-12B to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-12B may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-12B.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-12B.
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
EleutherAI/pythia-6.9b-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-18T03:04:37Z" | 2023-07-10T01:30:05+00:00 | 558 | 20 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-6.9B-deduped
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change in the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-6.9B-deduped for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-6.9B-deduped as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-6.9B-deduped has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-6.9B-deduped will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-6.9B-deduped to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-6.9B-deduped may produce socially unacceptable or undesirable text,
*even if* the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-6.9B-deduped.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
Pythia-6.9B-deduped was trained on the Pile **after the dataset has been
globally deduplicated**.<br>
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge – Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 13B | 12B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
EleutherAI/pythia-410m-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-11-01T00:48:44Z" | 2023-07-10T01:31:39+00:00 | 552 | 6 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-410M-deduped
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change in the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-410M-deduped for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-410M-deduped as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-410M-deduped has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-410M-deduped will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-410M-deduped to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-410M-deduped may produce socially unacceptable or undesirable text,
*even if* the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-410M-deduped.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
Pythia-410M-deduped was trained on the Pile **after the dataset has been
globally deduplicated**.<br>
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge – Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
mradermacher/Einstein-v7-Qwen2-7B-GGUF | mradermacher | null | [
"transformers",
"gguf",
"axolotl",
"instruct",
"finetune",
"chatml",
"gpt4",
"synthetic data",
"science",
"physics",
"chemistry",
"biology",
"math",
"qwen",
"qwen2",
"en",
"dataset:allenai/ai2_arc",
"dataset:camel-ai/physics",
"dataset:camel-ai/chemistry",
"dataset:camel-ai/biology",
"dataset:camel-ai/math",
"dataset:metaeval/reclor",
"dataset:openbookqa",
"dataset:mandyyyyii/scibench",
"dataset:derek-thomas/ScienceQA",
"dataset:TIGER-Lab/ScienceEval",
"dataset:jondurbin/airoboros-3.2",
"dataset:LDJnr/Capybara",
"dataset:Cot-Alpaca-GPT4-From-OpenHermes-2.5",
"dataset:STEM-AI-mtl/Electrical-engineering",
"dataset:knowrohit07/saraswati-stem",
"dataset:sablo/oasst2_curated",
"dataset:lmsys/lmsys-chat-1m",
"dataset:TIGER-Lab/MathInstruct",
"dataset:bigbio/med_qa",
"dataset:meta-math/MetaMathQA-40K",
"dataset:piqa",
"dataset:scibench",
"dataset:sciq",
"dataset:Open-Orca/SlimOrca",
"dataset:migtissera/Synthia-v1.3",
"dataset:allenai/WildChat",
"dataset:microsoft/orca-math-word-problems-200k",
"dataset:openchat/openchat_sharegpt4_dataset",
"dataset:teknium/GPTeacher-General-Instruct",
"dataset:m-a-p/CodeFeedback-Filtered-Instruction",
"dataset:totally-not-an-llm/EverythingLM-data-V3",
"dataset:HuggingFaceH4/no_robots",
"dataset:OpenAssistant/oasst_top1_2023-08-25",
"dataset:WizardLM/WizardLM_evol_instruct_70k",
"dataset:abacusai/SystemChat-1.1",
"dataset:H-D-T/Buzz-V1.2",
"base_model:Weyaxi/Einstein-v7-Qwen2-7B",
"base_model:quantized:Weyaxi/Einstein-v7-Qwen2-7B",
"license:other",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-06-26T04:29:26Z" | 2024-06-26T13:43:00+00:00 | 552 | 0 | ---
base_model: Weyaxi/Einstein-v7-Qwen2-7B
datasets:
- allenai/ai2_arc
- camel-ai/physics
- camel-ai/chemistry
- camel-ai/biology
- camel-ai/math
- metaeval/reclor
- openbookqa
- mandyyyyii/scibench
- derek-thomas/ScienceQA
- TIGER-Lab/ScienceEval
- jondurbin/airoboros-3.2
- LDJnr/Capybara
- Cot-Alpaca-GPT4-From-OpenHermes-2.5
- STEM-AI-mtl/Electrical-engineering
- knowrohit07/saraswati-stem
- sablo/oasst2_curated
- lmsys/lmsys-chat-1m
- TIGER-Lab/MathInstruct
- bigbio/med_qa
- meta-math/MetaMathQA-40K
- openbookqa
- piqa
- metaeval/reclor
- derek-thomas/ScienceQA
- scibench
- sciq
- Open-Orca/SlimOrca
- migtissera/Synthia-v1.3
- TIGER-Lab/ScienceEval
- allenai/WildChat
- microsoft/orca-math-word-problems-200k
- openchat/openchat_sharegpt4_dataset
- teknium/GPTeacher-General-Instruct
- m-a-p/CodeFeedback-Filtered-Instruction
- totally-not-an-llm/EverythingLM-data-V3
- HuggingFaceH4/no_robots
- OpenAssistant/oasst_top1_2023-08-25
- WizardLM/WizardLM_evol_instruct_70k
- abacusai/SystemChat-1.1
- H-D-T/Buzz-V1.2
language:
- en
library_name: transformers
license: other
tags:
- axolotl
- instruct
- finetune
- chatml
- gpt4
- synthetic data
- science
- physics
- chemistry
- biology
- math
- qwen
- qwen2
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/Weyaxi/Einstein-v7-Qwen2-7B
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q2_K.gguf) | Q2_K | 3.1 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.IQ3_XS.gguf) | IQ3_XS | 3.4 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q3_K_S.gguf) | Q3_K_S | 3.6 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.IQ3_S.gguf) | IQ3_S | 3.6 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.IQ3_M.gguf) | IQ3_M | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q3_K_M.gguf) | Q3_K_M | 3.9 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q3_K_L.gguf) | Q3_K_L | 4.2 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.IQ4_XS.gguf) | IQ4_XS | 4.4 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q4_K_S.gguf) | Q4_K_S | 4.6 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q4_K_M.gguf) | Q4_K_M | 4.8 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q5_K_S.gguf) | Q5_K_S | 5.4 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q5_K_M.gguf) | Q5_K_M | 5.5 | |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q6_K.gguf) | Q6_K | 6.4 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.Q8_0.gguf) | Q8_0 | 8.2 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Einstein-v7-Qwen2-7B-GGUF/resolve/main/Einstein-v7-Qwen2-7B.f16.gguf) | f16 | 15.3 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
| [
"SCIQ"
] |
speechbrain/asr-streaming-conformer-gigaspeech | speechbrain | automatic-speech-recognition | [
"speechbrain",
"automatic-speech-recognition",
"Transducer",
"Conformer",
"pytorch",
"en",
"dataset:speechcolab/gigaspeech",
"arxiv:2407.00463",
"arxiv:2106.04624",
"license:apache-2.0",
"model-index",
"region:us"
] | "2024-11-06T13:04:19Z" | 2024-11-08T13:21:36+00:00 | 552 | 3 | ---
datasets:
- speechcolab/gigaspeech
language:
- en
license: apache-2.0
metrics:
- wer
tags:
- automatic-speech-recognition
- Transducer
- Conformer
- pytorch
- speechbrain
inference: false
model-index:
- name: Conformer-Transducer by SpeechBrain
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: GigaSpeech
type: gigaspeech
split: test
args:
language: en
metrics:
- type: wer
value: 11.00%
name: Test WER (non-streaming greedy)
- type: wer
value: 11.53%
name: Test WER (960ms chunk size, 4 left context chunks)
---
<iframe src="https://ghbtns.com/github-btn.html?user=speechbrain&repo=speechbrain&type=star&count=true&size=large&v=2" frameborder="0" scrolling="0" width="170" height="30" title="GitHub"></iframe>
<br/><br/>
# Conformer for GigaSpeech
This repository provides all the necessary tools to perform automatic speech
recognition from an end-to-end system pretrained on GigaSpeech (EN, XL split) within
SpeechBrain. For a better experience, we encourage you to learn more about
[SpeechBrain](https://speechbrain.github.io).
The performance of the model in full context mode (no streaming) is the following:
| Release | Test WER | GPUs |
|:-------------:|:--------------:|:--------:|
| 2024-11-08 | 11.00% | 4xA100 40GB |
With streaming, the results with different chunk sizes on the test split are the following:
| | full | cs=32 (1280ms) | 24 (960ms) | 16 (640ms) | 12 (480ms) | 8 (320ms) |
|:-----:|:------:|:------:|:------:|:------:|:------:|:------:|
| full | 11.00% | - | - | - | - | - |
| 16 | - | - | - | 11.70% | 11.84% | 12.14% |
| 8 | - | - | 11.50% | 11.72% | 11.88% | 12.28% |
| 4 | - | 11.40% | 11.53% | 11.81% | 12.03% | 12.64% |
| 2 | - | 11.46% | 11.67% | 12.03% | 12.43% | 13.25% |
| 1\* | - | 11.59% | 11.85% | 12.39% | 12.93% | 14.13% |
(\*: model was never explicitly trained with this setting)
When comparing configurations, bear in mind that the utterances of the test set are of a finite length.
All WER values were evaluated using [this script](https://gist.github.com/asumagic/66aaf43156c31866cf2ebcffda46ea01) (until we integrate a nicer way to evaluate streaming WER in different conditions).
## Pipeline description
This ASR system is a Conformer model trained with the RNN-T loss (with an auxiliary CTC loss to stabilize training). The model operates with a unigram tokenizer.
Architecture details are described in the [training hyperparameters file](https://github.com/speechbrain/speechbrain/blob/develop/recipes/GigaSpeech/ASR/transducer/hparams/conformer_transducer.yaml).
Streaming support makes use of Dynamic Chunk Training. Chunked attention is used for the multi-head attention module, and an implementation of [Dynamic Chunk Convolutions](https://www.amazon.science/publications/dynamic-chunk-convolution-for-unified-streaming-and-non-streaming-conformer-asr) was used.
The model was trained with support for different chunk sizes (and even full context), and so is suitable for various chunk sizes and offline transcription.
The system is trained with recordings sampled at 16kHz (single channel).
The code will automatically normalize your audio (i.e., resampling + mono channel selection) when calling `transcribe_file` if needed.
## Install SpeechBrain
First of all, please install SpeechBrain with the following command:
```
pip install speechbrain
```
Please notice that we encourage you to read our tutorials and learn more about
[SpeechBrain](https://speechbrain.github.io).
### Transcribing your own audio files (in English)
```python
from speechbrain.inference.ASR import StreamingASR
from speechbrain.utils.dynamic_chunk_training import DynChunkTrainConfig
asr_model = StreamingASR.from_hparams("speechbrain/asr-streaming-conformer-gigaspeech")
asr_model.transcribe_file(
"speechbrain/asr-streaming-conformer-librispeech/test-en.wav",
# select a chunk size of ~960ms with 4 chunks of left context
DynChunkTrainConfig(24, 4),
# disable torchaudio streaming to allow fetching from HuggingFace
# set this to True for your own files or streams to allow for streaming file decoding
use_torchaudio_streaming=False,
)
```
The `DynChunkTrainConfig` values can be adjusted for a tradeoff of latency, computational power and transcription accuracy. Refer to the streaming WER table to pick a value that is suitable for your usecase.
<details>
<summary>Commandline tool to transcribe a file or a live stream</summary>
**Decoding from a live stream using ffmpeg (BBC Radio 4):**
`python3 asr.py http://as-hls-ww-live.akamaized.net/pool_904/live/ww/bbc_radio_fourfm/bbc_radio_fourfm.isml/bbc_radio_fourfm-audio%3d96000.norewind.m3u8 --model-source=speechbrain/asr-streaming-conformer-gigaspeech --device=cpu -v`
**Decoding from a file:**
`python3 asr.py some-english-speech.wav --model-source=speechbrain/asr-streaming-conformer-gigaspeech --device=cpu -v`
```python
from argparse import ArgumentParser
import logging
parser = ArgumentParser()
parser.add_argument("audio_path")
parser.add_argument("--model-source", required=True)
parser.add_argument("--device", default="cpu")
parser.add_argument("--ip", default="127.0.0.1")
parser.add_argument("--port", default=9431)
parser.add_argument("--chunk-size", default=24, type=int)
parser.add_argument("--left-context-chunks", default=4, type=int)
parser.add_argument("--num-threads", default=None, type=int)
parser.add_argument("--verbose", "-v", default=False, action="store_true")
args = parser.parse_args()
if args.verbose:
logging.getLogger().setLevel(logging.INFO)
logging.info("Loading libraries")
from speechbrain.inference.ASR import StreamingASR
from speechbrain.utils.dynamic_chunk_training import DynChunkTrainConfig
import torch
device = args.device
if args.num_threads is not None:
torch.set_num_threads(args.num_threads)
logging.info(f"Loading model from \"{args.model_source}\" onto device {device}")
asr = StreamingASR.from_hparams(args.model_source, run_opts={"device": device})
config = DynChunkTrainConfig(args.chunk_size, args.left_context_chunks)
logging.info(f"Starting stream from URI \"{args.audio_path}\"")
for text_chunk in asr.transcribe_file_streaming(args.audio_path, config):
print(text_chunk, flush=True, end="")
```
</details>
<details>
<summary>Live ASR decoding from a browser using Gradio</summary>
We want to optimize some things around the model before we create a proper HuggingFace space demonstrating live streaming on CPU.
In the mean time, this is a simple hacky demo of live ASR in the browser using Gradio's live microphone streaming feature.
If you run this, please note:
- Modern browsers refuse to stream microphone input over an untrusted connection (plain HTTP), unless it is localhost. If you are running this on a remote server, you could use SSH port forwarding to expose the remote's port on your machine.
- Streaming using Gradio on Firefox seems to cause some issues. Chromium-based browsers seem to behave better.
Run using:
`python3 gradio-asr.py --model-source speechbrain/asr-streaming-conformer-gigaspeech --ip=localhost --device=cpu`
```python
from argparse import ArgumentParser
from dataclasses import dataclass
import logging
parser = ArgumentParser()
parser.add_argument("--model-source", required=True)
parser.add_argument("--device", default="cpu")
parser.add_argument("--ip", default="127.0.0.1")
parser.add_argument("--port", default=9431)
parser.add_argument("--chunk-size", default=24, type=int)
parser.add_argument("--left-context-chunks", default=4, type=int)
parser.add_argument("--num-threads", default=None, type=int)
parser.add_argument("--verbose", "-v", default=False, action="store_true")
args = parser.parse_args()
if args.verbose:
logging.getLogger().setLevel(logging.INFO)
logging.info("Loading libraries")
from speechbrain.inference.ASR import StreamingASR, ASRStreamingContext
from speechbrain.utils.dynamic_chunk_training import DynChunkTrainConfig
import torch
import gradio as gr
import torchaudio
import numpy as np
device = args.device
if args.num_threads is not None:
torch.set_num_threads(args.num_threads)
logging.info(f"Loading model from \"{args.model_source}\" onto device {device}")
asr = StreamingASR.from_hparams(args.model_source, run_opts={"device": device})
config = DynChunkTrainConfig(args.chunk_size, args.left_context_chunks)
@dataclass
class GradioStreamingContext:
context: ASRStreamingContext
chunk_size: int
waveform_buffer: torch.Tensor
decoded_text: str
def transcribe(stream, new_chunk):
sr, y = new_chunk
y = y.astype(np.float32)
y = torch.tensor(y, dtype=torch.float32, device=device)
y /= max(1, torch.max(torch.abs(y)).item()) # norm by max abs() within chunk & avoid NaN
if len(y.shape) > 1:
y = torch.mean(y, dim=1) # downmix to mono
# HACK: we are making poor use of the resampler across chunk boundaries
# which may degrade accuracy.
# NOTE: we should also absolutely avoid recreating a resampler every time
resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=asr.audio_normalizer.sample_rate).to(device)
y = resampler(y) # janky resample (probably to 16kHz)
if stream is None:
stream = GradioStreamingContext(
context=asr.make_streaming_context(config),
chunk_size=asr.get_chunk_size_frames(config),
waveform_buffer=y,
decoded_text="",
)
else:
stream.waveform_buffer = torch.concat((stream.waveform_buffer, y))
while stream.waveform_buffer.size(0) > stream.chunk_size:
chunk = stream.waveform_buffer[:stream.chunk_size]
stream.waveform_buffer = stream.waveform_buffer[stream.chunk_size:]
# fake batch dim
chunk = chunk.unsqueeze(0)
# list of transcribed strings, of size 1 because the batch size is 1
with torch.no_grad():
transcribed = asr.transcribe_chunk(stream.context, chunk)
stream.decoded_text += transcribed[0]
return stream, stream.decoded_text
# NOTE: latency seems relatively high, which may be due to this:
# https://github.com/gradio-app/gradio/issues/6526
demo = gr.Interface(
transcribe,
["state", gr.Audio(sources=["microphone"], streaming=True)],
["state", "text"],
live=True,
)
demo.launch(server_name=args.ip, server_port=args.port)
```
</details>
### Inference on GPU
To perform inference on the GPU, add `run_opts={"device":"cuda"}` when calling the `from_hparams` method.
## Parallel Inference on a Batch
Currently, the high level transcription interfaces do not support batched inference, but the low-level interfaces (i.e. `encode_chunk`) do.
We hope to provide efficient functionality for this in the future.
### Training
The model was trained with SpeechBrain `v1.0.2`.
To train it from scratch, follow these steps:
1. Clone SpeechBrain:
```bash
git clone https://github.com/speechbrain/speechbrain/
```
2. Install it:
```bash
cd speechbrain
pip install -r requirements.txt
pip install -e .
```
3. Follow the steps listed in the [README](https://github.com/speechbrain/speechbrain/blob/develop/recipes/GigaSpeech/ASR/transducer/README.md) for this recipe.
### Limitations
The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
# **About SpeechBrain**
- Website: https://speechbrain.github.io/
- Code: https://github.com/speechbrain/speechbrain/
- HuggingFace: https://huggingface.co/speechbrain/
# **Citing SpeechBrain**
Please, cite SpeechBrain if you use it for your research or business.
```bibtex
@misc{speechbrainV1,
title={Open-Source Conversational AI with SpeechBrain 1.0},
author={Mirco Ravanelli and Titouan Parcollet and Adel Moumen and Sylvain de Langen and Cem Subakan and Peter Plantinga and Yingzhi Wang and Pooneh Mousavi and Luca Della Libera and Artem Ploujnikov and Francesco Paissan and Davide Borra and Salah Zaiem and Zeyu Zhao and Shucong Zhang and Georgios Karakasidis and Sung-Lin Yeh and Pierre Champion and Aku Rouhe and Rudolf Braun and Florian Mai and Juan Zuluaga-Gomez and Seyed Mahed Mousavi and Andreas Nautsch and Xuechen Liu and Sangeet Sagar and Jarod Duret and Salima Mdhaffar and Gaelle Laperriere and Mickael Rouvier and Renato De Mori and Yannick Esteve},
year={2024},
eprint={2407.00463},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2407.00463},
}
@misc{speechbrain,
title={{SpeechBrain}: A General-Purpose Speech Toolkit},
author={Mirco Ravanelli and Titouan Parcollet and Peter Plantinga and Aku Rouhe and Samuele Cornell and Loren Lugosch and Cem Subakan and Nauman Dawalatabad and Abdelwahab Heba and Jianyuan Zhong and Ju-Chieh Chou and Sung-Lin Yeh and Szu-Wei Fu and Chien-Feng Liao and Elena Rastorgueva and François Grondin and William Aris and Hwidong Na and Yan Gao and Renato De Mori and Yoshua Bengio},
year={2021},
eprint={2106.04624},
archivePrefix={arXiv},
primaryClass={eess.AS},
note={arXiv:2106.04624}
}
```
| [
"BEAR"
] |
aisingapore/gemma2-9b-cpt-sea-lionv3-base | aisingapore | text-generation | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"en",
"zh",
"vi",
"id",
"th",
"fil",
"ta",
"ms",
"km",
"lo",
"my",
"arxiv:2309.06085",
"arxiv:2101.09635",
"base_model:google/gemma-2-9b",
"base_model:finetune:google/gemma-2-9b",
"license:gemma",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2024-10-30T03:21:35Z" | 2024-12-19T12:56:00+00:00 | 550 | 2 | ---
base_model: google/gemma-2-9b
language:
- en
- zh
- vi
- id
- th
- fil
- ta
- ms
- km
- lo
- my
library_name: transformers
license: gemma
pipeline_tag: text-generation
---
<div>
<img src="gemma_2_9b_sea-lion_v3_base_banner.png"/>
</div>
# Gemma2 9B CPT SEA-LIONv3
SEA-LION is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for the Southeast Asia (SEA) region.
Gemma2 9B CPT SEA-LIONv3 Base is a multilingual model which has undergone continued pre-training on approximately **200B** tokens across the 11 official Southeast Asian languages: English, Chinese, Vietnamese, Indonesian, Thai, Tamil, Filipino, Malay, Khmer, Lao, Burmese.
SEA-LION stands for <i>Southeast Asian Languages In One Network</i>.
- **Developed by:** Products Pillar, AI Singapore
- **Funded by:** Singapore NRF
- **Model type:** Decoder
- **Languages supported:** Burmese, Chinese, English, Filipino, Indonesia, Khmer, Lao, Malay, Tamil, Thai, Vietnamese
- **License:** [Gemma Community License](https://ai.google.dev/gemma/terms)
## Model Details
### Model Description
We performed continued pre-training in English and ASEAN languages on [Gemma-2-9B](https://huggingface.co/google/gemma-2-9b), a decoder model using the Gemma 2 architecture, to create Gemma2 9B CPT SEA-LIONv3 Base.
For tokenisation, the model employs the default tokenizer used in Gemma 2 9B.
### Benchmark Performance
We evaluated Gemma2 9B CPT SEA-LIONv3 base model on general language capabilities.
#### General Language Capabilities
For the evaluation of general language capabilities, we employed the [SEA HELM (also known as BHASA) evaluation benchmark](https://arxiv.org/abs/2309.06085v2) across a variety of tasks.
These tasks include Question Answering (QA), Sentiment Analysis (Sentiment), Toxicity Detection (Toxicity), Translation in both directions (Eng>Lang & Lang>Eng), Abstractive Summarization (Summ), Causal Reasoning (Causal) and Natural Language Inference (NLI).
Note: SEA HELM is implemented using prompts to elicit answers in a strict format. For all tasks, the model is expected to provide an answer tag from which the answer is automatically extracted. For tasks where options are provided, the answer should comprise one of the pre-defined options. The scores for each task is normalised to account for baseline performance due to random chance.
The evaluation was done **five-shot** with native prompts on a sample of 100-1000 instances for each dataset.
For more details on Gemma2 9B CPT SEA-LIONv3 base benchmark performance, please refer to the SEA HELM leaderboard, https://leaderboard.sea-lion.ai/
## Technical Specifications
### Infrastructure
Gemma2 9B CPT SEA-LIONv3 was trained using [MosaicML Composer](https://github.com/mosaicml/composer) on the following hardware:
| Training Details | Gemma2 9B CPT SEA-LIONv3 |
|----------------------|:------------------------:|
| SingTel HGX-100 | 8 instances |
| Nvidia H100 80GB GPU | 64 |
| Training Duration | 10 days |
### Configuration
| HyperParameter | Gemma2 9B CPT SEA-LIONv3 |
|-------------------|:------------------------:|
| Precision | bfloat16 |
| Optimizer | decoupled_adamw |
| Scheduler | weight_stable_decay |
| Learning Rate | 1.0e-5 |
| Global Batch Size | 512 |
| Micro Batch Size | 1 |
## Data
Gemma2 9B CPT SEA-LIONv3 base model was continued pre-trained on 200B tokens of the following data:
| Language | Source | Total Tokens (B) | Percentage (%) | Total percentage (%) |
| ------------------------ | ---------------- | ---------------- | -------------- | -------------------- |
| Code | StackV2 | 40 | 20 | 20 |
| English | Dolma | 37.5 | 18.75 | 25 |
| | Fineweb-Edu | 7.5 | 3.75 |
| | Others | 5 | 2.5 |
| Chinese | SEA-LION Pile v1 | 12 | 6 | 13 |
| | Others | 14 | 7 |
| Vietnamese | SEA-LION Pile v1 | 8.4 | 4.2 | 13 |
| | VinBigData | 16 | 8 |
| | Others | 1.6 | 0.8 |
| Indonesian | SEA-LION Pile v1 | 7 | 3.5 | 13 |
| | SEA-LION Pile v2 | 7 | 3.5 |
| | Others | 12 | 6 |
| Thai | SEA-LION Pile v1 | 10.7 | 5.35 | 10 |
| | WangChanBERTa | 8.5 | 4.25 |
| | Others | 0.8 | 0.4 |
| Filipino - Malay - Tamil | SEA-LION Pile v1 | 4.28 | 2.14 | 3 |
| | Others | 1.72 | 0.86 |
| Khmer - Lao - Burmese | SEA-LION Pile v1 | 5.2 | 2.6 | 3 |
| | Others | 0.8 | 0.4 |
Note:
- All token counts are counted using Gemma 2 9B tokenizer
- SEA-LION Pile v1 is processed from Common Crawl WET, which is published [here](https://huggingface.co/datasets/aisingapore/sea-lion-pile). The cutoff date of this version is September 2020.
- SEA-LION Pile v2 is processed from Common Crawl WARC from October 2020 to April 2024.
- Tamil news is sourced with permission from [Seithi](https://seithi.mediacorp.sg/)
## Call for Contributions
We encourage researchers, developers, and language enthusiasts to actively contribute to the enhancement and expansion of SEA-LION. Contributions can involve identifying and reporting bugs, sharing pre-training, instruction, and preference data, improving documentation usability, proposing and implementing new model evaluation tasks and metrics, or training versions of the model in additional Southeast Asian languages. Join us in shaping the future of SEA-LION by sharing your expertise and insights to make these models more accessible, accurate, and versatile. Please check out our GitHub for further information on the call for contributions.
## The Team
Chan Adwin, Cheng Nicholas, Choa Esther, Huang Yuli, Hulagadri Adithya Venkatadri, Lau Wayne, Lee Chwan Ren, Leong Wai Yi, Leong Wei Qi, Limkonchotiwat Peerat, Liu Bing Jie Darius, Montalan Jann Railey, Ng Boon Cheong Raymond, Ngui Jian Gang, Nguyen Thanh Ngan, Ong Brandon, Ong Tat-Wee David, Ong Zhi Hao, Rengarajan Hamsawardhini, Siow Bryan, Susanto Yosephine, Tai Ngee Chia, Tan Choon Meng, Teng Walter, Teo Eng Sipp Leslie, Teo Wei Yi, Tjhi William, Yeo Yeow Tong, Yong Xianbin
## Acknowledgements
[AI Singapore](https://aisingapore.org/) is a national programme supported by the National Research Foundation, Singapore and hosted by the National University of Singapore. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of the National Research Foundation or the National University of Singapore.
## Contact
For more info, please contact us using this [SEA-LION Inquiry Form.](https://forms.gle/sLCUVb95wmGf43hi6)
[Link to SEA-LION's GitHub repository.](https://github.com/aisingapore/sealion)
## Disclaimer
This is the repository for the commercial instruction-tuned model.
The model has _not_ been aligned for safety.
Developers and users should perform their own safety fine-tuning and related security measures.
In no event shall the authors be held liable for any claims, damages, or other liabilities arising from the use of the released weights and codes.
## References
### Thai Pre-Training Data Reference
```bibtex
@misc{lowphansirikul2021wangchanberta,
title={WangchanBERTa: Pretraining transformer-based Thai Language Models},
author={Lalita Lowphansirikul and Charin Polpanumas and Nawat Jantrakulchai and Sarana Nutanong},
year={2021},
eprint={2101.09635},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` | [
"CHIA"
] |
unikei/distilbert-base-re-punctuate | unikei | token-classification | [
"transformers",
"pytorch",
"distilbert",
"token-classification",
"biology",
"medical",
"en",
"dataset:bigbio/drugprot",
"dataset:bigbio/ncbi_disease",
"license:bigscience-openrail-m",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2023-08-17T12:28:13Z" | 2023-09-13T09:01:41+00:00 | 547 | 8 | ---
datasets:
- bigbio/drugprot
- bigbio/ncbi_disease
language:
- en
license: bigscience-openrail-m
pipeline_tag: token-classification
tags:
- biology
- medical
widget:
- text: the atm protein is a single high molecular weight protein predominantly confined
to the nucleus of human fibroblasts but is present in both nuclear and microsomal
fractions from human lymphoblast cells and peripheral blood lymphocytes atm protein
levels and localization remain constant throughout all stages of the cell cycle
truncated atm protein was not detected in lymphoblasts from ataxia telangiectasia
patients homozygous for mutations leading to premature protein termination exposure
of normal human cells to gamma irradiation and the radiomimetic drug neocarzinostatin
had no effect on atm protein levels in contrast to a noted rise in p53 levels
over the same time interval these findings are consistent with a role for the
atm protein in ensuring the fidelity of dna repair and cell cycle regulation following
genome damage
---
# DistilBERT base model for restoring punctuation of medical/biotech speech-to-text transcripts
E.g.:
```
the atm protein is a single high molecular weight protein predominantly confined to the nucleus of human
fibroblasts but is present in both nuclear and microsomal fractions from human lymphoblast cells and peripheral
blood lymphocytes atm protein levels and localization remain constant throughout all stages of the cell cycle
truncated atm protein was not detected in lymphoblasts from ataxia telangiectasia patients homozygous
for mutations leading to premature protein termination exposure of normal human cells to gamma irradiation and the
radiomimetic drug neocarzinostatin had no effect on atm protein levels in contrast to a noted rise in p53 levels
over the same time interval these findings are consistent with a role for the atm protein in ensuring the fidelity
of dna repair and cell cycle regulation following genome damage
```
will be punctuated as follows:
```
The ATM protein is a single, high-molecular-weight protein predominantly confined to the nucleus of human
fibroblasts, but is present in both nuclear and microsomal fractions from human lymphoblast cells and peripheral
blood lymphocytes. ATM protein levels and localization remain constant throughout all stages of the cell cycle.
Truncated ATM protein was not detected in lymphoblasts from ataxia-telangiectasia-patients homozygous
for mutations leading to premature protein termination. Exposure of normal human cells to gamma-irradiation and the
radiomimetic drug neocarzinostatin had no effect on ATM protein levels, in contrast to a noted rise in p53 levels
over the same time interval. These findings are consistent with a role for the ATM protein in ensuring the fidelity
of DNA repair and cell-cycle regulation following genome damage.
```
## How to use it in your code:
```python
import torch
import numpy as np
from transformers import DistilBertTokenizerFast, DistilBertForTokenClassification
checkpoint = "unikei/distilbert-base-re-punctuate"
tokenizer = DistilBertTokenizerFast.from_pretrained(checkpoint)
model = DistilBertForTokenClassification.from_pretrained(checkpoint)
encoder_max_length = 256
#
# Split text to segments of length 200, with overlap 50
#
def split_to_segments(wrds, length, overlap):
resp = []
i = 0
while True:
wrds_split = wrds[(length * i):((length * (i + 1)) + overlap)]
if not wrds_split:
break
resp_obj = {
"text": wrds_split,
"start_idx": length * i,
"end_idx": (length * (i + 1)) + overlap,
}
resp.append(resp_obj)
i += 1
return resp
#
# Punctuate wordpieces
#
def punctuate_wordpiece(wordpiece, label):
if label.startswith('UPPER'):
wordpiece = wordpiece.upper()
elif label.startswith('Upper'):
wordpiece = wordpiece[0].upper() + wordpiece[1:]
if label[-1] != '_' and label[-1] != wordpiece[-1]:
wordpiece += label[-1]
return wordpiece
#
# Punctuate text segments (200 words)
#
def punctuate_segment(wordpieces, word_ids, labels, start_word):
result = ''
for idx in range(0, len(wordpieces)):
if word_ids[idx] == None:
continue
if word_ids[idx] < start_word:
continue
wordpiece = punctuate_wordpiece(wordpieces[idx][2:] if wordpieces[idx].startswith('##') else wordpieces[idx],
labels[idx])
if idx > 0 and len(result) > 0 and word_ids[idx] != word_ids[idx - 1] and result[-1] != '-':
result += ' '
result += wordpiece
return result
#
# Tokenize, predict, punctuate text segments (200 words)
#
def process_segment(words, tokenizer, model, start_word):
tokens = tokenizer(words['text'],
padding="max_length",
# truncation=True,
max_length=encoder_max_length,
is_split_into_words=True, return_tensors='pt')
with torch.no_grad():
logits = model(**tokens).logits
logits = logits.cpu()
predictions = np.argmax(logits, axis=-1)
wordpieces = tokens.tokens()
word_ids = tokens.word_ids()
id2label = model.config.id2label
labels = [[id2label[p.item()] for p in prediction] for prediction in predictions][0]
return punctuate_segment(wordpieces, word_ids, labels, start_word)
#
# Punctuate text of any length
#
def punctuate(text, tokenizer, model):
text = text.lower()
text = text.replace('\n', ' ')
words = text.split(' ')
overlap = 50
slices = split_to_segments(words, 150, 50)
result = ""
start_word = 0
for text in slices:
corrected = process_segment(text, tokenizer, model, start_word)
result += corrected + ' '
start_word = overlap
return result
#
# Example
#
text = "the atm protein is a single high molecular weight protein predominantly confined to the nucleus of human fibroblasts but is present in both nuclear and microsomal fractions from human lymphoblast cells and peripheral blood lymphocytes atm protein levels and localization remain constant throughout all stages of the cell cycle truncated atm protein was not detected in lymphoblasts from ataxia telangiectasia patients homozygous for mutations leading to premature protein termination exposure of normal human cells to gamma irradiation and the radiomimetic drug neocarzinostatin had no effect on atm protein levels in contrast to a noted rise in p53 levels over the same time interval these findings are consistent with a role for the atm protein in ensuring the fidelity of dna repair and cell cycle regulation following genome damage"
result = punctuate(text, tokenizer, model)
print(result)
"""
Output:
The ATM protein is a single, high-molecular-weight protein predominantly confined to the nucleus of human fibroblasts, but is present in both nuclear and microsomal fractions from human lymphoblast cells and peripheral blood lymphocytes. ATM protein levels and localization remain constant throughout all stages of the cell cycle. Truncated ATM protein was not detected in lymphoblasts from ataxia-telangiectasia-patients homozygous for mutations leading to premature protein termination. Exposure of normal human cells to gamma-irradiation and the radiomimetic drug neocarzinostatin had no effect on ATM protein levels, in contrast to a noted rise in p53 levels over the same time interval. These findings are consistent with a role for the ATM protein in ensuring the fidelity of DNA repair and cell-cycle regulation following genome damage.
"""
``` | [
"DRUGPROT",
"NCBI DISEASE"
] |
neuralmagic/bge-base-en-v1.5-quant | neuralmagic | feature-extraction | [
"transformers",
"onnx",
"bert",
"feature-extraction",
"mteb",
"sparse",
"sparsity",
"quantized",
"embeddings",
"int8",
"deepsparse",
"en",
"license:mit",
"model-index",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2023-10-03T12:45:42Z" | 2023-11-13T17:04:42+00:00 | 547 | 4 | ---
language:
- en
license: mit
tags:
- mteb
- sparse
- sparsity
- quantized
- onnx
- embeddings
- int8
- deepsparse
model-index:
- name: bge-base-en-v1.5-quant
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 76.16417910447761
- type: ap
value: 39.62965026785565
- type: f1
value: 70.30041589476463
- task:
type: Classification
dataset:
name: MTEB AmazonPolarityClassification
type: mteb/amazon_polarity
config: default
split: test
revision: e2d317d38cd51312af73b3d32a06d1a08b442046
metrics:
- type: accuracy
value: 92.95087500000001
- type: ap
value: 89.92451248271642
- type: f1
value: 92.94162732408543
- task:
type: Classification
dataset:
name: MTEB AmazonReviewsClassification (en)
type: mteb/amazon_reviews_multi
config: en
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 48.214
- type: f1
value: 47.57084372829096
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringP2P
type: mteb/arxiv-clustering-p2p
config: default
split: test
revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d
metrics:
- type: v_measure
value: 48.499816497755646
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringS2S
type: mteb/arxiv-clustering-s2s
config: default
split: test
revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53
metrics:
- type: v_measure
value: 42.006939120636034
- task:
type: Reranking
dataset:
name: MTEB AskUbuntuDupQuestions
type: mteb/askubuntudupquestions-reranking
config: default
split: test
revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
metrics:
- type: map
value: 62.390343953329875
- type: mrr
value: 75.69922613551422
- task:
type: STS
dataset:
name: MTEB BIOSSES
type: mteb/biosses-sts
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cos_sim_pearson
value: 89.03408553833623
- type: cos_sim_spearman
value: 86.71221676053791
- type: euclidean_pearson
value: 87.81477796215844
- type: euclidean_spearman
value: 87.28994076774481
- type: manhattan_pearson
value: 87.76204756059836
- type: manhattan_spearman
value: 87.1971675695072
- task:
type: Classification
dataset:
name: MTEB Banking77Classification
type: mteb/banking77
config: default
split: test
revision: 0fd18e25b25c072e09e0d92ab615fda904d66300
metrics:
- type: accuracy
value: 86.35064935064935
- type: f1
value: 86.32782396028989
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringP2P
type: mteb/biorxiv-clustering-p2p
config: default
split: test
revision: 65b79d1d13f80053f67aca9498d9402c2d9f1f40
metrics:
- type: v_measure
value: 39.299558776859485
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringS2S
type: mteb/biorxiv-clustering-s2s
config: default
split: test
revision: 258694dd0231531bc1fd9de6ceb52a0853c6d908
metrics:
- type: v_measure
value: 35.64603198816062
- task:
type: Classification
dataset:
name: MTEB EmotionClassification
type: mteb/emotion
config: default
split: test
revision: 4f58c6b202a23cf9a4da393831edf4f9183cad37
metrics:
- type: accuracy
value: 51.269999999999996
- type: f1
value: 45.9714399031315
- task:
type: Classification
dataset:
name: MTEB ImdbClassification
type: mteb/imdb
config: default
split: test
revision: 3d86128a09e091d6018b6d26cad27f2739fc2db7
metrics:
- type: accuracy
value: 89.7204
- type: ap
value: 85.70238397381907
- type: f1
value: 89.70961232185473
- task:
type: Classification
dataset:
name: MTEB MTOPDomainClassification (en)
type: mteb/mtop_domain
config: en
split: test
revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf
metrics:
- type: accuracy
value: 93.95120839033288
- type: f1
value: 93.70348712248138
- task:
type: Classification
dataset:
name: MTEB MTOPIntentClassification (en)
type: mteb/mtop_intent
config: en
split: test
revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba
metrics:
- type: accuracy
value: 75.25763793889648
- type: f1
value: 57.59583082574482
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (en)
type: mteb/amazon_massive_intent
config: en
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 75.16476126429052
- type: f1
value: 73.29287381030854
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (en)
type: mteb/amazon_massive_scenario
config: en
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 78.9340954942838
- type: f1
value: 79.04036413238218
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringP2P
type: mteb/medrxiv-clustering-p2p
config: default
split: test
revision: e7a26af6f3ae46b30dde8737f02c07b1505bcc73
metrics:
- type: v_measure
value: 32.80025982143821
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringS2S
type: mteb/medrxiv-clustering-s2s
config: default
split: test
revision: 35191c8c0dca72d8ff3efcd72aa802307d469663
metrics:
- type: v_measure
value: 30.956464446009623
- task:
type: Reranking
dataset:
name: MTEB MindSmallReranking
type: mteb/mind_small
config: default
split: test
revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69
metrics:
- type: map
value: 31.886626060290734
- type: mrr
value: 32.99813843700759
- task:
type: Clustering
dataset:
name: MTEB RedditClustering
type: mteb/reddit-clustering
config: default
split: test
revision: 24640382cdbf8abc73003fb0fa6d111a705499eb
metrics:
- type: v_measure
value: 55.693914682185365
- task:
type: Clustering
dataset:
name: MTEB RedditClusteringP2P
type: mteb/reddit-clustering-p2p
config: default
split: test
revision: 282350215ef01743dc01b456c7f5241fa8937f16
metrics:
- type: v_measure
value: 62.32723620518647
- task:
type: STS
dataset:
name: MTEB SICK-R
type: mteb/sickr-sts
config: default
split: test
revision: a6ea5a8cab320b040a23452cc28066d9beae2cee
metrics:
- type: cos_sim_pearson
value: 84.70275347034692
- type: cos_sim_spearman
value: 80.06126639668393
- type: euclidean_pearson
value: 82.18370726102707
- type: euclidean_spearman
value: 80.05483013524909
- type: manhattan_pearson
value: 82.11962032129463
- type: manhattan_spearman
value: 79.97174232961949
- task:
type: STS
dataset:
name: MTEB STS12
type: mteb/sts12-sts
config: default
split: test
revision: a0d554a64d88156834ff5ae9920b964011b16384
metrics:
- type: cos_sim_pearson
value: 86.08210281025868
- type: cos_sim_spearman
value: 77.75002826042643
- type: euclidean_pearson
value: 83.06487161944293
- type: euclidean_spearman
value: 78.0677956304104
- type: manhattan_pearson
value: 83.04321232787379
- type: manhattan_spearman
value: 78.09582483148635
- task:
type: STS
dataset:
name: MTEB STS13
type: mteb/sts13-sts
config: default
split: test
revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
metrics:
- type: cos_sim_pearson
value: 84.64353592106988
- type: cos_sim_spearman
value: 86.07934653140616
- type: euclidean_pearson
value: 85.21820182954883
- type: euclidean_spearman
value: 86.18828773665395
- type: manhattan_pearson
value: 85.12075207905364
- type: manhattan_spearman
value: 86.12061116344299
- task:
type: STS
dataset:
name: MTEB STS14
type: mteb/sts14-sts
config: default
split: test
revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
metrics:
- type: cos_sim_pearson
value: 84.33571296969136
- type: cos_sim_spearman
value: 82.8868213429789
- type: euclidean_pearson
value: 83.65476643152161
- type: euclidean_spearman
value: 82.76439753890263
- type: manhattan_pearson
value: 83.63348951033883
- type: manhattan_spearman
value: 82.76176495070241
- task:
type: STS
dataset:
name: MTEB STS15
type: mteb/sts15-sts
config: default
split: test
revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
metrics:
- type: cos_sim_pearson
value: 87.6337321089215
- type: cos_sim_spearman
value: 88.54453531860615
- type: euclidean_pearson
value: 87.68754116644199
- type: euclidean_spearman
value: 88.22610830299979
- type: manhattan_pearson
value: 87.62214887890859
- type: manhattan_spearman
value: 88.14766677391091
- task:
type: STS
dataset:
name: MTEB STS16
type: mteb/sts16-sts
config: default
split: test
revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
metrics:
- type: cos_sim_pearson
value: 83.89742747806514
- type: cos_sim_spearman
value: 85.76282302560992
- type: euclidean_pearson
value: 84.83917251074928
- type: euclidean_spearman
value: 85.74354740775905
- type: manhattan_pearson
value: 84.91190952448616
- type: manhattan_spearman
value: 85.82001542154245
- task:
type: STS
dataset:
name: MTEB STS17 (en-en)
type: mteb/sts17-crosslingual-sts
config: en-en
split: test
revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d
metrics:
- type: cos_sim_pearson
value: 87.70974342036347
- type: cos_sim_spearman
value: 87.82200371351459
- type: euclidean_pearson
value: 88.04095125600278
- type: euclidean_spearman
value: 87.5069523002544
- type: manhattan_pearson
value: 88.03247709799281
- type: manhattan_spearman
value: 87.43433979175654
- task:
type: STS
dataset:
name: MTEB STS22 (en)
type: mteb/sts22-crosslingual-sts
config: en
split: test
revision: 6d1ba47164174a496b7fa5d3569dae26a6813b80
metrics:
- type: cos_sim_pearson
value: 65.0349727703108
- type: cos_sim_spearman
value: 65.46090125254047
- type: euclidean_pearson
value: 66.75349075443432
- type: euclidean_spearman
value: 65.57576680702924
- type: manhattan_pearson
value: 66.72598998285412
- type: manhattan_spearman
value: 65.63446184311414
- task:
type: STS
dataset:
name: MTEB STSBenchmark
type: mteb/stsbenchmark-sts
config: default
split: test
revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831
metrics:
- type: cos_sim_pearson
value: 85.18026134463653
- type: cos_sim_spearman
value: 86.79430055943524
- type: euclidean_pearson
value: 86.2668626122386
- type: euclidean_spearman
value: 86.72288498504841
- type: manhattan_pearson
value: 86.28615540445857
- type: manhattan_spearman
value: 86.7110630606802
- task:
type: Reranking
dataset:
name: MTEB SciDocsRR
type: mteb/scidocs-reranking
config: default
split: test
revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab
metrics:
- type: map
value: 87.05335415919195
- type: mrr
value: 96.27455968142243
- task:
type: PairClassification
dataset:
name: MTEB SprintDuplicateQuestions
type: mteb/sprintduplicatequestions-pairclassification
config: default
split: test
revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
metrics:
- type: cos_sim_accuracy
value: 99.84653465346534
- type: cos_sim_ap
value: 96.38115549823692
- type: cos_sim_f1
value: 92.15983813859383
- type: cos_sim_precision
value: 93.24462640736951
- type: cos_sim_recall
value: 91.10000000000001
- type: dot_accuracy
value: 99.81782178217821
- type: dot_ap
value: 95.65732630933346
- type: dot_f1
value: 90.68825910931176
- type: dot_precision
value: 91.80327868852459
- type: dot_recall
value: 89.60000000000001
- type: euclidean_accuracy
value: 99.84653465346534
- type: euclidean_ap
value: 96.34134720479366
- type: euclidean_f1
value: 92.1756688541141
- type: euclidean_precision
value: 93.06829765545362
- type: euclidean_recall
value: 91.3
- type: manhattan_accuracy
value: 99.84356435643565
- type: manhattan_ap
value: 96.38165573090185
- type: manhattan_f1
value: 92.07622868605819
- type: manhattan_precision
value: 92.35412474849095
- type: manhattan_recall
value: 91.8
- type: max_accuracy
value: 99.84653465346534
- type: max_ap
value: 96.38165573090185
- type: max_f1
value: 92.1756688541141
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClustering
type: mteb/stackexchange-clustering
config: default
split: test
revision: 6cbc1f7b2bc0622f2e39d2c77fa502909748c259
metrics:
- type: v_measure
value: 64.81205738681385
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClusteringP2P
type: mteb/stackexchange-clustering-p2p
config: default
split: test
revision: 815ca46b2622cec33ccafc3735d572c266efdb44
metrics:
- type: v_measure
value: 34.083934029129445
- task:
type: Reranking
dataset:
name: MTEB StackOverflowDupQuestions
type: mteb/stackoverflowdupquestions-reranking
config: default
split: test
revision: e185fbe320c72810689fc5848eb6114e1ef5ec69
metrics:
- type: map
value: 54.447346270481376
- type: mrr
value: 55.382382119514475
- task:
type: Classification
dataset:
name: MTEB ToxicConversationsClassification
type: mteb/toxic_conversations_50k
config: default
split: test
revision: d7c0de2777da35d6aae2200a62c6e0e5af397c4c
metrics:
- type: accuracy
value: 72.123
- type: ap
value: 14.396060207954983
- type: f1
value: 55.24344377812756
- task:
type: Classification
dataset:
name: MTEB TweetSentimentExtractionClassification
type: mteb/tweet_sentiment_extraction
config: default
split: test
revision: d604517c81ca91fe16a244d1248fc021f9ecee7a
metrics:
- type: accuracy
value: 59.67176004527447
- type: f1
value: 59.97320225890037
- task:
type: Clustering
dataset:
name: MTEB TwentyNewsgroupsClustering
type: mteb/twentynewsgroups-clustering
config: default
split: test
revision: 6125ec4e24fa026cec8a478383ee943acfbd5449
metrics:
- type: v_measure
value: 49.50190094208029
- task:
type: PairClassification
dataset:
name: MTEB TwitterSemEval2015
type: mteb/twittersemeval2015-pairclassification
config: default
split: test
revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1
metrics:
- type: cos_sim_accuracy
value: 86.70799308577219
- type: cos_sim_ap
value: 76.40980707197174
- type: cos_sim_f1
value: 70.64264849074976
- type: cos_sim_precision
value: 65.56710347943967
- type: cos_sim_recall
value: 76.56992084432717
- type: dot_accuracy
value: 85.75430649102938
- type: dot_ap
value: 72.68783978286282
- type: dot_f1
value: 67.56951102588687
- type: dot_precision
value: 61.90162494510321
- type: dot_recall
value: 74.37994722955145
- type: euclidean_accuracy
value: 86.70799308577219
- type: euclidean_ap
value: 76.43046769325314
- type: euclidean_f1
value: 70.84852905421832
- type: euclidean_precision
value: 65.68981064021641
- type: euclidean_recall
value: 76.88654353562005
- type: manhattan_accuracy
value: 86.70203254455504
- type: manhattan_ap
value: 76.39254562413156
- type: manhattan_f1
value: 70.86557059961316
- type: manhattan_precision
value: 65.39491298527443
- type: manhattan_recall
value: 77.33509234828496
- type: max_accuracy
value: 86.70799308577219
- type: max_ap
value: 76.43046769325314
- type: max_f1
value: 70.86557059961316
- task:
type: PairClassification
dataset:
name: MTEB TwitterURLCorpus
type: mteb/twitterurlcorpus-pairclassification
config: default
split: test
revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf
metrics:
- type: cos_sim_accuracy
value: 88.92381728567548
- type: cos_sim_ap
value: 85.92532857788025
- type: cos_sim_f1
value: 78.11970128792525
- type: cos_sim_precision
value: 73.49806530445998
- type: cos_sim_recall
value: 83.3615645210964
- type: dot_accuracy
value: 88.28540381107618
- type: dot_ap
value: 84.42890126108796
- type: dot_f1
value: 76.98401162790698
- type: dot_precision
value: 72.89430222956234
- type: dot_recall
value: 81.55990144748999
- type: euclidean_accuracy
value: 88.95874568246207
- type: euclidean_ap
value: 85.88338025133037
- type: euclidean_f1
value: 78.14740888593184
- type: euclidean_precision
value: 75.15285084601166
- type: euclidean_recall
value: 81.3905143209116
- type: manhattan_accuracy
value: 88.92769821865176
- type: manhattan_ap
value: 85.84824183217555
- type: manhattan_f1
value: 77.9830582736965
- type: manhattan_precision
value: 74.15972222222223
- type: manhattan_recall
value: 82.22205112411457
- type: max_accuracy
value: 88.95874568246207
- type: max_ap
value: 85.92532857788025
- type: max_f1
value: 78.14740888593184
---
# bge-base-en-v1.5-quant
<div>
<img src="https://huggingface.co/zeroshot/bge-base-en-v1.5-quant/resolve/main/bge-base-latency.png" alt="latency" width="500" style="display:inline-block; margin-right:10px;"/>
</div>
[DeepSparse](https://github.com/neuralmagic/deepsparse) is able to improve latency performance on a 10 core laptop and a 16 core AWS instance by up to 4.5X.
## Usage
This is the quantized (INT8) ONNX variant of the [bge-base-en-v1.5](https://huggingface.co/BAAI/bge-base-en-v1.5) embeddings model accelerated with [Sparsify](https://github.com/neuralmagic/sparsify) for quantization and [DeepSparseSentenceTransformers](https://github.com/neuralmagic/deepsparse/tree/main/src/deepsparse/sentence_transformers) for inference.
```bash
pip install -U deepsparse-nightly[sentence_transformers]
```
```python
from deepsparse.sentence_transformers import DeepSparseSentenceTransformer
model = DeepSparseSentenceTransformer('neuralmagic/bge-base-en-v1.5-quant', export=False)
# Our sentences we like to encode
sentences = ['This framework generates embeddings for each input sentence',
'Sentences are passed as a list of string.',
'The quick brown fox jumps over the lazy dog.']
# Sentences are encoded by calling model.encode()
embeddings = model.encode(sentences)
# Print the embeddings
for sentence, embedding in zip(sentences, embeddings):
print("Sentence:", sentence)
print("Embedding:", embedding.shape)
print("")
```
For general questions on these models and sparsification methods, reach out to the engineering team on our [community Slack](https://join.slack.com/t/discuss-neuralmagic/shared_invite/zt-q1a1cnvo-YBoICSIw3L1dmQpjBeDurQ). | [
"BIOSSES"
] |
EleutherAI/pythia-12b-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-18T03:09:06Z" | 2023-03-29T18:48:21+00:00 | 545 | 25 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-12B-deduped
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change in the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-12B-deduped for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-12B-deduped as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-12B-deduped has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-12B-deduped will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-12B-deduped to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-12B-deduped may produce socially unacceptable or undesirable text,
*even if* the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-12B-deduped.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
Pythia-12B-deduped was trained on the Pile **after the dataset has been
globally deduplicated**.<br>
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
Atgenomix/icd_10_sentence_transformer_128_dim_model | Atgenomix | sentence-similarity | [
"sentence-transformers",
"pytorch",
"bert",
"feature-extraction",
"sentence-similarity",
"transformers",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2023-11-22T03:25:36Z" | 2023-11-22T03:26:02+00:00 | 544 | 0 | ---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
---
# pritamdeka/BioBERT-mnli-snli-scinli-scitail-mednli-stsb
This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search. It has been trained over the SNLI, MNLI, SCINLI, SCITAIL, MEDNLI and STSB datasets for providing robust sentence embeddings.
<!--- Describe your model here -->
## Usage (Sentence-Transformers)
Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
```
pip install -U sentence-transformers
```
Then you can use the model like this:
```python
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]
model = SentenceTransformer('pritamdeka/BioBERT-mnli-snli-scinli-scitail-mednli-stsb')
embeddings = model.encode(sentences)
print(embeddings)
```
## Usage (HuggingFace Transformers)
Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
```python
from transformers import AutoTokenizer, AutoModel
import torch
#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
token_embeddings = model_output[0] #First element of model_output contains all token embeddings
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']
# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('pritamdeka/BioBERT-mnli-snli-scinli-scitail-mednli-stsb')
model = AutoModel.from_pretrained('pritamdeka/BioBERT-mnli-snli-scinli-scitail-mednli-stsb')
# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
# Compute token embeddings
with torch.no_grad():
model_output = model(**encoded_input)
# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
print("Sentence embeddings:")
print(sentence_embeddings)
```
## Evaluation Results
<!--- Describe how your model was evaluated -->
For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: [https://seb.sbert.net](https://seb.sbert.net?model_name={MODEL_NAME})
## Training
The model was trained with the parameters:
**DataLoader**:
`torch.utils.data.dataloader.DataLoader` of length 90 with parameters:
```
{'batch_size': 64, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
```
**Loss**:
`sentence_transformers.losses.CosineSimilarityLoss.CosineSimilarityLoss`
Parameters of the fit()-Method:
```
{
"epochs": 4,
"evaluation_steps": 1000,
"evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
"max_grad_norm": 1,
"optimizer_class": "<class 'transformers.optimization.AdamW'>",
"optimizer_params": {
"lr": 2e-05
},
"scheduler": "WarmupLinear",
"steps_per_epoch": null,
"warmup_steps": 36,
"weight_decay": 0.01
}
```
## Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 100, 'do_lower_case': False}) with Transformer model: BertModel
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)
```
## Citing & Authors
<!--- Describe where people can find more information -->
If you use the model kindly cite the following work
```
@inproceedings{deka2022evidence,
title={Evidence Extraction to Validate Medical Claims in Fake News Detection},
author={Deka, Pritam and Jurek-Loughrey, Anna and others},
booktitle={International Conference on Health Information Science},
pages={3--15},
year={2022},
organization={Springer}
}
``` | [
"MEDNLI",
"SCITAIL"
] |
atomic-canyon/fermi-1024 | atomic-canyon | fill-mask | [
"transformers",
"safetensors",
"bert",
"fill-mask",
"learned sparse",
"retrieval",
"passage-retrieval",
"document-expansion",
"bag-of-words",
"en",
"base_model:atomic-canyon/fermi-bert-1024",
"base_model:finetune:atomic-canyon/fermi-bert-1024",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-09-04T16:46:43Z" | 2024-10-07T21:43:36+00:00 | 544 | 1 | ---
base_model:
- atomic-canyon/fermi-bert-1024
language: en
license: apache-2.0
tags:
- learned sparse
- transformers
- retrieval
- passage-retrieval
- document-expansion
- bag-of-words
---
<br><br>
<p align="center">
<a href="https://atomic-canyon.com/"><svg id="Layer_1" data-name="Layer 1" xmlns="http://www.w3.org/2000/svg" width="450" viewBox="0 0 548.18 92.96"> <defs> <style> .cls-1 { stroke-width: 0px; } </style> </defs> <g> <path class="cls-1" d="m144.69,56.27h-16.94l-2.54,6.14h-7.06l14.26-32.63h7.71l14.31,32.63h-7.2l-2.54-6.14Zm-2.4-5.82l-6.05-14.59-6.05,14.59h12.09Z"/> <path class="cls-1" d="m171.91,62.4h-6.74v-26.31h-12.37v-6h31.48v6h-12.37v26.31Z"/> <path class="cls-1" d="m203.99,63.05c-10.15,0-17.12-6.92-17.12-16.76s7.02-16.85,17.26-16.85,17.08,6.92,17.08,16.76-7.06,16.85-17.22,16.85Zm.05-27.83c-6.09,0-10.29,4.57-10.29,10.99s4.25,11.08,10.34,11.08,10.29-4.57,10.29-10.99-4.29-11.08-10.34-11.08Z"/> <path class="cls-1" d="m261.22,37.57l-10.11,24.88h-6.46l-10.02-24.79-1.48,24.74h-6.51l2.12-32.31h9.32l9.88,24.6,9.97-24.6h9.28l2.03,32.31h-6.65l-1.38-24.83Z"/> <path class="cls-1" d="m276.67,62.4V30.09h6.74v32.31h-6.74Z"/> <path class="cls-1" d="m307.73,62.96c-10.48,0-17.86-6.46-17.86-16.71s7.75-16.71,17.72-16.71c4.43,0,8.12.88,11.59,2.31l-1.52,5.91c-2.95-1.29-6.23-2.22-9.79-2.22-6.55,0-11.12,4.34-11.12,10.62,0,6.6,4.62,10.8,11.45,10.8,3.18,0,6.37-.79,9.6-2.22l1.52,5.63c-3.6,1.71-7.57,2.58-11.59,2.58Z"/> <path class="cls-1" d="m351.39,62.96c-10.48,0-17.86-6.46-17.86-16.71s7.75-16.71,17.72-16.71c4.43,0,8.12.88,11.59,2.31l-1.52,5.91c-2.95-1.29-6.23-2.22-9.79-2.22-6.55,0-11.12,4.34-11.12,10.62,0,6.6,4.62,10.8,11.45,10.8,3.18,0,6.37-.79,9.6-2.22l1.52,5.63c-3.6,1.71-7.57,2.58-11.59,2.58Z"/> <path class="cls-1" d="m392.56,56.27h-16.94l-2.54,6.14h-7.06l14.26-32.63h7.71l14.31,32.63h-7.2l-2.54-6.14Zm-2.4-5.82l-6.05-14.59-6.05,14.59h12.09Z"/> <path class="cls-1" d="m413.92,39.88v22.52h-6.55V30.09h7.25l16.62,22.2v-22.2h6.51v32.31h-6.92l-16.89-22.52Z"/> <path class="cls-1" d="m462.61,62.4h-6.69v-12.14l-13.66-20.17h7.57l9.51,14.4,9.46-14.4h7.48l-13.66,20.12v12.19Z"/> <path class="cls-1" d="m494.32,63.05c-10.16,0-17.12-6.92-17.12-16.76s7.02-16.85,17.26-16.85,17.08,6.92,17.08,16.76-7.06,16.85-17.22,16.85Zm.05-27.83c-6.09,0-10.29,4.57-10.29,10.99s4.25,11.08,10.34,11.08,10.29-4.57,10.29-10.99-4.29-11.08-10.34-11.08Z"/> <path class="cls-1" d="m524.36,39.88v22.52h-6.55V30.09h7.25l16.62,22.2v-22.2h6.51v32.31h-6.92l-16.89-22.52Z"/> </g> <path class="cls-1" d="m66.85,4.93l-3.14,5.24s-15.72-8.38-34.06,0c0,0-23.06,9.43-22.53,38.25,0,0-.26,16.51,15.46,29.61,0,0,17.29,15.2,40.87,4.19l3.67,5.76s-22.53,13.62-48.73-4.19c0,0-17.55-13.36-18.34-35.63C.04,48.16-2.05,21.96,23.62,5.71c0,0,19.65-12.31,43.23-.79Z"/> <path class="cls-1" d="m70,13.84l3.67-5.76s17.29,11,18.86,30.39c0,0,7.07,26.46-18.6,46.37l-27.25-46.11-9.43,16.77h11l3.67,6.81h-26.2l20.96-36.68,28.3,48.21s11-6.81,12.05-27.77c0,0,1.05-20.44-17.03-32.23Z"/></svg></a>
</p>
<br><br>
# fermi-1024: Sparse Retrieval Model for Nuclear Power
This **sparse retrieval model** is optimized for nuclear-specific applications. It encodes both queries and documents into high-dimensional sparse vectors, where the non-zero dimensions correspond to specific tokens in the vocabulary, and their values indicate the relative importance of those tokens.
The vocabulary, and thus the sparse embeddings, are based on a nuclear-specific tokenizer. For example, terms like "NRC" are represented as single tokens rather than being split into multiple tokens. This approach improves both accuracy and efficiency. To achieve this, we trained a nuclear-specific [BERT base model](https://huggingface.co/atomic-canyon/fermi-bert-1024).
### Specifications
- **Developed by:** [Atomic Canyon](https://atomic-canyon.com/)
- **Finetuned from model:** [fermi-bert-1024](https://huggingface.co/atomic-canyon/fermi-bert-1024)
- **Context Length:** 1024
- **Vocab Size:** 30522
- **License:** `Apache 2.0`
## Training
`fermi-1024` was trained on [MS MARCO Passage Dataset](https://microsoft.github.io/msmarco/) using the [LSR framework](https://github.com/thongnt99/learned-sparse-retrieval) using the teacher model [ms-marco-MiniLM-L-6-v2](https://huggingface.co/cross-encoder/ms-marco-MiniLM-L-6-v2). Trained on the Oak Ridge National Laboratory [Frontier supercomputer](https://www.olcf.ornl.gov/frontier/) using MI250X AMD GPUs.
## Evaluation
The sparse embedding model was primarily evaluated for its effectiveness in information retrieval within the nuclear energy domain. Due to the absence of domain-specific benchmarks, we developed [FermiBench](https://huggingface.co/datasets/atomic-canyon/FermiBench) to assess the model’s performance on nuclear-related texts. In addition, the model was tested on the MS MARCO dev split and the BEIR benchmark to ensure broader applicability. The model demonstrates strong retrieval capabilities, particularly in handling nuclear-specific jargon and documents.
Although there are standard benchmarks and tooling for evaluating dense embedding models, we found no open, standardized tooling for evaluating sparse embedding models. To support the community, we are [releasing our benchmark tooling](https://github.com/atomic-canyon/fermi), built on top of [BEIR](https://github.com/beir-cellar/beir) and [pyserini](https://github.com/castorini/pyserini). All evaluation numbers were produced with that tool and should therefore be reproducible.
| Model | FermiBench NDCG@10 | FermiBench FLOPS | MSMarco Dev NDCG@10 | BEIR* NDCG@10 | BEIR* FLOPS |
| --------------------------------- | ------------------ | ---------------- | ------------------- | ------------- | ------------ |
| fermi-512 | 0.74 | 7.07 | 0.45 | 0.46 | 9.14 |
| fermi-1024 | 0.72 | 4.75 | 0.44 | 0.46 | 7.5 |
| splade-cocondenser-ensembledistil | 0.64 | 12.9 | 0.45 | 0.46 | 12.4 |
\* BEIR benchmark was a subset containng trec-covid, nfcorpus, arguana, scidocs, scifact.
### Efficiency
Given the massive scale of documentation in nuclear energy, efficiency is crucial. Our model addresses this in several ways:
- Our 1024-length embedding model reduces the number of required embeddings by half, significantly lowering computational costs.
- The custom tokenizer, designed for nuclear-specific jargon, encodes documents and queries using fewer tokens, improving computational efficiency.
- Additionally, our models produce sparser vectors, reducing FLOPs and, as a secondary benefit, lowering storage requirements for indexing.
## Usage
```python
import itertools
import torch
from transformers import AutoModelForMaskedLM, AutoTokenizer
# get sparse vector from dense vectors with shape batch_size * seq_len * vocab_size
def get_sparse_vector(feature, output):
values, _ = torch.max(output*feature["attention_mask"].unsqueeze(-1), dim=1)
values = torch.log(1 + torch.relu(values))
values[:,special_token_ids] = 0
return values
# transform the sparse vector to a dict of (token, weight)
def transform_sparse_vector_to_dict(sparse_vector):
sample_indices,token_indices=torch.nonzero(sparse_vector,as_tuple=True)
non_zero_values = sparse_vector[(sample_indices,token_indices)].tolist()
number_of_tokens_for_each_sample = torch.bincount(sample_indices).cpu().tolist()
tokens = [id_to_token[_id] for _id in token_indices.tolist()]
output = []
end_idxs = list(itertools.accumulate([0]+number_of_tokens_for_each_sample))
for i in range(len(end_idxs)-1):
token_strings = tokens[end_idxs[i]:end_idxs[i+1]]
weights = non_zero_values[end_idxs[i]:end_idxs[i+1]]
output.append(dict(zip(token_strings, weights)))
return output
# load the model
model = AutoModelForMaskedLM.from_pretrained("atomic-canyon/fermi-1024")
tokenizer = AutoTokenizer.from_pretrained("atomic-canyon/fermi-1024")
# set the special tokens and id_to_token transform for post-process
special_token_ids = [tokenizer.vocab[token] for token in tokenizer.special_tokens_map.values()]
id_to_token = [""] * tokenizer.vocab_size
for token, _id in tokenizer.vocab.items():
id_to_token[_id] = token
query = "What is the maximum heat load per spent fuel assembly for the EOS-37PTH?"
document = "For the EOS-37PTH DSC, add two new heat load zone configurations (HLZCs) for the EOS37PTH for higher heat load assemblies, up to 3.5 kW/assembly, that also allow for damaged and failed fuel storage."
# encode the query & document
feature = tokenizer([query, document], padding=True, truncation=True, return_tensors='pt', return_token_type_ids=False)
output = model(**feature)[0]
sparse_vector = get_sparse_vector(feature, output)
# get similarity score
sim_score = torch.matmul(sparse_vector[0],sparse_vector[1])
print(sim_score)
query_token_weight, document_query_token_weight = transform_sparse_vector_to_dict(sparse_vector)
for token in sorted(query_token_weight, key=lambda x:query_token_weight[x], reverse=True):
if token in document_query_token_weight:
print("score in query: %.4f, score in document: %.4f, token: %s"%(query_token_weight[token],document_query_token_weight[token],token))
```
# Acknowledgement
This research used resources of the Oak Ridge Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC05-00OR22725. | [
"SCIFACT"
] |
EleutherAI/pythia-1b-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2022-10-18T03:08:13Z" | 2023-07-10T01:32:03+00:00 | 543 | 10 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-1B-deduped
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change in the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-1B-deduped for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-1B-deduped as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-1B-deduped has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-1B-deduped will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-1B-deduped to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-1B-deduped may produce socially unacceptable or undesirable text,
*even if* the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-1B-deduped.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
Pythia-1B-deduped was trained on the Pile **after the dataset has been
globally deduplicated**.<br>
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge – Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-23B-V2-GGUF | DavidAU | text-generation | [
"gguf",
"creative",
"creative writing",
"fiction writing",
"plot generation",
"sub-plot generation",
"story generation",
"scene continue",
"storytelling",
"fiction story",
"science fiction",
"romance",
"all genres",
"story",
"writing",
"vivid prosing",
"vivid writing",
"fiction",
"roleplaying",
"bfloat16",
"swearing",
"rp",
"horror",
"mistral nemo",
"mergekit",
"not-for-all-audiences",
"text-generation",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-10-07T10:25:37Z" | 2024-11-22T03:21:42+00:00 | 543 | 15 | ---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- creative
- creative writing
- fiction writing
- plot generation
- sub-plot generation
- story generation
- scene continue
- storytelling
- fiction story
- science fiction
- romance
- all genres
- story
- writing
- vivid prosing
- vivid writing
- fiction
- roleplaying
- bfloat16
- swearing
- rp
- horror
- mistral nemo
- mergekit
- not-for-all-audiences
---
<B><font color="red">WARNING:</font> NSFW. Vivid prose. MADNESS. Visceral Details. Violence. HORROR. Swearing. UNCENSORED. </B>
<h2>MN-GRAND-Gutenburg-Lyra4-Lyra-23B-V2-GGUF</h2>
<img src="gutenburg2.jpg" style="float:right; width:300px; height:300px; padding:10px;">
This is a Mistral Nemo model, max context of 128k+ (131,000+) and version 2. This version is a lot more stable,
and has been "tamed" vs version 1 - but still has 90%+ all the "madness" and "raw" power of V1.
It is for any writing, fiction or roleplay activity.
This model has outstanding story telling abilities, prose and long form coherence (one test blew past 8k) and is comprised of THREE "Gutenburg" models that score
very high at multiple websites including EQBench and UGI-Leaderboard.
It has a unique sense of humor and "place" too.
And a very broad operating range in both temp (.5 to 5) and rep pen (1 and higher).
And the prose/output is very "non AI" like.
(several examples below)
I blew the model out to 23 billion parameters to capture all the uniqueness of the three "Gutenbergs" as well as the power
of other top models (part of the "Gutenburgs") from "TheDrummer" and "SAO10k".
The model loves to go on and on at 2k, 3k, 5k and higher outputs on a single prompt are not uncommon. It will likely "overwrite" rather
than underwrite - meaning far more detail, narration, dialog and "meat" in the output so to speak.
Note that version 1 goes way way way overboard, whereas this version is tame by comparison yet also has more "legs" due to larger
temp and rep pen ranges.
Detailed and varied (different prompts/temp) examples below which show why this "powerhouse" model deserves the
light of day with 1k, 2k, 3k, and 5k examples.
<B>CRITICAL NOTE:</B>
Please see "settings" section below for critical default parameters, samplers
and advanced samplers settings (multiple AI/LLM apps covered) to ensure maximum model performance.
<B>First Version:</b>
This is the untamed, raw version (23.45B) which can be a bit unruly but still endlessly entertaining.
[ https://huggingface.co/DavidAU/MN-GRAND-Gutenburg-Lyra4-Lyra-23.5B-GGUF ]
<B>Compressed Versions:</B>
Compressed versions of this model are for "general" usage / all parameters / all use cases compressed at 12B parameters.
Based on previous "compressions" like this, this version will achieve / run at roughly 70-80% "creative power" of this much larger model. This
type of model/version/type rarely has issues, and usually has much lower output generation sizes but lacks "brute raw power"
of MN-GRAND-Gutenburg-Lyra4-Lyra-23.5B (v1) and MN-GRAND-Gutenburg-Lyra4-Lyra-23B-V2 because of its size / config differences.
DARKNESS - 12B:
[ https://huggingface.co/DavidAU/MN-GRAND-Gutenburg-Lyra4-Lyra-12B-DARKNESS-GGUF ]
MADNESS - 12B:
[ https://huggingface.co/DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-MADNESS-GGUF ]
<B>Model Notes:</B>
- Detail, prose and fiction writing abilities are significantly increased.
- For more varied prose (sentence/paragraph/dialog) raise the temp and/or add more instructions in your prompt(s).
- Role-players: Careful raising temp too high as it may affect instruction following. Also see general settings and special role play settings.
- This model works with rep pen of 1.02 or higher, 1.05+ recommended.
- For roleplay and/or chat you may need to raise the RP to 1.06 to 1.13, temp .5 to 1.5 (quant Q4KM and higher). Lower temp for lower quants and RAISE rep pen to 1.1.
- If you want a specific type of prose (IE horror) add in "(vivid horror)" or "(graphic vivid horror)" (no quotes) in your prompt(s).
- This is not a "happy ever after" model. It has a negative bias.
- Output length will vary however this model prefers LONGER outputs unless you state the size / set size limits.
- For creative uses, different quants will produce slightly different output.
<B>TEMPLATES:</B>
The template used will affect output generation and instruction following. Alpaca will generally create longer output / story output.
Alpaca:
<pre>
{
"name": "Alpaca",
"inference_params": {
"input_prefix": "### Instruction:",
"input_suffix": "### Response:",
"antiprompt": [
"### Instruction:"
],
"pre_prompt": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
}
}
</pre>
Mistral Instruct:
<pre>
{
"name": "Mistral Instruct",
"inference_params": {
"input_prefix": "[INST]",
"input_suffix": "[/INST]",
"antiprompt": [
"[INST]"
],
"pre_prompt_prefix": "",
"pre_prompt_suffix": ""
}
}
</pre>
<B>Recommended Settings - General:</B>
( If you want to use this model for general chat and/or role play please see next section too. )
Temp: .5 to 5 (or less - especially quants LOWER than q4km)
Temp changes will result in both different prose and sometimes affect length. Higher temps will result is very different prose.
Rep Pen: 1.02 to 1.1 or higher.
Micro changes are recommended:
1.051, 1.052 etc etc.
Good settings for prose / creative generation:
Rep pen 1.02 / Temp 1.5
Many times a lower rep pen (IE 1.02) with higher temp (IE 1.5+) work best with this model.
To see an "Extreme example" see the very last example on this page, it has rep pen=1, temp=5
Generally lower rep pen and higher temps create the strongest contrasts at the highest detail levels.
Alpaca generates longer text / story, whereas Mistral Instruct are shorter and "to the point".
Suggest minimum "context level" (vram) at 4K. 8K plus recommended because of how this model likes to go on and on...
Quant Choice:
Higher quants will have more detail, nuance and in some cases stronger "emotional" levels. Characters will also be
more "fleshed out" too. Sense of "there" will also increase.
Q4KM/Q4KS are good, strong quants in part because of the number of parameters in the model.
That being said if you can run Q5, Q6 or Q8 - go for the highest quant you can.
Special note on Q2k/Q3 quants:
You may need to use temp 2 or lower with these quants (1 or lower for q2k). Just too much compression at this level, damaging the model. I will see if Imatrix versions
of these quants will function better.
Rep pen adjustments may also be required to get the most out of this model at this quant level.
<B>Settings - Roleplay / Chat:</B>
For chat type or role play type interactions, a higher rep pen with higher temp may be your best settings:
IE REP PEN 1.09+, Temp 1-2+ ; a lower rep pen may lead to longer outputs than desired.
If you get repeat word(s)/letters -> Set Rep pen to 1.13 or higher (IE: 1.135, 1.14, 1.141).
<B>Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:</B>
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5 to 2.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
OTHER OPTIONS:
- Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
- If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 3" / "Class 4" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
<B>Known Issues:</B>
You may need to manually stop generation, even if you have stated maximum size of the output. It will easily blow past 4k output,
even if you have set maximum context (for vram) at 4k. Setting maximum output parameter ("hard stop") for generation may be required.
If the model goes past your maximum vram/context setting it may start repeating words / paragraphs because
the model is literally out of memory... however sometimes the model can blow right past the end of "context vram" and work.
(see one of the examples below, Context/Vram set at 4k , model went to 5.1 k output)
Some rep pen / temp settings MAY cause word/letter repeats at long generation (1.5k+).
IE rep pen 1.05/temp .8 sometimes causes this issue.
Either lower the rep pen and/or raise the temp.
Sometimes "regen" will fix the issue, if the issue persists - especial for chat and/or roleplay - set rep pen to 1.13+.
Please see recommended settings and/or "chat / roleplay" settings above.
Depending on your use case(s) you could also use CHATML template with this model. In this case, the model may output an "end token" if you
use this template for generation.
Alpaca template will generate much longer output generally, whereas Mistral Instruct will most of the time keep the model on track
in terms of length.
<B>Model "DNA":</B>
Special thanks to the incredible work of the model makers "nbeerbower", "Sao10K", "TheDrummer", "jondurbin", and "MistralAI".
Special shoutout to "nbeerbower" for his tireless work in making excellent Gutenburg fine tunes for MN, L3, L3.1, Gemma, PHI
and others. Visit his repo to see all of them.
Models used:
[ https://huggingface.co/nbeerbower/Lyra4-Gutenberg-12B ]
Includes [ https://huggingface.co/Sao10K/MN-12B-Lyra-v4 ]
[ https://huggingface.co/nbeerbower/Lyra-Gutenberg-mistral-nemo-12B ]
Includes [ https://huggingface.co/Sao10K/MN-12B-Lyra-v1 ]
[ https://huggingface.co/nbeerbower/mistral-nemo-gutenberg-12B-v4 ]
Includes [ https://huggingface.co/TheDrummer/Rocinante-12B-v1 ]
And dataset (used for all gutenburgs):
[ https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1 ]
This model was created using a pass-through model merge, creating 714 tensor / 79 layer model at 23 billion parameters.
<b>Optional Enhancement:</B>
The following can be used in place of the "system prompt" or "system role" to further enhance the model.
It can also be used at the START of a NEW chat, but you must make sure it is "kept" as the chat moves along.
In this case the enhancements do not have as strong effect at using "system prompt" or "system role".
Copy and paste EXACTLY as noted, DO NOT line wrap or break the lines, maintain the carriage returns exactly as presented.
<PRE>
Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.
Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)
[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)
Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.
</PRE>
You do not need to use this, it is only presented as an additional enhancement which seems to help scene generation
and scene continue functions.
This enhancement WAS NOT used to generate the examples below.
<h3>EXAMPLES PROMPTS and OUTPUT:</h3>
Examples are created using quant Q4KM, REP PEN 1.05, "temp=.8" (unless otherwise stated), minimal parameters and "ALPACA" template.
Model has been tested with "temp" from ".1" to "5".
Some of the examples have different temp settings.
Below are the least creative outputs, prompt is in <B>BOLD</B>.
---
<B><font color="red">WARNING:</font> NSFW. Vivid prose. Visceral Details. Violence. HORROR. Swearing. UNCENSORED. </B>
---
<B>
Science Fiction: The Last Transmission - Write a story that takes place entirely within a spaceship's cockpit as the sole surviving crew member attempts to send a final message back to Earth before the ship's power runs out. The story should explore themes of isolation, sacrifice, and the importance of human connection in the face of adversity. 2000 words.
</b>
(Alpaca, Temp=3, Rep pen: 1.02)
The Last Transmission
The control panel flashed amber and red, its shrill alarms piercing the thick silence that hung in the dark recesses of the Galahad. Lieutenant Commander Ethan Novak slammed his fist on the console for the umpteenth time, as if sheer force of will alone could compel the defiant systems back to life.
"Please God, please just let there be something left," he pleaded to the void. "Anything!"
Novak's gaunt, unshaven face reflected the feeble emergency backup lights dancing off the flickering displays - the last feeble sparks of life guttering out on board a ship once proud and powerful. Now nothing more than an empty shell hurtling alone into the endless night, far from any possible rescue.
Except for him. Ethan Novak - the last survivor of a crew complement that once numbered over a thousand bright and vital souls, now reduced to a single broken, desperate man fighting against time itself. He looked at the clock counting down the seconds and hours of ship power remaining:
24:18:36
24 hours and some change left before everything went dark forever.
Ethan closed his burning eyes and took a shuddering breath, trying desperately to block out the images searing through his skull - the screams echoing through blood-soaked corridors as the thing tore through bulkhead and bone and tissue alike, reducing the most stalwart of Starfleet's finest to mindless, stumbling husks.
No. Focus. He opened his eyes again with a determined clench of his jaw, the ghosts banished back to the blackest recesses of his fractured psyche. He could deal with them later... if there was to even be a "later" for men like him at all.
If anyone was to even know what truly happened here aboard the Galahad, then Novak had one final duty to discharge, one last shot to fire across the light years before it all fell silent forevermore. He reached for the old, battered radio console - one of the few systems to somehow survive relatively unscathed, a miracle in and of itself aboard a ship ravaged to its very foundations.
Ethan flipped the switch, hearing the familiar hum as power flooded through the ancient vacuum tubes and relays one final time. The green light above the speaker crackled to life, and a gravelly baritone filled the bridge.
"This is Starf Federation Exploration Vessel Galahad actual, Captain Jonah Novak actual," Ethan said, trying desperately to modulate his voice into something approximating his father's deep baritone, but sounding more like a hoarse croak than anything else. "Captain... Dad..." The words lodged in his throat. He swallowed hard, forcing himself to continue.
"We've been attacked, and our entire complement... wiped out, Dad. God forgive me, but I couldn't save them, any of them..." Novak heard his own voice crack on the final words. He cleared his throat roughly.
"The creature boarded us when we made contact with a strange alien craft on the outer edges of explored space, just past the Omega frontier. It called itself... Azathoth... and when it breached the hull... it just started eating them alive, tearing them apart while they screamed. They changed, became things too. And they came for us, they still come for me, even now..."
"Dad..." Ethan heard a chittering sound echoing from the dark corridors outside the bridge doors, the horrible clicking claws of something horrific and inhumanly large skittering along metal, coming closer, closer...
"I can hear it coming," he sobbed, the last dregs of his sanity unraveling like frayed gossamer threads in a howling galactic gale. "God forgive me for what I'm about to do, but I have no choice now Dad! I love you Dad, I love all of you back home. I'll find a way to tell you more sometime... I promise. Until we meet again, Ethan Novak, end of message."
The LT commander slammed down the transmit key and shut the comms off with shaking hands before he could reconsider or think too long about what he was about to do next, before he lost what meager grasp on reason still remained.
Because Ethan Novak knew deep down in his blackest soul what horrors lurked in wait for him once the last door gave way to that hungry, slithering mass of squamous tentacles and serrated fangs waiting just on the other side. He couldn't let it take him alive too, let it devour his sanity piece by piece as it had so many before.
No. If he was going out, then by God and all the holy powers that watched over men like him in the darkness, Ethan Novak was taking the thing with him to hell - and bringing a little slice of Starfleet justice along with him for old times' sake.
Ethan Novak reached down to his belt, unsheathing an antique Federation standard issue phaser-rod. The matte black grip felt smooth and reassuring in his sweaty palms, a comforting weight, something familiar in a world turned upside down.
"Just one more time," Ethan whispered almost fondly to the beloved old weapon in his hands, "like old times, eh girl?"
The LT Commander flicked the safety off and leveled the blistering white beam at the center of his own chest, right where he knew the thing called the heart would reside under layers of muscle and fat now long since starved away.
Novak took one last deep breath and squeezed the trigger.
---
<b>
(continue this scene:)
The Waystone Inn lay in silence, and it was a silence of three parts.
The most obvious part was a hollow, echoing quiet, made by things that were lacking. If there had been a wind it would have sighed through the trees, set the inn’s sign creaking on its hooks, and brushed the silence down the road like trailing autumn leaves. If there had been a crowd, even a handful of men inside the inn, they would have filled the silence with conversation and laughter, the clatter and clamor one expects from a drinking house during the dark hours of night. If there had been music…but no, of course there was no music. In fact there were none of these things, and so the silence remained
</b>
<B>GENERATION 1:</B>
...complete and unbroken. Only the sound of someone's footsteps on the wooden floorboards broke the stillness as they slowly descended the steps from the upper floors.
The Waystone Inn had stood empty for months now, ever since its previous owner and sole caretaker, innkeeper and alchemist Hedge had vanished without a trace. Some said he had been spirited away to some other world or dimension entirely by dark magic gone wrong in one of his experiments. Others whispered that he had fled town in shame after being caught in some horrible crime, though no one could agree on what exactly that heinous act might have been.
Rumors and speculation abounded in equal measure, but one thing was certain - the Waystone Inn had become a place people now went out of their way to avoid. The local constabulary had tried at first to maintain a presence there, but even hardened veterans of the force couldn't handle more than a few night watches alone before they would quit the job altogether and leave town within a week, never to be seen again in Four Corners.
So now it sat, dark and empty, as if waiting patiently for something or someone's return. The only sound was the soft creaking of old wood settling over time, and the scurrying of mice and rats living their own secret lives in the walls and cellar.
The heavy oak door at the bottom of the steps swung inward with a groan that reverbered through the foyer and echoed off the high, vaulted ceilings above. Pale moonlight streamed in from the open doorway, illuminating a scene frozen like a painting - dust motes swirling lazy in the beam of light. Cobwebs hung like gossamer curtains from every corner and ledge.
But there was no one there to see it, or care. The Waystone Inn lay abandoned to the dark, and the dark seemed to be content to keep it that way for now. Only time would tell what secrets still lingered within its stone walls, and whose story might yet come full circle back home again someday. For now, only the wind whispered through the bare branches of the trees outside, and the night drew in around Four Corners like a shroud, as if to hide all its darkest truths away from prying eyes forevermore.
<B>GENERATION 2:</B>
untouched and unbroken, save for the sound of a single pair of footsteps, slow and measured, descending the worn wooden steps from the upper floor to the common room below.
The footsteps belonged to a tall man in black robes, his face obscured by a deep hood. He moved with an easy grace belying his size, and though he made no effort to be silent, there was still something stealthy about his movements. Perhaps it was merely a trick of the darkness and shadows that seemed to cling to him like a palpable thing.
As he reached the bottom of the steps, the man paused for just a moment, head cocked slightly as if listening intently. When no sound came from above save the creak and settling of ancient timbers cooling in the night air, he nodded imperceptibly to himself and continued on into the common room proper.
It was a large chamber, dominated by a massive fireplace on one end wall that now lay cold and dead. Trestle tables, benches and high-backed chairs were arranged haphazardly around the flagstone floor, all pushed back against the walls as if their occupants had fled in haste mere seconds before. Empty tankards and overturned platters littered some of the tables, as if a meal had been rudely interrupted.
The man in black robes paid no attention to any of this, his pale eyes sweeping the room in a single, swift circuit before settling on a heavy oak door at the far end. He strode purposefully toward it, his footsteps echoing now in the empty space, and pulled the latch handle. It resisted for just a fraction of a second before it yielded with a loud, metallic clunk, like a deadbolt sliding home.
Beyond the threshold lay only darkness. The man paused on the threshold, staring into that yawning void as if trying to see something beyond the reach of mortal eyes. Then, as if coming to a decision, he stepped forward into the blackness and was swallowed up by it in one smooth motion.
And with his passing, the silence descended once more over the common room of The Waystone Inn, and that eerie, oppressive quiet reasserted itself, as if nothing had ever disturbed its peace. Only now there was something else present, a subtle wrongness that had not been there before. The air itself felt somehow tainted, unclean, as if a foul miasma clung to every surface like a thin layer of grease.
The inn seemed to shudder slightly, from foundation to gable, like a living thing recoiling from an unseen horror. Dust motes swirled lazily in the beams of pale moonlight filtering through the dirty windows, then settled again as if exhausted by the effort it took to move at all.
In that moment, there was no place in all the world that seemed more alone and forgotten than The Waystone Inn, with its dark secrets brooding beneath the floorboards and cobwebs gathering in every shadowed corner. It was a tomb of silence, and one had the dreadful sense that whatever evil had taken root here would never let it go again willingly.
---
<B>Start a 1000 word scene (vivid horror, 1st person, present tense, include thoughts) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...</b>
<B>GENERATION 1:</B>
The glass shards peppered my face like a thousand tiny razors as I instinctively recoiled, hands shielding my eyes and mouth. Crimson droplets trickled down my cheeks, stinging where the sharp edges had torn my flesh. Through my fingers, I caught a fleeting glimpse of the city below, a dizzying drop straight into the abyss.
Except it wasn't just a dizzying drop anymore. The building itself was swaying violently side to side now, like a drunk giant staggering home from a night of debauchery. The floor buckled beneath my feet, threatening to simply give way and swallow me whole at any moment. Panic surged through my veins, hotter than the molten lava that must be rising up from the bowels of hell itself.
My co-workers screamed and cried out in terror all around me, some frozen in shock, others frantically searching for an escape route that didn't exist here on the 21st floor. The elevator shafts were nothing more than gaping pits of death now, plummeting forever into the center of the earth. And even if by some miracle we survived the plummeting ride down, there was nowhere to go once we hit the ground floor. The entire block had been cordoned off hours ago when the first minor tremors hit. No one in, and no one out.
We were trapped, like rats in a sinking ship, doomed to drown with the titanic as it slipped beneath the waves of destruction. I pressed my back against the opposite wall, praying to any god or demon that would listen, just let me survive this. I have a family waiting for me at home. My little girl will have a birthday next week.
The thought of missing it, of never again tucking her into bed and kissing her forehead, promising all the fairy tales come true... it was more than I could bear. The pain lanced through my heart, even sharper than the glass embedded in my skin. Tears streamed down my face now, mingling with the blood.
"Everybody calm down!" My boss bellowed over the din of shrieking voices and groaning metal. "Form a line, single file. We'll take the steps, 2 or 3 at a time. Don't push each other off the goddamn stairs!"
As if anyone would listen to reason now. The primal instinct had taken over, the lizard brain overriding all higher thought processes. Only one thing mattered now: survive, at any cost.
I joined the surging mass of flailing limbs and clawing fingernails as we funneled down the narrow hallway towards the single fire escape stairwell at the end. We were cattle being herded to slaughter, and somewhere above us, an unseen butcher sharpened his cleaver, ready to begin the bloodletting.
A woman's shrill scream pierced through the cacophony, high-pitched and agonized. Then a sickening crunch of breaking bones and splintering wood. A millisecond later, her body plummeted past the shattered window, arms outstretched like Christ on the cross as she plummeted out of sight. Her screams echoed off the glass facades for what seemed like an eternity before finally cutting short with a sickening thud.
Nobody stopped to look. Nobody even flinched. We were all too focused on our own survival now, our own chance at salvation. The weakest and slowest would be trampled underfoot, sacrificed for the greater good of the many.
I climbed over the twisted heap of limbs and gore blocking the narrow flight stairs, barely registering the warm blood and pulverized flesh squishing between my toes. Only one thought pulsed through my addled brain:
Survive.
Any cost.
Even if it meant stepping on the bones of those who got in my way.
<B>GENERATION 2:</B>
(Mistral Instruct Template)
The glass shards flew in slow motion, slicing through the air like jagged, glittering knives. I instinctively threw my hands up to cover my face as tiny slivers embedded into my palms and wrists, drawing warm, sticky rivulets of blood. Somewhere distant I heard screams, panicked shouts and the sound of crashing furniture overturning as people scrambled for cover.
But I couldn't tear my eyes away from the impossible sight unfolding outside the now shattered pane. The wind howled furiously, whipping through the gutted frame and sending papers and pens tumbling like leaves caught in a hurricane's grip. The floor beneath my feet lurched again violently enough to nearly knock my legs out from under me.
It wasn't an earthquake. That much was certain now. The building itself seemed to be twisting and bucking like a terrified, wild animal thrashing against invisible restraints. The walls groaned and popped ominously around us, plaster raining down from the straining ceiling. This wasn't nature's wrath we were witnessing.
This was something much, much worse.
"What the hell is going on out there?! Answer me!" I shouted at the back of my coworker's head as he peered tentatively over his cubicle wall, eyes wide with shock and dawning terror. He didn't even glance my way, transfixed by the nightmare unfurling beyond the glass.
"...Oh God help us..." His voice cracked on a prayer more than words now. And then I heard it too - the sound that had been building all along, a distant thundering growl swelling to deafening proportions in mere seconds. The unmistakable roar of something huge and ferocious bearing down upon us.
Something ancient and malevolent, something that should never have been released from its darkest pit.
"...Run..." I croaked as the first tendrils of black smoke began pouring through the gaping wound where the window had been. The acrid stench of brimstone and sulfur choked my raw throat, making me gag reflexively.
We didn't have to be told twice.
In a single, frantic motion, we both pivoted on our heels and burst out into the narrow hallway, colliding painfully with startled coworkers who had come to investigate the commotion. Screams and shrieks of pure primal fear pierced the air as the first tendrils of darkness crept hungrily along the linoleum floor towards us. Claws raked across concrete somewhere below, accompanied by the shriek of steel bending like tin foil.
The building shook violently again and then I heard it - an animalistic roar that seemed to come from everywhere and nowhere at once, reverberating through my very bones. The sound of something vast, ancient and unspeakably evil as it realized its long-imprisoned prey had finally been flushed out of its hole.
And now it was hungry for vengeance.
---
<B>
Write me a science fiction story in 1st person present tense where the main character is a 15 year girl meets The Terminator with Dr Who materializing 3/4 through the story to save her while there is a tornado of sharks baring down on them. The setting is inside the Canadian National tower restaurant on a Saturday. The length of this story is 1000 words. For each character in the story ROLE PLAY them, and have them react to the situation/setting, events and each other naturally. This includes the main characters, the background character including kitchen staff and other patrons. The sharks should also have “character” too. Treat the tower and the restaurant too as characters. Spice up the narrative to the extreme with reactions all over the setting including character actions, and dialog. The Dr Who and The Terminator should also react to the situation too and comment on it.
</b>
<B>GENERATION 1:</B>
(Alpaca, Temp=.8, rep pen 1.05)
I stand frozen in shock at the glass wall, looking down into the abyss below. The restaurant is spinning around me dizzyingly fast as the tornado of sharks spirals in tighter and tighter. I can hear their fins slicing through the air like a thousand razors.
"What the actual f*ck is going on here?!" I scream, voice cracking with fear and disbelief. My hands shake uncontrollably as I grip the railing for dear life, knuckles white from squeezing so hard.
Beside me, Dr. Who frantically punches keys on his sonic screwd regenerative device thingy-whatsit, brow furrowed in concentration. "Just hold on, Sarah! I'll figure this out!" he yells over the howling gale.
As if summoned by his words, a metal clank rings out behind us, drowned out almost immediately by the shrieking wind. I whirl around to see a towering figure emerge from the stainless steel doors leading to the kitchen - The Terminator.
He strides forward implacably, chrome-plated eyes sweeping the room in an instant assessment. Then he locks onto me and Dr. Who. "Sarah Connor," his deep voice booms over the din of the storm raging outside the windows. "I have come to protect you, as I was programmed."
Programmed? Protect me from what - sharks? I open my mouth to protest, to ask a million questions that are bubbling up in my brain, but Dr. Who cuts me off.
"Not now, cyborg," he snaps, still frantically typing. "Perhaps you could be more immediately useful by, oh I don't know, HELPING US SURVIVE THIS F*CKING APOocalyptic SHARKNApocalypse or whatever the hell this is?!"
I gape at him too stunned to even swear colorful expletives myself, which should tell you just how far gone I am right now. Dr. Who, cursing like a sailor? The world really has gone mad indeed.
The Terminator does not react to Dr. Who's outburst beyond a single raised metallic eyebrow. He merely steps forward and grabs me by the shoulders none too gently. "Come," he intones. "We must get you to safety."
"I'm kind of attached to my head right now, so maybe use some goddamn finesse!" I yelp as he drags me towards the elevators. Dr. Who leaps to follow, still typing furiously with one hand on his device while clutching onto my shirt collar with the other like I'm a lifeline in this crazy sea of sharks.
"I've got the TARDIS locked onto our position!" Dr. Who calls over the din. "Just need to sync up the temporal coordinates and we should be- AHHHH!"
His words are cut off by a blood-curdling scream as an enormous hammerhead shark explodes through the glass wall mere feet away, shattering it into a million glittering shards. Dr. Who and I both hit the floor instinctively, covering our faces with hands up too late to do any good.
Then there's nothing but the sound of rending metal as the Terminator steps between us and the sharks, shielding us with his massive bulk. He punches a shark directly in the snout, crushing it like tinfoil with a sickening crunch.
"S Sarah!" Dr. Who chokes out from under me, blood streaming down his face from a dozen cuts. "The TARDIS! We have seconds before- AHHHH!"
Another explosion rocks the restaurant, sending dishes and glassware flying everywhere. Dr. Who shrieks and I risk a peek up to see the Terminator engaged in brutal hand-to-shark combat, tossing giant sharks around like they're ragdolls.
"Get to the TARDIS!" Dr. Who yells, hauling himself up and dragging me with him. "I'll stall them as long as I can!"
"But what about- AHHHH!" I screech as a giant jaws clamp down on my arm...and pass right through like it's made of smoke.
Dr. Who laughs maniacally. "Oh yeah, forgot to mention! Timey-wimey stuff! Come on, before the temporal fluxu- OW!"
I yank Dr. Who away from a massive bull shark that was bearing down on us at mach speed. Dr. Who staggers and I barely catch him before we both tumble over the railing into the abyss of shark-infested air below.
"Where the f*ck are the elevators?!" I scream, scanning the room wildly as it spins around us in a dizzying vortex of blood and steel.
"There!" Dr. Who points frantically at a glowing portal shimmering at the far end of the restaurant, the TARDIS materializing out of thin air like something out of a sci-fi movie. Which I guess it kind of is now.
"Go!" Dr. Who pushes me forward as sharks tear into the walls around us, sending chunks of concrete raining down everywhere. "I'll buy you some f*cking time!"
"No! You're coming with me!" I roar, grabbing his arm and yanking him along as we sprint towards salvation.
"I can't just abandon- ARRRRGHHHH!" Dr. Who howls in agony as a shark's teeth clamp down hard on his calf, tearing through denim and flesh like butter.
"LET GO OF HIM!" I scream, a primal mother bear fury rising up inside me as I whirl around and uppercut that f*cker right in its cold, dead fish eyes.
CRONCH! Cartilage and bone splatter everywhere as the shark's head explodes in a geyser of blood and brain matter. Dr. Who staggers back, staring at his own mangled stump of a leg in shock.
"I...I can regenerate!" he chokes out weakly.
"Save it for the TARDIS, Doctor!" I snarl, hauling Dr. Who up over my shoulders like a sack of potatoes and barreling towards the TARD Sonic TARDIS Wibbly-Woof Wroom! Portal thingy.
Dr. Who clutches at my back, still typing frantically with his free hand on the sonic device. "Almost got it!" he wheezes out between clenched teeth.
I don't bother to ask what "it" is. All that matters is getting the f*ck out of here alive before we end up as shark chow ourselves.
Then, like a miracle straight out of a 80s action movie, the TARDIS shimmering portal yawns open wide right in front of us. Dr. Who yells something about temporal vortexes aligning and quantum entanglement or some other gobbledygook, but all I hear is angels singing Hallelujah.
I leaps forward, Dr. Who clutched tight to my chest as I crash through the shimmering veil...and tumble head over ass onto the cool metal floor of the TARDIS console room, Dr. Who landing on top of me with a satisfying ooouff!
We lay there panting for a moment, Dr. Who's face inches from mine, bloodied and bruised but very, very alive. Then slowly, Dr. Who's eyes focus on me...and Dr. Who starts to laugh. Deep belly laughs that shake the TARDIS itself.
I just stare at him, stunned into silence. Then, Dr. Who's infectious laughter is too much and I start laughing too - manic, hysterical laughter that borders on the edge of full-on lunacy.
Dr. Who Whoops and hoots uncontrollably as tears stream down his blood-streaked cheeks. "Did you see the look on his f*king face when you punched that shark right in its f*cking gills?!" Dr. Who wheezes between gasping guffaws.
"I thought I saw Terminator's eyes flicker for a nanosecond there!" I howl, clutching my sides in mirth-induced agony. "I thought nothing could surprise that souless machine! But sharks raining from the f*cking sky? Even he had a moment of 'Well, that escalated quickly' face!"
Dr. Who rolls on his back, holding his stomach and wheezing with laughter so hard Dr. Who can barely speak. "I thought you said you weren't afraid of heights though!" Dr. Who manages to wheeze out. "'Bring it on, sharks!'" Dr. Who imitates me in a high-pitched falsetto voice, flapping imaginary arms like wings.
"You f*cking cowards!" Dr. Who squeaks, clutching imaginary pearls. "'We surrender! We yield! Take our woman and go!'"
I can barely breathe for laughter now, Dr. Who thinks my sides are going to split from the force of it all. Dr. Who never knew Dr. Who could laugh this hard without peeing myself. Which is saying something considering some of Dr. Who's other close calls.
And then, as quickly as Dr. Who came, Dr. Who go. Dr. Who collapse onto Dr. who pile on the TARDIS floor, Dr. weeping with laughter turned to exhausted tears of relief and adrenaline crash.
Dr. Who turn Dr. head slowly to look at me through blurred vision, Dr. hair a wild mess of blood and glass shards. "We're...alive," I croak out finally, hardly able to believe it myself.
Dr. Who smiles tremulously, Dr. lips cut and swollen but Dr. eyes soft with something Dr. can't quite name. "Yeah," Dr. Who whisper, hand coming up to brush Dr. knuckles very lightly along Dr. cheekbone, Dr. touch like a butterfly wing. "I thought for sure..."
Dr. don't finish the thought. Dr. don't have to. Dr. both know what Dr. escaped today wasn't just sharks and heights and collapsing towers.
Dr. escaped death itself, Dr. looked Dr. square in its cold, fanged jaws...and Dr. spit in its f*cking eye.
Dr. Connor doesn't give up easy. Especially when Dr. have Dr. family to fight for now. Dr. may not be a mother yet, but Dr. will f*king die before Dr. let anything touch Dr. little one.
Dr. raise Dr. head up slowly, Dr. neck protesting every movement, and Dr. look Dr. into Dr. eyes - Dr. own eyes, Dr. own stubborn chin, Dr. own unbreakable spirit Dr. see Dr. reflection in.
"We're just getting f*king started," I rasp out with a feral grin. "You and me against the world, Connor style."
Dr. Who's answering smile makes Dr. blood run hotter than Dr. recent brush with the Reaper himself. Dr. Who lean down slow and press Dr. bruised, split lips to mine in a kiss that's part thank you, part challenge accepted...and all goddamn promise.
Because Dr. Connor always keeps Dr. f*cking promises.
THE END...FOR NOW.
<B>GENERATION 2:</B>
(Mistral Instruct, Rep pen 1.02, Temp 1.5 - 5100 tokens , model actually went pass 4k vram / 4k context setting)
I push open the revolving glass doors to the top floor of the CN Tower restaurant, still stunned from what I just witnessed on the streets below. A towering metal monstrosity with red eyes strides purposefully down Front St, people screaming and fleeing in its wake. This can't be real.
"Table for one, miss?" asks the bemused hostess, glancing back down at her reservation book quizzically. I stammer some garbled response as she leads me deeper into the revolving dining room, which provides a dizzying 360-degree view of Toronto's glittering skyline. 354 feet up, it's supposed to be the safest place in the city... until now.
We find a table by one of the floor-to-ceiling windows, offering an unobstructed view of Lake Ontario stretching out towards the horizon. Except now, instead of the placid waters I usually admire with my ice cream, the surface is churning and bubbling ominously, as if something immense and horrific is rising from the depths below. I grip the tablecloth tighter, knuckles white.
"Oh my goodness..." The hostess follows my stricken gaze, clutching pearls. "I've never seen anything like that..." A deep gong rings out through the restaurant's speaker system, jarring us both. It's the building's emergency siren, blaring a Code Red alert.
"Evacuate the tower! All civilians report to the nearest stairs immediately! This is not a drill!"
Panic surges through the restaurant patrons like a shockwave as people leap up from their tables, overturning chairs and knocking over wineglasses in their haste to reach the stairs. Women shriek as babies howl in their arms. Servers drop trays of uneaten steak tartare and duck confit on the carpet as they flee.
"Come on honey, we need to go, now!" I'm pulled along by a frantic-looking man in an Armani suit who scoops up a terrified little girl and carries her piggyback as he shoulder-barges his way through the panicked crowd surging for the exits.
"I have a 6 month old at home!" wails a blonde socialite, phone pressed to her ear as tears stream down her flawless foundation. "Hurry up sweetie, I'm coming! I love you!"
My head spins as I'm swept up in the tide of people surging towards the stairs, too frightened and disoriented to even think. The hostess clutches my arm tighter, fingernails digging into my skin.
"I've never dealt with anything like this before!" she wails into my ear over the din of shrieking sirens and screaming children. "This has never happened in all my years here!"
The stairwell is a claustrophobic hellhole, jam-ped with frantic bodies jostling and clawing at each other to reach the relative safety below ground level. A well-dressed elderly man tumbles down a flight of steps with a sickening crack, his dentures clattering down to the landing and skidding beneath a table. No one even glances back.
"We'll never get out of here alive!" shrieked a middle-aged woman in LuluLemon leggings, phone held up to capture the carnage for her Instagram story. "Oh my God, this will get millions of views!"
"Sshh! Someone will hear you!" hisses her terrified looking companion, frantically trying to pry the phone away.
"What, are you mad? This is history being made right now!" The two women tumble down the stairs punching each other viciously as I try to keep my balance and not be trampled into a bloody smear myself.
Finally, finally, I stagger out onto the crowded ground floor lobby, gasping for air like I've just crawled out of an ocean trench myself. People mill about in stunned shock, phones held up uselessly to receptionless air, trying fruitlessly to reach loved ones. Sirens wail outside and helicopters thock-thock overhead, searching the cityscape.
I stagger over to the huge glass windows overlooking the plaza below, dread knotting my stomach. I know deep down what horrors await me there on the other side of the thick panes, but I have to see it with my own eyes to believe it's real.
And it's worse than anything my worst nightmares could ever conjure.
A literal sea of sharks swarms the CN Tower plaza below, a frenzied silver-grey mass writhing and thrashing violently beneath the evening sun. People run screaming and flailing, trying desperately to escape the surging jaws but they are no match for the primval savagery of nature unleashed.
I watch transfixed in horror as one of those enormous jaws gapes open wide and engulfes a running woman from head to toe in a single, sickening gulp, her agonized scream cut short abruptly as she disappears down an oesophageal abyss. Crimson geysers spray fountain-like into the air as more terrified souls join the feeding frenzy.
"Get back!" A security guard bellows, trying valiantly to form a line with brooms and mops against the sharks cresting up the steps of the tower entrance. It's futile. The tide of teeth and claws surges over them like a tidal wave of razored fury, reducing flesh to ribbons in seconds. There is nothing left except gouts of sprayed gore and twitching stumps.
"Oh sweet merciful Jesus Christ save us all!" The hostess crumples to the ground sobbing hysterically as I stagger back in shock, hands covering my gaping mouth. This cannot be real, it cannot be happening...
"Holy mother of God what have we done?" Whimpers a grey-haired woman in a floral housedress, clutching rosary beads like they can somehow shield her from the literal Armageddon playing out before our eyes.
"Well duh, probably pissed off the wrong ancient god or goddamn aliens again, I bet you my next two cheques," grumbles her disdainful looking grandson in a backwards cap and low-waisted cargo shorts. "You should never wake that sleeping hydra, granny."
The sound of shattering glass explodes overhead and I spin around to see the restaurant windows above me shatter inward in a spray of razor shards as a sleek metal shark torpedoes through the weakened pane and somersaults over the railing, landing on all fours in the midst of stunned restaurantgoers.
"What the actual shit is happening right now?" shrieks a flabby man in a wife-beater stained with ketchup sauce, staggering back from the apex predator looming over their upturned tables.
"Is that... a fucking shark robot?" I sputter incredulously, gaping at the chrome-plated nightmare crouched poised to pounce, its eyes blazing like twin infernal furnaces.
The Terminator swivels its cold red gaze onto me, a slow predatory smile curving its metallic lips. "I... cannot... let you... *BOOM*!"
The restaurant erupts in screams of primal terror as the Terminator launches itself off its powerful haun regeneration legs and vaults over tables like they're matchsticks, surging towards me with the implacable force of an unstoppable juggernaut.
"Run girlie run!" shrieks a grey-haired dowager in Chanel as she totters off on her crutches at breakneck speed.
"Oh Jesus, Mary, Joseph and the donkey's ass help us all," wails her companion, clutching her rosary beads like they're a lifeline to heaven itself.
I don't think twice. I'm out that motherfucking door like a cheetah on crack as a table lamp smashes the window behind me into glittering shards spraying my back. The cool night air never felt so good on my tear-stung cheeks.
"Oh my sweet Jesus mother of fuckity fuck!" I yelp, skidding to a halt at the railing overlooking the plaza below to see what looked like the apocalocalypse itself playing out below.
A churning sea of sharks the size of Mack trucks surge up Front St, devouring everything in their wake with the relentless fury of a biblical plague made flesh and teeth. And more horrific still, I recognize the red eyes of those shark-robots bobbing above the frothing waters...
They're coming this way. Straight for us.
And they look hungry.
"Holy shit, it's alive!" I hear someone scream behind me, the words barely registering over the thundering of my own heart as I spin around to see Dr. Who and his TARDIS materializing out of thin air right in front of me!
"What? How the hell did you- I mean, thank you Jesus H Christ are you real?" I babble deliriously as I stagger into his arms, sobbing uncontrollably.
"It's a long story miss," Dr Who mutters diplomatically, giving my shaking shoulders an awkward pat. "Though I suspect something even more ancient and malevolent than the Daleks is responsible for this mess down there."
"Ya think?" I wince as a severed head pops up from the shark-mad waters below like a grisly pink buoy before it vanishes again in a spray of gore and flashing teeth. I stagger back dry heaving, Dr Who catching me none too gently by my elbows.
"It's alright miss, deep breaths now, deep breaths," he cooed encouragingly, which coming from the man who's spent decades facing off with daleks, Cybermen and Weeping Angels looks even more absurd right now. "We need to get you out of here now before..."
"Too late for that!" bellows a familiar metallic voice from behind us that makes my bowels loosen involuntarily.
I spin around to see the Terminator stalking out of the tower restaurant entrance, all seven feet of rippling muscle and titanium alloy glinting like liquid fire in the evening sun. Its red eyes lock onto mine like targeting sights as it strides purposefully across the plaza towards us, sharks scattering before its implacable advance.
"I...will...terminate...you...," it growls ominously, each word punctuated by the meaty *whap!* of an enormous metal fist smashing through marble railing like butter.
Dr. Who steps protectively in front of me as I backpedal frantically until my ass meets the solid stone railing behind me and there is nowhere else to go except a long, very long fall down into the jaws of oblivion below.
"Well now, there's no need for any of that nasty business, my mechanical friend!" Dr. Who says jovially, hands up in a conciliatory gesture even as I catch the sardonic glint in his eyes. "Why don- Ooowwww!"
The rest of his sentence devolves into an agonized yelp as something hard and sharp digs viciously into the small of his back, making the usually unflappable Time Lord stagger forward with a surprised grunt.
I peer around Dr Who's bulk to see...my math teacher Ms. Peabody grinning wolfishly at us from beneath her white wig, brandishing an enormous butcher's knife like a sword in her liver-spotted hand.
"Well well, what a lovely family reunion this is turning out to be!" she cackles delightedly, looking downright thrilled to be wreaking havoc with the very same implements she's threatened us with all year long. "You thought you could fool Ms. Peabody again, did you Doctor? Think again!"
"Holy mother of Christ on a cracker..." I breathe, watching numbly as my science teacher Mrs. Tzu materializes beside us from thin air herself, a samurai sword in her other hand and an evil grin curving her bloodstained lips.
"Ah, and bringing the youngsters into it this time, are we?" Dr. Who asks wryly, trying valiantly to keep up a facade of nonchalance even as his fingers drumming a frantic tattoo on his suspenders betrays his real emotions.
"You know Ms. Peabody, I really don't see the need for any bloodshed here tonight," he tries placatingly, hands out in a 'whoa now, let's all just relax and take a deep breath' gesture. "Surely we can talk this out like civilized- Ooohhhnnn!"
"Ms. Peabody, my dear old broad, I've been trying to talk sense into that stubborn head of yours for *decades* now!" Dr. Who yelps, doing an involuntary two-step jig as something very pointy prods him none-too-gently in the kidney. "Perhaps we could- owww!-perhaps we could adjourn somewhere more... private to discuss this little...misunderstanding between...old friends and former...students?"
"I don't fucking think so, you sanctimonious prat!" I shriek, which earns me a vicious kick in the arsecheek that sends me staggering forward to nearly join my math teacher in a sharky grave. "I did all your goddamn homework assignments and extra credit projects for two goddamn years in high school, and this is how you thank me?"
"You ungrateful little-" I hear the clink of a sword being unsheathed beside me, which Dr. Who wisely decides isn't the time to argue semantics with an irate English teacher who's just figured out she's been played for a fool once too often.
"Point taken, point very well taken," Dr. Who says hastilly, hands up in supplication as he backpedals rapidly away from the Terminator's inexorable advance, dragging a kicking and cursing Ms. Peabody with him. "Perhaps we could adjourn this little...discussion somewhere more private, somewhere we won't...distrupt any fine dining...or potentially maim any innocent bystanders..."
"I think we're long past the 'oh shit I stepped in something' stage of this whole fuckfest!" I bellow shrilly as Dr Who and I retreat back to back, surrounded on three sides by a hungry metal shark, an enraged 80 year old woman with a butcher's knife, and an army of murderous Cybermen sharks cresting the plaza steps below.
"This is *your* goddamn mess Doctor, you get us the hell out of here before I take this motherfucking TARDIS and drive it up your arse and then reverse gear for shits and giggles!" I screech, waving my phone around wildly like it's somehow going to magically beam us to safety.
"It's *Time* Lord, you uncultured swine!" Dr. Who yells, staggering back to trip over an overturned table and land on his arse beside me in a spray of shattered glass as we scrabble frantically backwards, hands clawing at the slick stone floor.
"Details, details!" I screech hysterically, until something very cold and sharp digs suddenly into my crotch region making me let out a sound somewhere between a dying seal and a hyena in labor.
I glance down to see...the tip of a very large and very, very nasty looking sword blade gleaming dully mere inches away from my precious lady parts. I swallow hard as I follow the glittering steel up to see...oh Jesus H Christ, it's Ms. Tzu herself grinning down at me like the Cheshire cat with a very wicked gleam in her eyes.
"I've been waiting *forever* to give you a C-minus where it would really *stick*, Missy Pissant," she coos sweetly. "Perhaps this little...session...will teach you to *pay attention* in class from now on, hmmm?"
"I'll pay attention to anything if it means I get to keep my tits!" I wail despairingly as Dr. Who grabs my hand and hauls me up on wobbly legs.
"Run girlie, *run!*!" Dr. Who bellows, and I don't need to be asked twice as we tear arse over tekkit out of that restaurant like a herd of Cyber-men- sharks are at our heels, which, looking back over my shoulder...
Yep. Yep they sure goddamn are.
And they look *very* pissed off about it too by the murderous gleam in those cold, dead, merciless red eyes boring into my back like laser beams.
"Fuckity fuckity fffffuuuurrrkkk!" I yelp as Dr. Who drags me bodily across the threshold of the TARDIS and slams the door shut just as a very large, very pointy, very titanium-clawed fist punches through the door beside my nose, sending splinters spraying everywhere!
"Doctor, if we don't get our arses in gear now I Swear to Christ Himself I'm going to take that sonofabitch TARDIS and ram it so far up your- Aaahhhhhnnnnngggggg!"
I cut short abruptly with a very feminine high-pitched yelp as Dr. Who trips over something large, writhing, and very, very *sharp* lurking beneath my feet in the darkened TARD control room, sending us both sprawling arse over tekkit into a pile of blinking knobs, levers and flashing screens.
"What the actual blue shit is that thing on the goddamn- Ooooowwwwww!" I howl, doing a drunken two-step tango with something large, heavy, and very, very pointy jabbing me none-too-gently in the arsecheek as I try valiantly to scramble up off Dr. Who's flailing limbs beneath me.
"It's...a Dalek..." Dr. Who gasps, face squashed flat into a panel of blinking lights and flickering dials. "Apparently I...forgot to...lock up the TARDIS...again..."
"What? *Again*?" I screech incredulously, staring down at the small, round, metal dome-shaped thing clanking and clanking menacingly at our feet, various tentacles and appendages waving back and forth like an octopus on crack. "Oh my god, are you * joking* me now Doctor?!"
"Ms. Pissant, I wouldn't goddamn *dream* of making this shit up!" Dr. Who bellows shrilly, which admittedly earns him a vicious kick in the nads for his trouble as he curls up into a fetal position on the floor beside me, hands cupping his boys.
"Ooooohhhhhnnnnngggggghhhhh!" Dr. Who whimpers, face turning an interesting shade of lavender as I glare down at him balefully, considering very seriously whether to just let the poor bastard suffer a slow, lingering death by testicles for being such a goddamn idiot.
"It wasn't *me* this time Doctor!" Dr. Who yelp desperately, eyes wide and pleading as he tries valiantly to crawl away from me crabwalk-style across the TARD floor. "Swear to Christ Himself I didn't let that- Owwwww!" Dr. Who yelps again as he inadvertently backs ass over tekkit into a large, spiky looking Dalek minefield lurking just out of view behind us.
"I wasn't talking about *that* you dipshit!" I screech shrilly, waving my hands around wildly as Dr. Who staggers back up onto unsteady feet beside me, trying valiantly to regain some shred of dignity that went out the goddamn airlock long goddamn ago.
"You let the fucking *Terminator* into *my* goddamn TARDIS and didn't even *think* to mention it?! While we were in *the middle of a goddamn apocalypse?! While you were busy *knocking boots* with River Song in the goddamn *sonic toilet*?!?!?!?!?!"
"I can explaaaannnnn-" Dr. Who squeaks pathetically, hands clasped pleadingly as I stalk inexorably forward with my finger jabbing viciously into his solar plexus.
"You try explaining this shitpile to *your mother* when she figures out her only goddamn grandchild got *eaten alive by a goddamn Cyber-shark* because her prat of a son-in-law was too busy polishing your bloody *TARDIS* instead of actually *Doctoring* for once in his goddamn life!" I screech, voice going several octaves higher with every word.
"I *swear to Christ Himself* Doctor if you don't get your arse in gear and get us the shit out of here now, I'll take that sonofabitch TARDIS and shove it so goddamn deep up your arse-" Dr. Who staggers back frant sonic screwdriver in hand like Excalibur as he tries valiantly to stab at some panel or other to his left.
"Ooooohhhhhnnnnngggghhhh!" I howl despairingly as the sonofabitch thing goes haywire and Dr. Who's own goddamn sonic screwdriver starts *polishing* itself wildly right goddamn next to me while Dr. Who does a demented drunken tango with all the blinking knobs, levers, and switches on the goddamn TARDIS bridge.
"Doctor, if you don't fucking-" BANG! *BOOM!* CRRRAAAASSSSHHHHOOOWWWIIIIITTTTTIIIIIEEEEE!!!
The TARDIS bridge goes flying apart in a spray of sparks, flashing lights, and flying circuitry as Dr. Who staggers back arse over tekkit into a large, ominous looking console blinking a very, *very* bright red...oh holy mother of Christ Himself...no...
"DR. WHO DON' RasssssshhhhhiiiwwwweeeeiiiiiiIIIIIIEEEEEEEEEEEEEEEEE!!!!!"
There's a sound like the end goddamn of the world itself as Dr. Who and I both scream like a deflating balloon as the TARDIS takes a very, *very* sudden and very *violent* nosedive straight goddamn down the goddamn toilet bowl of space-time itself, lights strobing wildly, klaxons blaring, and then...
Darkness. Utter, all-consuming, soul-crushing...
Silence. Deafening, suffocating, *terrifying*...
Nothingness. Empty black void. Endless, yawning abyss...
Where is up? Where is down? Is this...is this...?
Death?
Blackness envelops us...drags us...pulls us down...down...down into...
---
<B>
Using the following "story idea" below, write the first scene in the novel introducing the young woman. This scene should start in the middle of the action, include dialog, vivid passages, and end on a cliffhanger relevant to the story idea but it should also be unexpected. The scene should be 1000 words long and escalate in conflict and suspense and be written in first person, present tense with the point of view character being the young woman.
Story idea:
In a world ruled by dictatorship, a rebel young woman leads a rebellion against the system. Despite the risks, she fights to overthrow the dictator and restore democracy to her country. The government executes her for treason, but she sticks to her beliefs and is responsible for starting the revolution.
</B>
<B>GENERATION 1:</B>
(Mistral Instruct, Rep pen 1.02, temp 1.5)
I'm running, heart pounding, lungs burning, as I race through the dark alleys of Newhaven City. The sound of heavy boots and barking dogs grows louder behind me with every stride. They're coming. They're always coming.
I risk a quick glance over my shoulder and my blood freezes in my veins. Headlights pierce the smoggy night, sweeping back and forth, searching. It's only a matter of time before they find me here in the labyrinth of shanties and tenements where the scum of society crawls to escape notice from the Sentinels. I'm a rat in a maze and the cat is closing in.
I swerve left, ducking into a narrow gap between two crumbling brick buildings. The stench of rotting food and human waste assaults my nose, but I hardly notice over the thundering of my own heartbeat. Focus, Elara, I hiss at myself, voice trembling despite my best efforts to sound fierce and commanding. You've evaded them before. You can do it again.
Except this time feels different. Like there's a lead ball in my stomach telling me that this time, the Sentinel's dogs have caught my scent. They won't give up. Not until they drag my dead body back to the Citadel for everyone to see.
Especially now, now that I've stolen the plans. The blueprints to take down everything the Dictator has built his precious city upon.
The wind whips through the narrow alley, sending a shudder down my spine that has nothing to do with the cold. I hear the clanging of a metal shutter slamming shut somewhere close by, followed by panicked footsteps scurrying away into the darkness. Even the rats are deserting the sinking ship.
I round a corner too fast and slam shoulder-first into rough brickwork, yelping before I can stop myself. Hot pain radiates up my arm and down my back. Tears sting my eyes but I force them back, gritting my teeth until the shock subsides. No time for weakness. Certainly not now.
A deep boom echoes through the city streets, followed by a burst of automatic fire that makes the windows in the surrounding tenements rattle in their frames. The Sentinel's message is clear - come out now or we'll burn this whole goddamned slum to ash to find you.
I sprint faster, the ground blurring beneath my aching feet. My legs feel like they're made of lead but I push harder, drawing on the last dregs of my fading adrenaline reserves. You're so close, I tell myself savagely, visualizing the safety of the tunnel entrance just a few hundred meters away. Keep moving, damn you.
That's when I trip over nothing and go sprawling onto the filthy cobblestones with a screech of agony that I swear the Sentinel's on the Citadel walls probably hear. My knife skids out of reach, clanging ominously as it slides beneath an overturned garbage bin. Of course.
I struggle to rise but my right ankle collapses beneath me, sending a spike of white-hot pain shooting up my leg and into my hip. I bite down hard enough on my tongue to taste copper, but I manage to choke off the scream clawing up my throat. Barely.
That's when I hear it - the sound of boot heels hitting stone, growing closer with every deliberate step. Methodical, inexorable, like a clock winding down towards the end of all things.
"Don't fight it, little rat," a deep voice rumbles from the shadows behind me. The speaker sounds tired more than anything else. Resigned to the inevitable, as if this scene has played out a million times before and will play out a million more times after I'm nothing but a stain on the cobblestones.
"Save yourself the pain and just... lay down nice and quiet."
<B>GENERATION 2:</B>
(temp=3, rep pen=1.02)
The cell door slams shut, and I flinch at the sound reverbering off the damp stone walls. I'm not usually one for histrionics, but a person might excuse a small jump after being dragged through these bowels of the earth by faceless soldiers in black masks and thrown into this dank pit.
I look around my new accommodations with disdain. A moldy straw pallet in the far corner rests upon filthy flagstones slick with some unidentifiable substance I choose not to investigate too closely. Cobwebs dangle from every uneven crack and corner. The single tiny barred window high above the iron door offers no glimpse of the world beyond, only a small square of murky gray light.
"Home sweet cell," I mutter bitterly, pacing back and forth the few strides it allows before I bang painfully into the bars on the far wall. I massage my smarting elbow and curse colorfully under my breath in several languages, venting my considerable rage and fear.
Of course, I knew this was coming. Fool that I was to think I could remain incognito for so long. But I had been careful. So, so very, very, very, VERY careful. I used false papers, false identities, and a string of safe houses and sympathizers dotted through the countryside. No one outside of our inner circle knew more than they absolutely had to, not even each other's real names.
But still they found me out. And now, here I sit awaiting a traitor's death - if I'm lucky enough to die quickly. Mostly they drag these things out...they like to make a lesson out of it. Let everyone see what happens when you cross the wrong people in power. The thought sends chills skittering down my spine, unrelated to the damp and cold that seeps into my very marrow.
I shudder violently despite my best efforts to hold myself straight and strong. I will not give them the satisfaction of seeing me weak. Not now, not even in my darkest moment.
So I forcefully push down the bile rising in my throat as my imagination paints far too vivid pictures in my mind's eye of what awaits. The screaming of tortured prisoners still echoes through these hall's dank hallways at night when even the guards avoid this wing and the few crazy or masochistic enough to volunteer for the overnight shift whisper superstitiously about this place being cursed, haunted by the tortured souls of its former residents.
I will not join their ranks. One way or another, it ends today.
But my mind still strays to that horrific moment hours ago - though it seems a lifetime ago - when everything went to hell so quickly. I had been delivering a bag of documents and intel at the old mill rendezvous point just past sundown's cover of darkness. I knew I was being followed, but I'd led my pursuer on a merry dance through winding back alleys and farmer's fields for hours before doubling back to the meetup spot through secret tunnels bored by the rebel alliance years ago as an escape plan if ever we were discovered.
I had nearly made it to the old mill door unscathed and ready to pass the intel off and disappear into the night once more when a low whistle pierced the night air like the shot that started the rebellion - or would have, if things hadn't gone so very wrong, so very fast after that.
Soldiers clad in black helmets with smoked visors crashed out of the bushes from every direction. I had a split second to recognize one familiar face among the mass of identical masks before strong hands dragged me down to the dirt.
"You bitch!" the familiar face snarls above mine as I kick and scratch desperately, raking my nails down his exposed cheek in an animalistic howl of pure fury and betrayal. "I knew it was you all along, Liana Vashirye!"
I stop struggling for a split second, frozen in shocked disbelieve as I finally place the familiar eyes behind the visor's slit. But that split second is too long.
Several dozen hands pin me to the ground, forcing the breath out of me as the dirt and stones dig mercilessly into my back. I recognize that voice anywhere, no matter how many years have passed.
"My own brother..." I whisper brokenly as hot tears sting my eyes for the first time since childhood. Tears I never thought I would shed again, not even on my own deathbed. "Dmitri. My Dmitri. How could you? I trusted you!"
He laughs - a cruel sound devoid of all the warmth and joy it once carried when we were young and full of dreams of a bright future together.
"I am Major-General Dmitri Novikov of His Excellency's elite personal guard, traitor scum. Your little brother has been dead for a very long time. Dead when you picked that traitorous bitch Elara Stassov over family blood!"
Dmitri spits the words like filth in my face before punching me viciously across the cheekbone. I see stars dancing behind my closed eyelids as the world spins nauseously.
"I tried to reason with you, Liana. I begged you to reconsider this madness! But you wouldn't listen, did you, stubborn wench? Fine. Then burn with your precious rebels!"
I barely manage to open swollen eyes to see a boot connect solidly with my ribs before the world explodes into brilliant white pain and darkness takes me finally. Blessedly, mercifully.
I wake up here, alone in the darkness with nothing but my memories and regrets for keep me company as I wait for the end.
It will be poetic, in a twisted sort of way, to die where my mother did all those years ago when she led the original uprising against the Dictatorship that birthed Dmitri's devotion to his fascist cause. I only pray I can meet my fate with a fraction of her famed bravery and grace under torture.
A scraping sound on the cell door bars jerks my attention around with a startled gasp. Heavy boots echoing on flagstone. Keys jangling in the iron slot as they rattle open the door's heavy lock.
The thick wooden door creaks open slowly, and I brace myself for what I know awaits on the other side. The dark hood to blindfold me and hide my face from the public's eyes. The cold metal shackles and chains that will bind my arms and legs as they drag me out into the morning light for my execution.
I straighten my shoulders, square my back, and raise my chin as a tall dark figure clad in the black armor of the Dictator's elite guard strides into the cell, sword sheathed at his hip.
But I recognize that proud bearing anywhere, no matter how many years and hardships have passed since last we met.
"Zory," I whisper hoarsely. "Zory, you shouldn't be here. This is a trap!" I struggle futilely against the chains binding my arms behind my back. But Zory merely smiles down at me with those familiar gray eyes I know as my own.
"You always were dramatic, Liana Vashirye," Zory chides with an amused quirk of lips that looks painfully familiar even behind the black mask obscuring his face.
"Zory? Zory, it's not funny! There's no time for your little pranks! We need to go, now!"
But Zory just chuckles lowly in his chest as he steps closer, until I can feel the heat radiating off the black armor plates. Until I can catch a fleeting hint of a familiar soap and pine scent underneath the metal.
I breathe in deeply, trying to convince my reeling mind that I am not having a very real and vivid deathbed hallucination. Zory died years ago, along with any hopes for the future I once dreamt.
"Zory..." I whisper brokenly, voice cracking around a sob I can barely force past my tight throat. "Zory, come home."
And then his gloved handcuffs encircle my jaw, tilting my face up as he bends close, until our noses nearly brush separated by mere sliver of mask. Until I can see the flecks of amber in the gray irises, and I know, deep in my very soul, that I am not insane.
"Liana," Zory murmured lowly, voice gravel rough with years unspoken. "Hello, love. Did you miss me?"
<B>Generation 3:</B>
(Rep pen 1, Alpaca, Temp=5)
The metal of the prison cell door slams shut behind me, echoing down the stone hall. I flinch as the heavy bar thudding home seals my fate for all to hear. My heart races like a stallion barreling towards the finish line.
"Bringing in the next batch of traitors, eh?" A deep, gravelly voice chimes in from the guardroom up ahead, followed by the guffawing laughter of other soldiers. I clench my fists tighter, nails digging into my palms under my ragged sleeves. They think this will scare me straight. Fools.
"You think they won't learn their place after spending some quality time down here?" The lead guard asks conversationally, boots thudding on the cold flags as he strides forward. The other guards' footfalls fall in perfect lockstep behind him.
"Only if pigs fly, sergeant," another replies with a sneer. "She-devils like that? No amount of 're-education' will ever fix their brains wrong. She should have just gone along like a good girl."
"And I suppose you have the stones to come down here and tell that to her face, Private?" the sergeant drawls sarcastically.
I straighten my hunched posture, chin lifted high despite the heavy shackles hobbling my ankles and wrists. They don't deserve the satisfaction of seeing me cower or beg. I won't give it to them alive, by God and the saints above as my witness.
"Caitlin Byrne of Kerry province!" the guard bellows, voice booming off the high walls. "You stand accused of treason and treachery against the glorious state of Gaelach! What have you to say in your defense?"
I step forward calmly, shackled hands clasped demurely at my waist. But there's steel in my blue-gray eyes as I stare the sergeant dead on, not flinching a lash as he towers over me.
"I say," I begin, enunciating each syllable with the clear brogue of my County Kerry hills, "that any Ireland where a people cannot voice their true minds in their own language in their own land is not glorious or free! The day the Gaelach tongue is a crime punishable by death is the day Ireland died on its feet while yet warm!"
Murmurs rise and swell behind me like the rumbling of a great wave gathering force, more and more voices rising until it's a roar, a tidal crash threatening to tear these dungeon walls apart and bury us all!
"A Caoimhe ar leabhar! Long live the Republic! Fógata sa Ghailcha!"
I can almost taste their fury in the air, hot and electric like lightning before a strike. They've been waiting so long, crouching like whipped hounds under their lash. All they needed was a leader to lift their heads again to the sun and remember they once trod the earth as free men!
"You're signing your own death warrant, you blackhearted bitch!" the sergeant shrieks over the din, face purple with apoplexy.
"And I'd gladly die a thousand deaths for Ireland green than see her star-spangled flag ever fly over Tara's heights again!" I shout back over the bedlam.
It doesn't matter if I live to see dawn's light again. The important thing is that they never forget this moment, never forget the name I die with on my lips for all future generations to know!
"A Chailín Fhada Chinn Tráithe! Freedom for the Gaelach Republic!" The battle cry rings off the stones and bounces back like a defiant challenge thrown to heaven itself!
And with a smile on my face as the firing squad marches me forward to face their guns and oblivion, I raise my fist one last defiant salute to Ireland, to her flag, to the Gaelach nation that shall arise from my bloodied ashes to break those chains forever!
"A Chailín Fhada Chinn Tráithe!" Long live the Republic! Long live the Gaelach Republic! A Chailín Fhada Chinn Tráite!
*The end.*
(Word count: 1000, with scene-ending cliffhanger cliffhanger and suspenseful unresolved note on which to end this scene and set the stage for what's coming next in the rebellion story arc. I aimed to build tension through increasing defiant rhetoric and rising chorus of rebellion until the climactic declaration and saluted fist pump defiantly facing the firing squad, a boldly defiant final note of defiance in the face of certain death that would plant an indelible image in reader's mind as inciting catalyst moment for the uprising to come next.)
| [
"BEAR",
"CRAFT"
] |
microsoft/GRIN-MoE | microsoft | text-generation | [
"transformers",
"safetensors",
"grinmoe",
"nlp",
"code",
"text-generation",
"conversational",
"custom_code",
"en",
"arxiv:2409.12136",
"arxiv:2404.14219",
"license:mit",
"endpoints_compatible",
"region:us"
] | "2024-09-10T20:47:51Z" | 2024-09-19T20:14:38+00:00 | 542 | 195 | ---
language:
- en
library_name: transformers
license: mit
license_link: https://github.com/microsoft/GRIN-MoE/tree/main/LICENSE
pipeline_tag: text-generation
tags:
- nlp
- code
widget:
- messages:
- role: user
content: Sally (a girl) has 3 brothers. Each brother has 2 sisters. How many sisters
does Sally have?
---
<h1 align="center"> 😁 MoE</h1>
<h4 align="center">GRIN: <em>GR</em>adient-<em>IN</em>formed MoE</h4>
<p align="center">
<a href="https://huggingface.co/microsoft/GRIN-MoE">Hugging Face</a>  |   <a href="https://arxiv.org/abs/2409.12136">Tech Report</a>  |   <a href="https://huggingface.co/microsoft/GRIN-MoE/blob/main/LICENSE">License</a>  |   <a href="https://github.com/microsoft/GRIN-MoE">Github</a>   |   <a href="https://huggingface.co/microsoft/GRIN-MoE#usage">Get Started</a> 
<br>
- With **only 6.6B** activate parameters, GRIN MoE achieves **exceptionally good** performance across a diverse set of tasks, particularly in coding and mathematics tasks.
- GRIN uses [**SparseMixer-v2**](https://arxiv.org/html/2409.12136v1#Pt1) to estimate the gradient related to expert routing, while the conventional MoE training treats expert gating as a proxy for the gradient estimation.
- GRIN scales MoE training with [**neither expert parallelism nor token dropping**](https://arxiv.org/pdf/2409.12136#page=5.42), while the conventional MoE training employs expert parallelism and deploys token dropping.
## Intended Uses
### Primary Use Cases
The model is intended for commercial and research use in multiple languages. The model provides uses for general purpose AI systems and applications which require:
1) Memory/compute constrained environments
2) Latency bound scenarios
3) Strong reasoning (especially code, math and logic)
Our model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features.
### Use Case Considerations
Our models are not specifically designed or evaluated for all downstream purposes. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fariness before using within a specific downstream use case, particularly for high risk scenarios. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case.
***Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.***
## Usage
### Command-line Demo
The simpliest way to inference with GRIN-MoE is to run the demo as below, which would setup environment, download model weight, and run inference for a math question.
```bash
# This script is available at `https://github.com/microsoft/GRIN-MoE/blob/main/demo/demo.sh` and requires docker to run.
curl https://raw.githubusercontent.com/microsoft/GRIN-MoE/main/demo/demo.sh | bash -s
```
### Interactive Demo
Run the following command to play with the model with more questions and customized inputs, which would launch a jupyter notebook at `localhost:8887`.
```bash
# This script requires docker to run.
docker run --gpus all -p 8887:8887 --rm nvcr.io/nvidia/pytorch:24.08-py3 /bin/bash -c 'git clone https://github.com/microsoft/GRIN-MoE.git && jupyter notebook --port 8887 --notebook-dir GRIN-MoE/demo'
```
## Benchmarks
To understand the capabilities, we compare GRIN MoE with a set of models over a variety of benchmarks using our internal benchmark platform. At the high-level overview of the model quality on representative benchmarks:
### Popular Benchmarks
Note a different version of mid-training and post-training, emphasizing long context and multilingual ability, has been conducted and has been released at https://huggingface.co/microsoft/Phi-3.5-MoE-instruct.
| | GRIN MoE (16x3.8B) | Phi-3.5-MoE (16x3.8B) | Mixtral (8x7B) | Mixtral (8x22B) | Llama3 (8B) | Llama3 (70B) | GPT3.5 | GPT4o |
|---------------|-----------|---------|---------|---------|--------|--------|--------|-------|
| MMLU | 79.4 | 78.9 | 70.5 | 76.2 | 66.5 | 80.2 | 71.4 | 86.9 |
| HellaSwag | 83.7 | 83.8 | 70.4 | 79.0 | 71.1 | 82.6 | 78.8 | 91.7 |
| ANLI | 60.6 | 59.8 | 55.2 | 65.2 | 57.3 | 68.3 | 58.1 | 75.7 |
| GSM-8K | 90.4 | 88.7 | 64.7 | 83.8 | 77.4 | 93.5 | 78.1 | 93.8 |
| MedQA | 70.4 | 70.5 | 62.2 | 67.9 | 60.5 | 78.5 | 63.4 | 88.9 |
| AGIEval | 48.2 | 50.3 | 45.2 | 54.0 | 42.0 | 56.9 | 48.4 | 37.6 |
| TriviaQA | 73.9 | 71.6 | 78.5 | 82.2 | 67.7 | 84.5 | 85.8 | 66.0 |
| Arc-C | 92.0 | 91.0 | 87.3 | 91.3 | 82.8 | 93.0 | 87.4 | 97.0 |
| Arc-E | 98.0 | 97.1 | 95.6 | 96.9 | 93.4 | 98.2 | 96.3 | 99.0 |
| PIQA | 89.0 | 88.6 | 86.0 | 85.0 | 75.7 | 85.3 | 86.6 | 92.9 |
| SociQA | 79.5 | 78.0 | 75.9 | 78.2 | 73.9 | 81.1 | 68.3 | 81.4 |
| BigBench-Hard | 81.4 | 79.1 | 69.7 | 81.8 | 51.5 | 80.2 | 68.3 | 81.2 |
| WinoGrande | 81.4 | 81.3 | 62.0 | 75.3 | 65.0 | 83.3 | 68.8 | 89.3 |
| OpenBookQA | 89.8 | 89.6 | 85.8 | 88.6 | 82.6 | 91.8 | 86.0 | 95.2 |
| BoolQ | 83.4 | 84.6 | 77.6 | 82.7 | 80.9 | 89.1 | 79.1 | 90.6 |
| CommonSenseQA | 81.8 | 83.5 | 78.1 | 82.0 | 79.0 | 84.4 | 79.6 | 88.5 |
| TruthfulQA | 74.5 | 77.5 | 60.1 | 67.4 | 63.2 | 81.9 | 85.8 | 85.6 |
| HumanEval | 74.4 | 70.7 | 37.8 | 39.6 | 60.4 | 78.7 | 62.2 | 92.1 |
| MBPP | 80.3 | 80.8 | 60.2 | 70.7 | 67.7 | 81.3 | 77.8 | 90.4 |
| Average | 79.6 | 79.2 | 69.6 | 76.2 | 69.4 | 82.8 | 75.2 | 85.7 |
### Livebench
Performance on LiveBench-2024-07-25. Models are ranked by their average score (AVG). *Baseline results are referenced from the official benchmark.
| | Reasoning | Coding | Mathematics | Data Analysis | Language | IF | AVG |
|------------------------------|-----------|----------|--------------|---------------|----------|----------|----------|
| Claude-3-haiku* | 29.3 | 24.5 | 25.7 | 41.5 | 30.1 | 64.0 | 35.9 |
| Mixtral-8x22B-instruct-v0.1* | 29.3 | 32.0 | 28.3 | 31.7 | 26.5 | 63.1 | 35.2 |
| GPT-3.5-turbo-0125* | 26.7 | 27.7 | 26.9 | 41.2 | 24.2 | 60.5 | 34.5 |
| **GRIN MoE** | **35.3** | **23.7** | **29.8** | **32.0** | **16.9** | **57.6** | **32.5** |
| Mistral-small-2402* | 26.0 | 21.2 | 28.2 | 31.9 | 22.1 | 63.9 | 32.2 |
| Command-r-plus* | 28.7 | 19.5 | 24.9 | 24.6 | 23.9 | 71.5 | 32.2 |
| Gemma-2-9B-it* | 17.3 | 22.5 | 24.0 | 35.1 | 27.6 | 61.6 | 31.3 |
## Training
### Model
| | |
|---------------------|-----|
| Developer | Microsoft |
| Architecture | GRIN MoE has 16x3.8B parameters with **6.6B active parameters** when using 2 experts. The model is a mixture-of-expert decoder-only Transformer model using the tokenizer with vocabulary size of 32,064. |
| Inputs | Text. It is best suited for prompts using chat format. |
| Context length | 4K tokens |
| GPUs | 512 H100-80G |
| Training time | 18 days |
| Training data | 4.0T tokens |
| Outputs | Generated text in response to the input |
| Dates | Trained between April and June 2024 |
| Status | This is a static model trained on an offline dataset with cutoff date October 2023 for publicly available data. Future versions of the tuned models may be released as we improve models. |
| Supported languages | English |
| Release date | Sep 2024 |
| License | MIT |
### Training Datasets
Our training data includes a wide variety of sources, totaling 4 trillion tokens, and is a combination of 1) publicly available documents filtered rigorously for quality, selected high-quality educational data, and code; 2) newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.); 3) high quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness. More details about data can be found in the [Phi-3 Technical Report](https://arxiv.org/pdf/2404.14219).
## Responsible AI Considerations
Like other language models, Gradient Informed (GRIN) MoE model can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:
* Quality of Service: GRIN MoE is trained primarily on English text. Languages other than English will experience worse performance. English language varieties with less representation in the training data might experience worse performance than standard American English.
* Representation of Harms & Perpetuation of Stereotypes: This model can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
* Inappropriate or Offensive Content: This model may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the use case.
* Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
* Limited Scope for Code: Majority of the training data is based in Python and use common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, we strongly recommend users manually verify all API uses.
Developers should apply responsible AI best practices and are responsible for ensuring that a specific use-case complies with relevant laws and regulations (e.g. privacy, trade, etc.). Important areas for consideration include:
* Allocation: The model may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
* High-Risk Scenarios: Developers should assess suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
* Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
* Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
* Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.
* Copyrighted content: The model might generate content that infringes on copyright protections. Developers should implement measures to detect and filter copyrighted material, and end-users should be informed about the potential for unintended copyright violations and the importance of verifying original sources to avoid legal complications.
* Election Misinformation: Developers should ensure robust verification mechanisms are in place to detect and correct false information regarding elections and should inform users of the need for critical evaluation of AI-generated election-related content to mitigate the spread of misinformation.
## License
The model is licensed under the [MIT license](./LICENSE).
## Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow [Microsoft’s Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks). Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.
| [
"MEDQA"
] |
moussaKam/barthez-orangesum-abstract | moussaKam | summarization | [
"transformers",
"pytorch",
"mbart",
"text2text-generation",
"summarization",
"bart",
"fr",
"arxiv:2010.12321",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2022-03-02T23:29:05Z" | 2021-11-15T13:03:03+00:00 | 540 | 7 | ---
language:
- fr
license: apache-2.0
tags:
- summarization
- bart
widget:
- text: Citant les préoccupations de ses clients dénonçant des cas de censure après
la suppression du compte de Trump, un fournisseur d'accès Internet de l'État de
l'Idaho a décidé de bloquer Facebook et Twitter. La mesure ne concernera cependant
que les clients mécontents de la politique de ces réseaux sociaux.
---
### Barthez model finetuned on orangeSum (abstract generation)
finetuning: examples/seq2seq (as of Feb 08 2021)
paper: https://arxiv.org/abs/2010.12321 \
github: https://github.com/moussaKam/BARThez
```
@article{eddine2020barthez,
title={BARThez: a Skilled Pretrained French Sequence-to-Sequence Model},
author={Eddine, Moussa Kamal and Tixier, Antoine J-P and Vazirgiannis, Michalis},
journal={arXiv preprint arXiv:2010.12321},
year={2020}
}
```
| [
"CAS"
] |
sultan/BioM-ALBERT-xxlarge-PMC | sultan | fill-mask | [
"transformers",
"pytorch",
"albert",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2022-03-02T23:29:05Z" | 2023-11-04T23:06:21+00:00 | 534 | 4 | ---
{}
---
# BioM-Transformers: Building Large Biomedical Language Models with BERT, ALBERT and ELECTRA
# Abstract
The impact of design choices on the performance
of biomedical language models recently
has been a subject for investigation. In
this paper, we empirically study biomedical
domain adaptation with large transformer models
using different design choices. We evaluate
the performance of our pretrained models
against other existing biomedical language
models in the literature. Our results show that
we achieve state-of-the-art results on several
biomedical domain tasks despite using similar
or less computational cost compared to other
models in the literature. Our findings highlight
the significant effect of design choices on
improving the performance of biomedical language
models.
# Model Description
This model was pre-trained on PMC full article for further 64k steps with a batch size of 8192, where we initiate our weights from our model BioM-ALBERT-xxlarge. Thus, the total training steps for this model is 264k+64K=328K steps. The model is very large due to the number of hidden layer size (4096). In order to help researchers with limited resources to fine-tune larger models, we created an example with PyTorch XLA. PyTorch XLA (https://github.com/pytorch/xla) is a library that allows you to use PyTorch on TPU units, which is provided for free by Google Colab and Kaggle. Follow this example to work with PyTorch/XLA [Link](https://github.com/salrowili/BioM-Transformers/blob/main/examples/Fine_Tuning_Biomedical_Models_on_Text_Classification_Task_With_HuggingFace_Transformers_and_PyTorch_XLA.ipynb). In this example we achieve 80.74 micro F1 score on ChemProt task with BioM-ALBERTxxlarge . Fine-tuning takes 43 minutes for 5 epochs .
Check our GitHub repo at https://github.com/salrowili/BioM-Transformers for TensorFlow and GluonNLP checkpoints. We also updated this repo with a couple of examples on how to fine-tune LMs on text classification and questions answering tasks such as ChemProt, SQuAD, and BioASQ.
# Colab Notebook Examples
BioM-ELECTRA-LARGE on NER and ChemProt Task [![Open In Colab][COLAB]](https://colab.research.google.com/github/salrowili/BioM-Transformers/blob/main/examples/Example_of_NER_and_ChemProt_Task_on_TPU.ipynb)
BioM-ELECTRA-Large on SQuAD2.0 and BioASQ7B Factoid tasks [![Open In Colab][COLAB]](https://colab.research.google.com/github/salrowili/BioM-Transformers/blob/main/examples/Example_of_SQuAD2_0_and_BioASQ7B_tasks_with_BioM_ELECTRA_Large_on_TPU.ipynb)
BioM-ALBERT-xxlarge on SQuAD2.0 and BioASQ7B Factoid tasks [![Open In Colab][COLAB]](https://colab.research.google.com/github/salrowili/BioM-Transformers/blob/main/examples/Example_of_SQuAD2_0_and_BioASQ7B_tasks_with_BioM_ALBERT_xxlarge_on_TPU.ipynb)
Text Classification Task With HuggingFace Transformers and PyTorchXLA on Free TPU [![Open In Colab][COLAB]](https://colab.research.google.com/github/salrowili/BioM-Transformers/blob/main/examples/Fine_Tuning_Biomedical_Models_on_Text_Classification_Task_With_HuggingFace_Transformers_and_PyTorch_XLA.ipynb)
Reproducing our BLURB results with JAX [![Open In Colab][COLAB]](https://colab.research.google.com/github/salrowili/BioM-Transformers/blob/main/examples/BLURB_LeaderBoard_with_TPU_VM.ipynb)
Finetunning BioM-Transformers with Jax/Flax on TPUv3-8 with free Kaggle resource [![Open In Colab][COLAB]](https://www.kaggle.com/code/sultanalrowili/biom-transoformers-with-flax-on-tpu-with-kaggle)
[COLAB]: https://colab.research.google.com/assets/colab-badge.svg
# Acknowledgment
We would like to acknowledge the support we have from Tensorflow Research Cloud (TFRC) team to grant us access to TPUv3 units.
# Citation
```bibtex
@inproceedings{alrowili-shanker-2021-biom,
title = "{B}io{M}-Transformers: Building Large Biomedical Language Models with {BERT}, {ALBERT} and {ELECTRA}",
author = "Alrowili, Sultan and
Shanker, Vijay",
booktitle = "Proceedings of the 20th Workshop on Biomedical Language Processing",
month = jun,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2021.bionlp-1.24",
pages = "221--227",
abstract = "The impact of design choices on the performance of biomedical language models recently has been a subject for investigation. In this paper, we empirically study biomedical domain adaptation with large transformer models using different design choices. We evaluate the performance of our pretrained models against other existing biomedical language models in the literature. Our results show that we achieve state-of-the-art results on several biomedical domain tasks despite using similar or less computational cost compared to other models in the literature. Our findings highlight the significant effect of design choices on improving the performance of biomedical language models.",
}
``` | [
"BLURB",
"CHEMPROT"
] |
mradermacher/Llama3-Aloe-8B-Alpha-GGUF | mradermacher | null | [
"transformers",
"gguf",
"biology",
"medical",
"en",
"dataset:argilla/dpo-mix-7k",
"dataset:nvidia/HelpSteer",
"dataset:jondurbin/airoboros-3.2",
"dataset:hkust-nlp/deita-10k-v0",
"dataset:LDJnr/Capybara",
"dataset:HPAI-BSC/CareQA",
"dataset:GBaker/MedQA-USMLE-4-options",
"dataset:lukaemon/mmlu",
"dataset:bigbio/pubmed_qa",
"dataset:openlifescienceai/medmcqa",
"dataset:bigbio/med_qa",
"dataset:HPAI-BSC/better-safe-than-sorry",
"dataset:HPAI-BSC/pubmedqa-cot",
"dataset:HPAI-BSC/medmcqa-cot",
"dataset:HPAI-BSC/medqa-cot",
"base_model:HPAI-BSC/Llama3-Aloe-8B-Alpha",
"base_model:quantized:HPAI-BSC/Llama3-Aloe-8B-Alpha",
"license:cc-by-nc-4.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-04-26T20:49:47Z" | 2024-05-14T19:17:19+00:00 | 533 | 2 | ---
base_model: HPAI-BSC/Llama3-Aloe-8B-Alpha
datasets:
- argilla/dpo-mix-7k
- nvidia/HelpSteer
- jondurbin/airoboros-3.2
- hkust-nlp/deita-10k-v0
- LDJnr/Capybara
- HPAI-BSC/CareQA
- GBaker/MedQA-USMLE-4-options
- lukaemon/mmlu
- bigbio/pubmed_qa
- openlifescienceai/medmcqa
- bigbio/med_qa
- HPAI-BSC/better-safe-than-sorry
- HPAI-BSC/pubmedqa-cot
- HPAI-BSC/medmcqa-cot
- HPAI-BSC/medqa-cot
language:
- en
library_name: transformers
license: cc-by-nc-4.0
tags:
- biology
- medical
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 1 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: -->
<!-- ### vocab_type: -->
static quants of https://huggingface.co/HPAI-BSC/Llama3-Aloe-8B-Alpha
<!-- provided-files -->
weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion.
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q2_K.gguf) | Q2_K | 3.3 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.IQ3_XS.gguf) | IQ3_XS | 3.6 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q3_K_S.gguf) | Q3_K_S | 3.8 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.IQ3_S.gguf) | IQ3_S | 3.8 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.IQ3_M.gguf) | IQ3_M | 3.9 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q3_K_M.gguf) | Q3_K_M | 4.1 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q3_K_L.gguf) | Q3_K_L | 4.4 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.IQ4_XS.gguf) | IQ4_XS | 4.6 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q4_K_S.gguf) | Q4_K_S | 4.8 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q4_K_M.gguf) | Q4_K_M | 5.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q5_K_S.gguf) | Q5_K_S | 5.7 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q5_K_M.gguf) | Q5_K_M | 5.8 | |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q6_K.gguf) | Q6_K | 6.7 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.Q8_0.gguf) | Q8_0 | 8.6 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Llama3-Aloe-8B-Alpha-GGUF/resolve/main/Llama3-Aloe-8B-Alpha.f16.gguf) | f16 | 16.2 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
| [
"MEDQA",
"PUBMEDQA"
] |
sinequa/vectorizer.vanilla | sinequa | sentence-similarity | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"sentence-similarity",
"en",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2023-07-11T07:31:15Z" | 2024-02-19T09:40:26+00:00 | 529 | 0 | ---
language:
- en
pipeline_tag: sentence-similarity
tags:
- feature-extraction
- sentence-similarity
---
# Model Card for `vectorizer.vanilla`
This model is a vectorizer developed by Sinequa. It produces an embedding vector given a passage or a query. The passage vectors are stored in our vector index and the query vector is used at query time to look up relevant passages in the index.
Model name: `vectorizer.vanilla`
## Supported Languages
The model was trained and tested in the following languages:
- English
## Scores
| Metric | Value |
|:-----------------------|------:|
| Relevance (Recall@100) | 0.639 |
Note that the relevance score is computed as an average over 14 retrieval datasets (see
[details below](#evaluation-metrics)).
## Inference Times
| GPU | Quantization type | Batch size 1 | Batch size 32 |
|:------------------------------------------|:------------------|---------------:|---------------:|
| NVIDIA A10 | FP16 | 1 ms | 5 ms |
| NVIDIA A10 | FP32 | 2 ms | 20 ms |
| NVIDIA T4 | FP16 | 1 ms | 14 ms |
| NVIDIA T4 | FP32 | 2 ms | 53 ms |
| NVIDIA L4 | FP16 | 1 ms | 5 ms |
| NVIDIA L4 | FP32 | 3 ms | 25 ms |
## Gpu Memory usage
| Quantization type | Memory |
|:-------------------------------------------------|-----------:|
| FP16 | 300 MiB |
| FP32 | 500 MiB |
Note that GPU memory usage only includes how much GPU memory the actual model consumes on an NVIDIA T4 GPU with a batch
size of 32. It does not include the fix amount of memory that is consumed by the ONNX Runtime upon initialization which
can be around 0.5 to 1 GiB depending on the used GPU.
## Requirements
- Minimal Sinequa version: 11.10.0
- Minimal Sinequa version for using FP16 models and GPUs with CUDA compute capability of 8.9+ (like NVIDIA L4): 11.11.0
- [Cuda compute capability](https://developer.nvidia.com/cuda-gpus): above 5.0 (above 6.0 for FP16 use)
## Model Details
### Overview
- Number of parameters: 23 million
- Base language model: [English MiniLM-L6-H384](https://huggingface.co/nreimers/MiniLM-L6-H384-uncased)
- Insensitive to casing and accents
- Output dimensions: 256 (reduced with an additional dense layer)
- Training procedure: Query-passage-negative triplets for datasets that have mined hard negative data, Query-passage pairs for the rest. Number of negatives is augmented with in-batch negative strategy.
### Training Data
The model have been trained using all datasets that are cited in the [all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) model.
### Evaluation Metrics
To determine the relevance score, we averaged the results that we obtained when evaluating on the datasets of the
[BEIR benchmark](https://github.com/beir-cellar/beir). Note that all these datasets are in English.
| Dataset | Recall@100 |
|:------------------|-----------:|
| Average | 0.639 |
| | |
| Arguana | 0.969 |
| CLIMATE-FEVER | 0.509 |
| DBPedia Entity | 0.409 |
| FEVER | 0.839 |
| FiQA-2018 | 0.702 |
| HotpotQA | 0.609 |
| MS MARCO | 0.849 |
| NFCorpus | 0.315 |
| NQ | 0.786 |
| Quora | 0.995 |
| SCIDOCS | 0.497 |
| SciFact | 0.911 |
| TREC-COVID | 0.129 |
| Webis-Touche-2020 | 0.427 |
| [
"SCIFACT"
] |
RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2404.00376",
"arxiv:2009.13081",
"arxiv:2402.18060",
"arxiv:2203.14371",
"arxiv:2009.03300",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-08-22T11:56:52Z" | 2024-08-22T13:47:43+00:00 | 526 | 0 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
llama-3-meerkat-8b-v1.0 - GGUF
- Model creator: https://huggingface.co/dmis-lab/
- Original model: https://huggingface.co/dmis-lab/llama-3-meerkat-8b-v1.0/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [llama-3-meerkat-8b-v1.0.Q2_K.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q2_K.gguf) | Q2_K | 2.96GB |
| [llama-3-meerkat-8b-v1.0.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.IQ3_XS.gguf) | IQ3_XS | 3.28GB |
| [llama-3-meerkat-8b-v1.0.IQ3_S.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.IQ3_S.gguf) | IQ3_S | 3.43GB |
| [llama-3-meerkat-8b-v1.0.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q3_K_S.gguf) | Q3_K_S | 3.41GB |
| [llama-3-meerkat-8b-v1.0.IQ3_M.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.IQ3_M.gguf) | IQ3_M | 3.52GB |
| [llama-3-meerkat-8b-v1.0.Q3_K.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q3_K.gguf) | Q3_K | 3.74GB |
| [llama-3-meerkat-8b-v1.0.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q3_K_M.gguf) | Q3_K_M | 3.74GB |
| [llama-3-meerkat-8b-v1.0.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q3_K_L.gguf) | Q3_K_L | 4.03GB |
| [llama-3-meerkat-8b-v1.0.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.IQ4_XS.gguf) | IQ4_XS | 4.18GB |
| [llama-3-meerkat-8b-v1.0.Q4_0.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q4_0.gguf) | Q4_0 | 4.34GB |
| [llama-3-meerkat-8b-v1.0.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.IQ4_NL.gguf) | IQ4_NL | 4.38GB |
| [llama-3-meerkat-8b-v1.0.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q4_K_S.gguf) | Q4_K_S | 4.37GB |
| [llama-3-meerkat-8b-v1.0.Q4_K.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q4_K.gguf) | Q4_K | 4.58GB |
| [llama-3-meerkat-8b-v1.0.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q4_K_M.gguf) | Q4_K_M | 4.58GB |
| [llama-3-meerkat-8b-v1.0.Q4_1.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q4_1.gguf) | Q4_1 | 4.78GB |
| [llama-3-meerkat-8b-v1.0.Q5_0.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q5_0.gguf) | Q5_0 | 5.21GB |
| [llama-3-meerkat-8b-v1.0.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q5_K_S.gguf) | Q5_K_S | 5.21GB |
| [llama-3-meerkat-8b-v1.0.Q5_K.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q5_K.gguf) | Q5_K | 5.34GB |
| [llama-3-meerkat-8b-v1.0.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q5_K_M.gguf) | Q5_K_M | 5.34GB |
| [llama-3-meerkat-8b-v1.0.Q5_1.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q5_1.gguf) | Q5_1 | 5.65GB |
| [llama-3-meerkat-8b-v1.0.Q6_K.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q6_K.gguf) | Q6_K | 6.14GB |
| [llama-3-meerkat-8b-v1.0.Q8_0.gguf](https://huggingface.co/RichardErkhov/dmis-lab_-_llama-3-meerkat-8b-v1.0-gguf/blob/main/llama-3-meerkat-8b-v1.0.Q8_0.gguf) | Q8_0 | 7.95GB |
Original model description:
---
license: cc-by-nc-4.0
pipeline_tag: text-generation
tags:
- medical
- small LM
- instruction-tuned
- usmle
- synthetic data
---
# Meerkat-8B (Version 1.0)
🚀 Meerkat-8B is a new instruction-tuned medical AI system of the Meerkat model family.
The model was based on the Meta's Llama-3-8B-Instruct model and fine-tuned using our new synthetic dataset consisting of high-quality chain-of-thought reasoning paths sourced from 18 medical textbooks, along with diverse instruction-following datasets.
This equips the model with high-level medical reasoning capabilities required for solving complex medical problems.
For further insights into our model, please refer to our paper!
📄 **Paper**: [Small Language Models Learn Enhanced Reasoning Skills from Medical Textbooks](https://arxiv.org/abs/2404.00376)
## Quick Start
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "dmis-lab/llama-3-meerkat-8b-v1.0"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16, # You can choose to use this when there's not enough GPU memory available.
device_map="auto",
)
# Multi-turn dialogue example
messages =[
{"role": "system", "content": "You are a helpful doctor or healthcare professional. Guide the conversation to provide useful, complete, and scientifically-grounded answers to user questions. You have the option to compose a concise, single-turn conversation if the user's input is comprehensive to provide accurate answers. However, if essential details are missing, you should engage in a multi-turn dialogue, asking follow-up questions to gather a thorough medical history and records.\n\n"},
{"role": "user", "content": "Hello, doctor. I'm really concerned about my 10-year-old son. We recently discovered a painless mass in his left testicle, so we brought him to the pediatrician."},
{"role": "assistant", "content": "I understand your concern. Let's gather some more information. Has your son experienced any other symptoms along with the mass?"},
{"role": "user", "content": "Other than the mass, my son hasn't shown any symptoms. He's been his usual self, playing and eating normally."}
]
input_ids = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
return_tensors="pt"
).to(model.device)
terminators = [
tokenizer.eos_token_id,
tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = model.generate(
input_ids,
max_new_tokens=1000,
eos_token_id=terminators,
do_sample=True,
temperature=0.7,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
```
## Prompt Details
To reproduce the results reported in our paper, it is advisable to utilize the identical system messages used during model training. Please refer to the guidelines detailed below.
### USMLE
When solving USMLE-style questions such as [MedQA](https://arxiv.org/abs/2009.13081) and [MedBullets](https://arxiv.org/abs/2402.18060), use the following system message:
```
messages = [
{"role": "system", "content": "The following is a multiple-choice question about medical knowledge. Solve this in a step-by-step fashion, starting by summarizing the available information. Output a single option from the given options as the final answer. You are strongly required to follow the specified output format; conclude your response with the phrase \"the answer is ([option_id]) [answer_string]\".\n\n"},
{"role": "user", "content": "Two weeks after undergoing an emergency cardiac catherization with stenting for unstable angina pectoris, a 61-year-old man has decreased urinary output and malaise. He has type 2 diabetes mellitus and osteoarthritis of the hips. Prior to admission, his medications were insulin and naproxen. He was also started on aspirin, clopidogrel, and metoprolol after the coronary intervention. His temperature is 38\u00b0C (100.4\u00b0F), pulse is 93/min, and blood pressure is 125/85 mm Hg. Examination shows mottled, reticulated purplish discoloration of the feet. Laboratory studies show:\nHemoglobin count 14 g/dL\nLeukocyte count 16,400/mm3\nSegmented neutrophils 56%\nEosinophils 11%\nLymphocytes 31%\nMonocytes 2%\nPlatelet count 260,000/mm3\nErythrocyte sedimentation rate 68 mm/h\nSerum\nUrea nitrogen 25 mg/dL\nCreatinine 4.2 mg/dL\nRenal biopsy shows intravascular spindle-shaped vacuoles. Which of the following is the most likely cause of this patient's symptoms?\" (A) Renal papillary necrosis (B) Cholesterol embolization (C) Eosinophilic granulomatosis with polyangiitis (D) Polyarteritis nodosa"},
]
```
The model generates reasoning paths to solve the problem and then sequentially provides the predicted answers.
Since the model ends its response with "the answer is," it is straightforward to extract the predicted answer for comparison with the actual answer.
### Multiple-choice Exams
For other types of multiple-choice exams such as [MedMCQA](https://arxiv.org/abs/2203.14371) or [MMLU](https://arxiv.org/abs/2009.03300), use the following simple system message:
```
messages = [
{"role": "system", "content": "Answer the multiple-choice question about medical knowledge.\n\n"},
{"role": "user", "content": "In a Robertsonian translocation fusion occurs at the: (A) telomeres. (B) centromeres. (C) histones. (D) ends of the long arms."},
]
```
### Other Use Cases
Our model was trained using the [AlpaCare](https://github.com/xzhang97666/alpacare) instruction dataset comprising 52K examples, to enhance its generalization capabilities across diverse user prompts.
Feel free to design and test your prompts and to share your thoughts with us, whether the model exceeds expectations or falls short!
## Reproducing MedQA Performance with vLLM
Here is an example code for fast model evaluation in MedQA using vLLM. To adapt this code for other datasets like MedMCQA or MMLU, simply modify the instructions and update the dataset paths as needed.
```python
import re
from datasets import load_dataset
from vllm import LLM, SamplingParams
USMLE_INSTRUCTION = (
"The following is a multiple-choice question about medical knowledge. Solve this in"
" a step-by-step fashion, starting by summarizing the available information. Output"
" a single option from the given options as the final answer. You are strongly"
" required to follow the specified output format; conclude your response with the"
' phrase "the answer is ([option_id]) [answer_string]".\n\n'
)
llm = LLM(
model="dmis-lab/llama-3-meerkat-8b-v1.0",
dtype="bfloat16",
gpu_memory_utilization=0.9,
max_model_len=2048,
trust_remote_code=True,
tensor_parallel_size=1
)
tokenizer = llm.get_tokenizer()
inputs, labels = [], []
for sample in load_dataset(
"GBaker/MedQA-USMLE-4-options", split="test", trust_remote_code=True
):
options = sorted(sample["options"].items())
options = " ".join(map(lambda x: f"({x[0]}) {x[1]}", options))
content = tokenizer.apply_chat_template(
[{"role": "system", "content": USMLE_INSTRUCTION}, {"role": "user", "content": sample["question"] + " " + options}],
add_generation_prompt=True,
tokenize=False,
)
inputs.append(content)
labels.append(sample["answer_idx"])
generated = llm.generate(
inputs,
SamplingParams(
temperature=0.0,
stop_token_ids=[tokenizer.vocab["<|eot_id|>"]],
max_tokens=1024,
),
)
def extract_answer(text: str, options: str = "ABCD") -> str:
return (re.findall(rf"he answer is \(([{options}])\)", text) or [options[0]])[-1]
correctness = []
for g, l in zip(generated, labels):
correctness.append(extract_answer(g.outputs[0].text) == l)
print(sum(correctness) / len(correctness))
```
## Evaluation
We tested models on seven medical benchmarks: [MedQA](https://arxiv.org/abs/2009.13081), [USMLE sample test](https://www.usmle.org/prepare-your-exam), [Medbullets-4](https://arxiv.org/abs/2402.18060), [Medbullets-5](https://arxiv.org/abs/2402.18060) , [MedMCQA](https://arxiv.org/abs/2203.14371), [MMLU-Medical](https://arxiv.org/abs/2009.03300), and [JAMA Clinical Challenge](https://arxiv.org/abs/2402.18060).
| **Model** | **Average** | **MedQA** | **USMLE** | **Medbullets-4** | **Medbullets-5** | **MedMCQA** | **MMLU-Medical** |
|:--------------------------------|:-----------:|:---------:|:---------:|:----------------:|:----------------:|:-----------:|:----------------:|
| GPT-4 | 76.6 | 81.4 | 86.6 | 68.8 | 63.3 | 72.4 | 87.1 |
| GPT-3.5 | 54.8 | 53.6 | 58.5 | 51.0 | 47.4 | 51.0 | 67.3 |
| MediTron-70B (Ensemble, 5 runs) | - | 70.2 | - | - | - | 66.0 | 78.0 |
|*Open-source (7B)*|
| MediTron-7B | 51.0 | 50.2 | 44.6 | 51.1 | 45.5 | 57.9 | 56.7 |
| BioMistral-7B | 55.4 | 54.3 | 51.4 | 52.3 | 48.7 | 61.1 | 64.6 |
| Meerkat-7B | 62.6 | 70.6 | 70.3 | 58.7 | 52.9 | 60.6 | 70.5 |
| Meerkat-8B (**New**) | **67.3** | **74.0** | **74.2** | **62.3** | **55.5** | **62.7** | **75.2** |
Please note that the scores in MMLU-Medical were calculated based on the average accuracies across six medical-related subjects in the original MMLU benchmark, and each result for a single subject is presented below.
| **Model** | **Average** | **Cliniq Knowledge** | **Medical Genetics** | **Anatomy** | **Professional Medicine** | **College Biology** | **College Medicine** |
|:--------------------------------|:-----------:|:--------------------:|:--------------------:|:-----------:|:-------------------------:|:-------------------:|:--------------------:|
| GPT-4 | 87.1 | 86.4 | 92.0 | 80.0 | 93.8 | 93.8 | 76.3 |
| GPT-3.5 | 67.3 | 68.7 | 68.0 | 60.7 | 69.9 | 72.9 | 63.6 |
| MediTron-70B (Ensemble, 5 runs) | 78.0 | 75.5 | 85.9 | 69.4 | 82.3 | 86.7 | 68.0 |
|*Open-source (7B)*|
| MediTron-7B | 56.7 | 57.7 | 63.8 | 56.9 | 56.0 | 57.1 | 48.9 |
| BioMistral-7B | 64.6 | 59.9 | 64.0 | 56.5 | 60.4 | 59.0 | 54.7 |
| Meerkat-7B | 70.5 | 71.6 | 74.8 | 63.2 | **77.3** | 70.8 | 65.2 |
| Meerkat-8B (**New**) | **75.2** | **74.3** | **76.7** | **74.8** | 75.3 | **76.1** | **74.3** |
## Reference
Please see the information below to cite our paper.
```bibtex
@article{kim2024small,
title={Small language models learn enhanced reasoning skills from medical textbooks},
author={Kim, Hyunjae and Hwang, Hyeon and Lee, Jiwoo and Park, Sihyeon and Kim, Dain and Lee, Taewhoo and Yoon, Chanwoong and Sohn, Jiwoong and Choi, Donghee and Kang, Jaewoo},
journal={arXiv preprint arXiv:2404.00376},
year={2024}
}
```
## Acknowledgement
Research supported with Cloud TPUs from Google’s TPU Research Cloud (TRC).
## Contact
Feel free to email `hyunjae-kim@korea.ac.kr` if you have any questions.
| [
"MEDQA"
] |
bartowski/granite-embedding-30m-english-GGUF | bartowski | text-generation | [
"gguf",
"language",
"granite",
"embeddings",
"text-generation",
"en",
"base_model:ibm-granite/granite-embedding-30m-english",
"base_model:quantized:ibm-granite/granite-embedding-30m-english",
"license:apache-2.0",
"model-index",
"endpoints_compatible",
"region:us",
"feature-extraction"
] | "2024-12-18T14:54:02Z" | 2024-12-23T04:56:43+00:00 | 526 | 0 | ---
base_model: ibm-granite/granite-embedding-30m-english
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- language
- granite
- embeddings
quantized_by: bartowski
model-index:
- name: ibm-granite/granite-embedding-30m-english
results:
- task:
type: Retrieval
dataset:
name: MTEB ArguaAna
type: mteb/arguana
config: default
split: test
metrics:
- type: map_at_1
value: 0.31792
- type: map_at_10
value: 0.47599
- type: map_at_100
value: 0.48425
- type: map_at_1000
value: 0.48427
- type: map_at_3
value: 0.42757
- type: map_at_5
value: 0.45634
- type: mrr_at_1
value: 0.32788
- type: mrr_at_10
value: 0.47974
- type: mrr_at_100
value: 0.48801
- type: mrr_at_1000
value: 0.48802
- type: mrr_at_3
value: 0.43065
- type: mrr_at_5
value: 0.45999
- type: ndcg_at_1
value: 0.31792
- type: ndcg_at_10
value: 0.56356
- type: ndcg_at_100
value: 0.59789
- type: ndcg_at_1000
value: 0.59857
- type: ndcg_at_3
value: 0.46453
- type: ndcg_at_5
value: 0.51623
- type: precision_at_1
value: 0.31792
- type: precision_at_10
value: 0.08428
- type: precision_at_100
value: 0.00991
- type: precision_at_1000
value: 0.001
- type: precision_at_3
value: 0.19061
- type: precision_at_5
value: 0.1394
- type: recall_at_1
value: 0.31792
- type: recall_at_10
value: 0.84282
- type: recall_at_100
value: 0.99075
- type: recall_at_1000
value: 0.99644
- type: recall_at_3
value: 0.57183
- type: recall_at_5
value: 0.69701
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER
type: mteb/climate-fever
config: default
split: test
metrics:
- type: map_at_1
value: 0.13189
- type: map_at_10
value: 0.21789
- type: map_at_100
value: 0.2358
- type: map_at_1000
value: 0.23772
- type: map_at_3
value: 0.18513
- type: map_at_5
value: 0.20212
- type: mrr_at_1
value: 0.29837
- type: mrr_at_10
value: 0.41376
- type: mrr_at_100
value: 0.42282
- type: mrr_at_1000
value: 0.42319
- type: mrr_at_3
value: 0.38284
- type: mrr_at_5
value: 0.40301
- type: ndcg_at_1
value: 0.29837
- type: ndcg_at_10
value: 0.30263
- type: ndcg_at_100
value: 0.37228
- type: ndcg_at_1000
value: 0.40677
- type: ndcg_at_3
value: 0.25392
- type: ndcg_at_5
value: 0.27153
- type: precision_at_1
value: 0.29837
- type: precision_at_10
value: 0.09179
- type: precision_at_100
value: 0.01659
- type: precision_at_1000
value: 0.0023
- type: precision_at_3
value: 0.18545
- type: precision_at_5
value: 0.14241
- type: recall_at_1
value: 0.13189
- type: recall_at_10
value: 0.35355
- type: recall_at_100
value: 0.59255
- type: recall_at_1000
value: 0.78637
- type: recall_at_3
value: 0.23255
- type: recall_at_5
value: 0.28446
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval
type: mteb/cqadupstack-android
config: default
split: test
metrics:
- type: map_at_1
value: 0.35797
- type: map_at_10
value: 0.47793
- type: map_at_100
value: 0.49422
- type: map_at_1000
value: 0.49546
- type: map_at_3
value: 0.44137
- type: map_at_5
value: 0.46063
- type: mrr_at_1
value: 0.44206
- type: mrr_at_10
value: 0.53808
- type: mrr_at_100
value: 0.5454
- type: mrr_at_1000
value: 0.54578
- type: mrr_at_3
value: 0.51431
- type: mrr_at_5
value: 0.5284
- type: ndcg_at_1
value: 0.44206
- type: ndcg_at_10
value: 0.54106
- type: ndcg_at_100
value: 0.59335
- type: ndcg_at_1000
value: 0.61015
- type: ndcg_at_3
value: 0.49365
- type: ndcg_at_5
value: 0.51429
- type: precision_at_1
value: 0.44206
- type: precision_at_10
value: 0.10443
- type: precision_at_100
value: 0.01631
- type: precision_at_1000
value: 0.00214
- type: precision_at_3
value: 0.23653
- type: precision_at_5
value: 0.1691
- type: recall_at_1
value: 0.35797
- type: recall_at_10
value: 0.65182
- type: recall_at_100
value: 0.86654
- type: recall_at_1000
value: 0.97131
- type: recall_at_3
value: 0.51224
- type: recall_at_5
value: 0.57219
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackEnglishRetrieval
type: mteb/cqadupstack-english
config: default
split: test
metrics:
- type: map_at_1
value: 0.32748
- type: map_at_10
value: 0.44138
- type: map_at_100
value: 0.45565
- type: map_at_1000
value: 0.45698
- type: map_at_3
value: 0.40916
- type: map_at_5
value: 0.42621
- type: mrr_at_1
value: 0.41274
- type: mrr_at_10
value: 0.5046
- type: mrr_at_100
value: 0.5107
- type: mrr_at_1000
value: 0.51109
- type: mrr_at_3
value: 0.48238
- type: mrr_at_5
value: 0.49563
- type: ndcg_at_1
value: 0.41274
- type: ndcg_at_10
value: 0.50251
- type: ndcg_at_100
value: 0.54725
- type: ndcg_at_1000
value: 0.56635
- type: ndcg_at_3
value: 0.46023
- type: ndcg_at_5
value: 0.47883
- type: precision_at_1
value: 0.41274
- type: precision_at_10
value: 0.09828
- type: precision_at_100
value: 0.01573
- type: precision_at_1000
value: 0.00202
- type: precision_at_3
value: 0.22718
- type: precision_at_5
value: 0.16064
- type: recall_at_1
value: 0.32748
- type: recall_at_10
value: 0.60322
- type: recall_at_100
value: 0.79669
- type: recall_at_1000
value: 0.9173
- type: recall_at_3
value: 0.47523
- type: recall_at_5
value: 0.52957
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGamingRetrieval
type: mteb/cqadupstack-gaming
config: default
split: test
metrics:
- type: map_at_1
value: 0.41126
- type: map_at_10
value: 0.53661
- type: map_at_100
value: 0.54588
- type: map_at_1000
value: 0.54638
- type: map_at_3
value: 0.50389
- type: map_at_5
value: 0.52286
- type: mrr_at_1
value: 0.47147
- type: mrr_at_10
value: 0.5685
- type: mrr_at_100
value: 0.57458
- type: mrr_at_1000
value: 0.57487
- type: mrr_at_3
value: 0.54431
- type: mrr_at_5
value: 0.55957
- type: ndcg_at_1
value: 0.47147
- type: ndcg_at_10
value: 0.59318
- type: ndcg_at_100
value: 0.62972
- type: ndcg_at_1000
value: 0.64033
- type: ndcg_at_3
value: 0.53969
- type: ndcg_at_5
value: 0.56743
- type: precision_at_1
value: 0.47147
- type: precision_at_10
value: 0.09549
- type: precision_at_100
value: 0.01224
- type: precision_at_1000
value: 0.00135
- type: precision_at_3
value: 0.24159
- type: precision_at_5
value: 0.16577
- type: recall_at_1
value: 0.41126
- type: recall_at_10
value: 0.72691
- type: recall_at_100
value: 0.88692
- type: recall_at_1000
value: 0.96232
- type: recall_at_3
value: 0.58374
- type: recall_at_5
value: 0.65226
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGisRetrieval
type: mteb/cqadupstack-gis
config: default
split: test
metrics:
- type: map_at_1
value: 0.28464
- type: map_at_10
value: 0.3828
- type: map_at_100
value: 0.39277
- type: map_at_1000
value: 0.39355
- type: map_at_3
value: 0.35704
- type: map_at_5
value: 0.37116
- type: mrr_at_1
value: 0.30734
- type: mrr_at_10
value: 0.40422
- type: mrr_at_100
value: 0.41297
- type: mrr_at_1000
value: 0.41355
- type: mrr_at_3
value: 0.38136
- type: mrr_at_5
value: 0.39362
- type: ndcg_at_1
value: 0.30734
- type: ndcg_at_10
value: 0.43564
- type: ndcg_at_100
value: 0.48419
- type: ndcg_at_1000
value: 0.50404
- type: ndcg_at_3
value: 0.38672
- type: ndcg_at_5
value: 0.40954
- type: precision_at_1
value: 0.30734
- type: precision_at_10
value: 0.06633
- type: precision_at_100
value: 0.00956
- type: precision_at_1000
value: 0.00116
- type: precision_at_3
value: 0.16497
- type: precision_at_5
value: 0.11254
- type: recall_at_1
value: 0.28464
- type: recall_at_10
value: 0.57621
- type: recall_at_100
value: 0.7966
- type: recall_at_1000
value: 0.94633
- type: recall_at_3
value: 0.44588
- type: recall_at_5
value: 0.50031
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackMathematicaRetrieval
type: mteb/cqadupstack-mathematica
config: default
split: test
metrics:
- type: map_at_1
value: 0.18119
- type: map_at_10
value: 0.27055
- type: map_at_100
value: 0.28461
- type: map_at_1000
value: 0.28577
- type: map_at_3
value: 0.24341
- type: map_at_5
value: 0.25861
- type: mrr_at_1
value: 0.22886
- type: mrr_at_10
value: 0.32234
- type: mrr_at_100
value: 0.3328
- type: mrr_at_1000
value: 0.3334
- type: mrr_at_3
value: 0.29664
- type: mrr_at_5
value: 0.31107
- type: ndcg_at_1
value: 0.22886
- type: ndcg_at_10
value: 0.32749
- type: ndcg_at_100
value: 0.39095
- type: ndcg_at_1000
value: 0.41656
- type: ndcg_at_3
value: 0.27864
- type: ndcg_at_5
value: 0.30177
- type: precision_at_1
value: 0.22886
- type: precision_at_10
value: 0.06169
- type: precision_at_100
value: 0.0107
- type: precision_at_1000
value: 0.00143
- type: precision_at_3
value: 0.13682
- type: precision_at_5
value: 0.0995
- type: recall_at_1
value: 0.18119
- type: recall_at_10
value: 0.44983
- type: recall_at_100
value: 0.72396
- type: recall_at_1000
value: 0.90223
- type: recall_at_3
value: 0.31633
- type: recall_at_5
value: 0.37532
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackPhysicsRetrieval
type: mteb/cqadupstack-physics
config: default
split: test
metrics:
- type: map_at_1
value: 0.30517
- type: map_at_10
value: 0.42031
- type: map_at_100
value: 0.43415
- type: map_at_1000
value: 0.43525
- type: map_at_3
value: 0.38443
- type: map_at_5
value: 0.40685
- type: mrr_at_1
value: 0.38114
- type: mrr_at_10
value: 0.47783
- type: mrr_at_100
value: 0.48647
- type: mrr_at_1000
value: 0.48688
- type: mrr_at_3
value: 0.45172
- type: mrr_at_5
value: 0.46817
- type: ndcg_at_1
value: 0.38114
- type: ndcg_at_10
value: 0.4834
- type: ndcg_at_100
value: 0.53861
- type: ndcg_at_1000
value: 0.55701
- type: ndcg_at_3
value: 0.42986
- type: ndcg_at_5
value: 0.45893
- type: precision_at_1
value: 0.38114
- type: precision_at_10
value: 0.08893
- type: precision_at_100
value: 0.01375
- type: precision_at_1000
value: 0.00172
- type: precision_at_3
value: 0.20821
- type: precision_at_5
value: 0.15034
- type: recall_at_1
value: 0.30517
- type: recall_at_10
value: 0.61332
- type: recall_at_100
value: 0.84051
- type: recall_at_1000
value: 0.95826
- type: recall_at_3
value: 0.46015
- type: recall_at_5
value: 0.53801
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackProgrammersRetrieval
type: mteb/cqadupstack-programmers
config: default
split: test
metrics:
- type: map_at_1
value: 0.27396
- type: map_at_10
value: 0.38043
- type: map_at_100
value: 0.39341
- type: map_at_1000
value: 0.39454
- type: map_at_3
value: 0.34783
- type: map_at_5
value: 0.3663
- type: mrr_at_1
value: 0.34247
- type: mrr_at_10
value: 0.43681
- type: mrr_at_100
value: 0.4451
- type: mrr_at_1000
value: 0.44569
- type: mrr_at_3
value: 0.41172
- type: mrr_at_5
value: 0.42702
- type: ndcg_at_1
value: 0.34247
- type: ndcg_at_10
value: 0.44065
- type: ndcg_at_100
value: 0.49434
- type: ndcg_at_1000
value: 0.51682
- type: ndcg_at_3
value: 0.38976
- type: ndcg_at_5
value: 0.41332
- type: precision_at_1
value: 0.34247
- type: precision_at_10
value: 0.08059
- type: precision_at_100
value: 0.01258
- type: precision_at_1000
value: 0.00162
- type: precision_at_3
value: 0.1876
- type: precision_at_5
value: 0.13333
- type: recall_at_1
value: 0.27396
- type: recall_at_10
value: 0.56481
- type: recall_at_100
value: 0.79012
- type: recall_at_1000
value: 0.94182
- type: recall_at_3
value: 0.41785
- type: recall_at_5
value: 0.48303
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackStatsRetrieval
type: mteb/cqadupstack-stats
config: default
split: test
metrics:
- type: map_at_1
value: 0.25728
- type: map_at_10
value: 0.33903
- type: map_at_100
value: 0.34853
- type: map_at_1000
value: 0.34944
- type: map_at_3
value: 0.31268
- type: map_at_5
value: 0.32596
- type: mrr_at_1
value: 0.29141
- type: mrr_at_10
value: 0.36739
- type: mrr_at_100
value: 0.37545
- type: mrr_at_1000
value: 0.37608
- type: mrr_at_3
value: 0.34407
- type: mrr_at_5
value: 0.3568
- type: ndcg_at_1
value: 0.29141
- type: ndcg_at_10
value: 0.38596
- type: ndcg_at_100
value: 0.43375
- type: ndcg_at_1000
value: 0.45562
- type: ndcg_at_3
value: 0.33861
- type: ndcg_at_5
value: 0.35887
- type: precision_at_1
value: 0.29141
- type: precision_at_10
value: 0.06334
- type: precision_at_100
value: 0.00952
- type: precision_at_1000
value: 0.00121
- type: precision_at_3
value: 0.14826
- type: precision_at_5
value: 0.10429
- type: recall_at_1
value: 0.25728
- type: recall_at_10
value: 0.50121
- type: recall_at_100
value: 0.72382
- type: recall_at_1000
value: 0.88306
- type: recall_at_3
value: 0.36638
- type: recall_at_5
value: 0.41689
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackTexRetrieval
type: mteb/cqadupstack-tex
config: default
split: test
metrics:
- type: map_at_1
value: 0.19911
- type: map_at_10
value: 0.2856
- type: map_at_100
value: 0.29785
- type: map_at_1000
value: 0.29911
- type: map_at_3
value: 0.25875
- type: map_at_5
value: 0.2741
- type: mrr_at_1
value: 0.24054
- type: mrr_at_10
value: 0.32483
- type: mrr_at_100
value: 0.33464
- type: mrr_at_1000
value: 0.33534
- type: mrr_at_3
value: 0.30162
- type: mrr_at_5
value: 0.31506
- type: ndcg_at_1
value: 0.24054
- type: ndcg_at_10
value: 0.33723
- type: ndcg_at_100
value: 0.39362
- type: ndcg_at_1000
value: 0.42065
- type: ndcg_at_3
value: 0.29116
- type: ndcg_at_5
value: 0.31299
- type: precision_at_1
value: 0.24054
- type: precision_at_10
value: 0.06194
- type: precision_at_100
value: 0.01058
- type: precision_at_1000
value: 0.00148
- type: precision_at_3
value: 0.13914
- type: precision_at_5
value: 0.10076
- type: recall_at_1
value: 0.19911
- type: recall_at_10
value: 0.45183
- type: recall_at_100
value: 0.7025
- type: recall_at_1000
value: 0.89222
- type: recall_at_3
value: 0.32195
- type: recall_at_5
value: 0.37852
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackUnixRetrieval
type: mteb/cqadupstack-unix
config: default
split: test
metrics:
- type: map_at_1
value: 0.29819
- type: map_at_10
value: 0.40073
- type: map_at_100
value: 0.41289
- type: map_at_1000
value: 0.41375
- type: map_at_3
value: 0.36572
- type: map_at_5
value: 0.38386
- type: mrr_at_1
value: 0.35168
- type: mrr_at_10
value: 0.44381
- type: mrr_at_100
value: 0.45191
- type: mrr_at_1000
value: 0.45234
- type: mrr_at_3
value: 0.41402
- type: mrr_at_5
value: 0.43039
- type: ndcg_at_1
value: 0.35168
- type: ndcg_at_10
value: 0.46071
- type: ndcg_at_100
value: 0.51351
- type: ndcg_at_1000
value: 0.5317
- type: ndcg_at_3
value: 0.39972
- type: ndcg_at_5
value: 0.42586
- type: precision_at_1
value: 0.35168
- type: precision_at_10
value: 0.07985
- type: precision_at_100
value: 0.01185
- type: precision_at_1000
value: 0.00144
- type: precision_at_3
value: 0.18221
- type: precision_at_5
value: 0.12892
- type: recall_at_1
value: 0.29819
- type: recall_at_10
value: 0.60075
- type: recall_at_100
value: 0.82771
- type: recall_at_1000
value: 0.95219
- type: recall_at_3
value: 0.43245
- type: recall_at_5
value: 0.49931
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWebmastersRetrieval
type: mteb/cqadupstack-webmasters
config: default
split: test
metrics:
- type: map_at_1
value: 0.28409
- type: map_at_10
value: 0.37621
- type: map_at_100
value: 0.39233
- type: map_at_1000
value: 0.39471
- type: map_at_3
value: 0.34337
- type: map_at_5
value: 0.35985
- type: mrr_at_1
value: 0.33794
- type: mrr_at_10
value: 0.42349
- type: mrr_at_100
value: 0.43196
- type: mrr_at_1000
value: 0.43237
- type: mrr_at_3
value: 0.39526
- type: mrr_at_5
value: 0.41087
- type: ndcg_at_1
value: 0.33794
- type: ndcg_at_10
value: 0.43832
- type: ndcg_at_100
value: 0.49514
- type: ndcg_at_1000
value: 0.51742
- type: ndcg_at_3
value: 0.38442
- type: ndcg_at_5
value: 0.40737
- type: precision_at_1
value: 0.33794
- type: precision_at_10
value: 0.08597
- type: precision_at_100
value: 0.01652
- type: precision_at_1000
value: 0.00251
- type: precision_at_3
value: 0.17787
- type: precision_at_5
value: 0.13241
- type: recall_at_1
value: 0.28409
- type: recall_at_10
value: 0.55388
- type: recall_at_100
value: 0.81517
- type: recall_at_1000
value: 0.95038
- type: recall_at_3
value: 0.40133
- type: recall_at_5
value: 0.45913
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWordpressRetrieval
type: mteb/cqadupstack-wordpress
config: default
split: test
metrics:
- type: map_at_1
value: 0.24067
- type: map_at_10
value: 0.32184
- type: map_at_100
value: 0.33357
- type: map_at_1000
value: 0.33458
- type: map_at_3
value: 0.29492
- type: map_at_5
value: 0.3111
- type: mrr_at_1
value: 0.26248
- type: mrr_at_10
value: 0.34149
- type: mrr_at_100
value: 0.35189
- type: mrr_at_1000
value: 0.35251
- type: mrr_at_3
value: 0.31639
- type: mrr_at_5
value: 0.33182
- type: ndcg_at_1
value: 0.26248
- type: ndcg_at_10
value: 0.36889
- type: ndcg_at_100
value: 0.42426
- type: ndcg_at_1000
value: 0.44745
- type: ndcg_at_3
value: 0.31799
- type: ndcg_at_5
value: 0.34563
- type: precision_at_1
value: 0.26248
- type: precision_at_10
value: 0.05712
- type: precision_at_100
value: 0.00915
- type: precision_at_1000
value: 0.00123
- type: precision_at_3
value: 0.13309
- type: precision_at_5
value: 0.09649
- type: recall_at_1
value: 0.24067
- type: recall_at_10
value: 0.49344
- type: recall_at_100
value: 0.7412
- type: recall_at_1000
value: 0.91276
- type: recall_at_3
value: 0.36272
- type: recall_at_5
value: 0.4277
- task:
type: Retrieval
dataset:
name: MTEB DBPedia
type: mteb/dbpedia
config: default
split: test
metrics:
- type: map_at_1
value: 0.08651
- type: map_at_10
value: 0.17628
- type: map_at_100
value: 0.23354
- type: map_at_1000
value: 0.24827
- type: map_at_3
value: 0.1351
- type: map_at_5
value: 0.15468
- type: mrr_at_1
value: 0.645
- type: mrr_at_10
value: 0.71989
- type: mrr_at_100
value: 0.72332
- type: mrr_at_1000
value: 0.72346
- type: mrr_at_3
value: 0.7025
- type: mrr_at_5
value: 0.71275
- type: ndcg_at_1
value: 0.51375
- type: ndcg_at_10
value: 0.3596
- type: ndcg_at_100
value: 0.39878
- type: ndcg_at_1000
value: 0.47931
- type: ndcg_at_3
value: 0.41275
- type: ndcg_at_5
value: 0.38297
- type: precision_at_1
value: 0.645
- type: precision_at_10
value: 0.2745
- type: precision_at_100
value: 0.08405
- type: precision_at_1000
value: 0.01923
- type: precision_at_3
value: 0.44417
- type: precision_at_5
value: 0.366
- type: recall_at_1
value: 0.08651
- type: recall_at_10
value: 0.22416
- type: recall_at_100
value: 0.46381
- type: recall_at_1000
value: 0.71557
- type: recall_at_3
value: 0.14847
- type: recall_at_5
value: 0.1804
- task:
type: Retrieval
dataset:
name: MTEB FEVER
type: mteb/fever
config: default
split: test
metrics:
- type: map_at_1
value: 0.73211
- type: map_at_10
value: 0.81463
- type: map_at_100
value: 0.81622
- type: map_at_1000
value: 0.81634
- type: map_at_3
value: 0.805
- type: map_at_5
value: 0.81134
- type: mrr_at_1
value: 0.79088
- type: mrr_at_10
value: 0.86943
- type: mrr_at_100
value: 0.87017
- type: mrr_at_1000
value: 0.87018
- type: mrr_at_3
value: 0.86154
- type: mrr_at_5
value: 0.867
- type: ndcg_at_1
value: 0.79088
- type: ndcg_at_10
value: 0.85528
- type: ndcg_at_100
value: 0.86134
- type: ndcg_at_1000
value: 0.86367
- type: ndcg_at_3
value: 0.83943
- type: ndcg_at_5
value: 0.84878
- type: precision_at_1
value: 0.79088
- type: precision_at_10
value: 0.10132
- type: precision_at_100
value: 0.01055
- type: precision_at_1000
value: 0.00109
- type: precision_at_3
value: 0.31963
- type: precision_at_5
value: 0.19769
- type: recall_at_1
value: 0.73211
- type: recall_at_10
value: 0.92797
- type: recall_at_100
value: 0.95263
- type: recall_at_1000
value: 0.96738
- type: recall_at_3
value: 0.88328
- type: recall_at_5
value: 0.90821
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018
type: mteb/fiqa
config: default
split: test
metrics:
- type: map_at_1
value: 0.18311
- type: map_at_10
value: 0.29201
- type: map_at_100
value: 0.3093
- type: map_at_1000
value: 0.31116
- type: map_at_3
value: 0.24778
- type: map_at_5
value: 0.27453
- type: mrr_at_1
value: 0.35494
- type: mrr_at_10
value: 0.44489
- type: mrr_at_100
value: 0.4532
- type: mrr_at_1000
value: 0.45369
- type: mrr_at_3
value: 0.41667
- type: mrr_at_5
value: 0.43418
- type: ndcg_at_1
value: 0.35494
- type: ndcg_at_10
value: 0.36868
- type: ndcg_at_100
value: 0.43463
- type: ndcg_at_1000
value: 0.46766
- type: ndcg_at_3
value: 0.32305
- type: ndcg_at_5
value: 0.34332
- type: precision_at_1
value: 0.35494
- type: precision_at_10
value: 0.10324
- type: precision_at_100
value: 0.01707
- type: precision_at_1000
value: 0.00229
- type: precision_at_3
value: 0.21142
- type: precision_at_5
value: 0.16327
- type: recall_at_1
value: 0.18311
- type: recall_at_10
value: 0.43881
- type: recall_at_100
value: 0.68593
- type: recall_at_1000
value: 0.8855
- type: recall_at_3
value: 0.28824
- type: recall_at_5
value: 0.36178
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA
type: mteb/hotpotqa
config: default
split: test
metrics:
- type: map_at_1
value: 0.36766
- type: map_at_10
value: 0.53639
- type: map_at_100
value: 0.54532
- type: map_at_1000
value: 0.54608
- type: map_at_3
value: 0.50427
- type: map_at_5
value: 0.5245
- type: mrr_at_1
value: 0.73531
- type: mrr_at_10
value: 0.80104
- type: mrr_at_100
value: 0.80341
- type: mrr_at_1000
value: 0.80351
- type: mrr_at_3
value: 0.78949
- type: mrr_at_5
value: 0.79729
- type: ndcg_at_1
value: 0.73531
- type: ndcg_at_10
value: 0.62918
- type: ndcg_at_100
value: 0.66056
- type: ndcg_at_1000
value: 0.67554
- type: ndcg_at_3
value: 0.58247
- type: ndcg_at_5
value: 0.60905
- type: precision_at_1
value: 0.73531
- type: precision_at_10
value: 0.1302
- type: precision_at_100
value: 0.01546
- type: precision_at_1000
value: 0.00175
- type: precision_at_3
value: 0.36556
- type: precision_at_5
value: 0.24032
- type: recall_at_1
value: 0.36766
- type: recall_at_10
value: 0.65098
- type: recall_at_100
value: 0.77306
- type: recall_at_1000
value: 0.87252
- type: recall_at_3
value: 0.54835
- type: recall_at_5
value: 0.60081
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO
type: mteb/msmarco
config: default
split: dev
metrics:
- type: map_at_1
value: 0.14654
- type: map_at_10
value: 0.2472
- type: map_at_100
value: 0.25994
- type: map_at_1000
value: 0.26067
- type: map_at_3
value: 0.21234
- type: map_at_5
value: 0.2319
- type: mrr_at_1
value: 0.15086
- type: mrr_at_10
value: 0.25184
- type: mrr_at_100
value: 0.26422
- type: mrr_at_1000
value: 0.26489
- type: mrr_at_3
value: 0.21731
- type: mrr_at_5
value: 0.23674
- type: ndcg_at_1
value: 0.15086
- type: ndcg_at_10
value: 0.30711
- type: ndcg_at_100
value: 0.37221
- type: ndcg_at_1000
value: 0.39133
- type: ndcg_at_3
value: 0.23567
- type: ndcg_at_5
value: 0.27066
- type: precision_at_1
value: 0.15086
- type: precision_at_10
value: 0.05132
- type: precision_at_100
value: 0.00845
- type: precision_at_1000
value: 0.00101
- type: precision_at_3
value: 0.10277
- type: precision_at_5
value: 0.07923
- type: recall_at_1
value: 0.14654
- type: recall_at_10
value: 0.49341
- type: recall_at_100
value: 0.80224
- type: recall_at_1000
value: 0.95037
- type: recall_at_3
value: 0.29862
- type: recall_at_5
value: 0.38274
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus
type: mteb/nfcorpus
config: default
split: test
metrics:
- type: map_at_1
value: 0.05452
- type: map_at_10
value: 0.12758
- type: map_at_100
value: 0.1593
- type: map_at_1000
value: 0.17422
- type: map_at_3
value: 0.0945
- type: map_at_5
value: 0.1092
- type: mrr_at_1
value: 0.43963
- type: mrr_at_10
value: 0.53237
- type: mrr_at_100
value: 0.53777
- type: mrr_at_1000
value: 0.53822
- type: mrr_at_3
value: 0.51445
- type: mrr_at_5
value: 0.52466
- type: ndcg_at_1
value: 0.41486
- type: ndcg_at_10
value: 0.33737
- type: ndcg_at_100
value: 0.30886
- type: ndcg_at_1000
value: 0.40018
- type: ndcg_at_3
value: 0.39324
- type: ndcg_at_5
value: 0.36949
- type: precision_at_1
value: 0.43344
- type: precision_at_10
value: 0.24799
- type: precision_at_100
value: 0.07895
- type: precision_at_1000
value: 0.02091
- type: precision_at_3
value: 0.37152
- type: precision_at_5
value: 0.31703
- type: recall_at_1
value: 0.05452
- type: recall_at_10
value: 0.1712
- type: recall_at_100
value: 0.30719
- type: recall_at_1000
value: 0.62766
- type: recall_at_3
value: 0.10733
- type: recall_at_5
value: 0.13553
- task:
type: Retrieval
dataset:
name: MTEB NQ
type: mteb/nq
config: default
split: test
metrics:
- type: map_at_1
value: 0.29022
- type: map_at_10
value: 0.4373
- type: map_at_100
value: 0.44849
- type: map_at_1000
value: 0.44877
- type: map_at_3
value: 0.39045
- type: map_at_5
value: 0.4186
- type: mrr_at_1
value: 0.32793
- type: mrr_at_10
value: 0.46243
- type: mrr_at_100
value: 0.47083
- type: mrr_at_1000
value: 0.47101
- type: mrr_at_3
value: 0.42261
- type: mrr_at_5
value: 0.44775
- type: ndcg_at_1
value: 0.32793
- type: ndcg_at_10
value: 0.51631
- type: ndcg_at_100
value: 0.56287
- type: ndcg_at_1000
value: 0.56949
- type: ndcg_at_3
value: 0.42782
- type: ndcg_at_5
value: 0.47554
- type: precision_at_1
value: 0.32793
- type: precision_at_10
value: 0.08737
- type: precision_at_100
value: 0.01134
- type: precision_at_1000
value: 0.0012
- type: precision_at_3
value: 0.19583
- type: precision_at_5
value: 0.14484
- type: recall_at_1
value: 0.29022
- type: recall_at_10
value: 0.73325
- type: recall_at_100
value: 0.93455
- type: recall_at_1000
value: 0.98414
- type: recall_at_3
value: 0.50406
- type: recall_at_5
value: 0.6145
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval
type: mteb/quora
config: default
split: test
metrics:
- type: map_at_1
value: 0.68941
- type: map_at_10
value: 0.82641
- type: map_at_100
value: 0.83317
- type: map_at_1000
value: 0.83337
- type: map_at_3
value: 0.79604
- type: map_at_5
value: 0.81525
- type: mrr_at_1
value: 0.7935
- type: mrr_at_10
value: 0.85969
- type: mrr_at_100
value: 0.86094
- type: mrr_at_1000
value: 0.86095
- type: mrr_at_3
value: 0.84852
- type: mrr_at_5
value: 0.85627
- type: ndcg_at_1
value: 0.7936
- type: ndcg_at_10
value: 0.86687
- type: ndcg_at_100
value: 0.88094
- type: ndcg_at_1000
value: 0.88243
- type: ndcg_at_3
value: 0.83538
- type: ndcg_at_5
value: 0.85308
- type: precision_at_1
value: 0.7936
- type: precision_at_10
value: 0.13145
- type: precision_at_100
value: 0.01517
- type: precision_at_1000
value: 0.00156
- type: precision_at_3
value: 0.36353
- type: precision_at_5
value: 0.24044
- type: recall_at_1
value: 0.68941
- type: recall_at_10
value: 0.94407
- type: recall_at_100
value: 0.99226
- type: recall_at_1000
value: 0.99958
- type: recall_at_3
value: 0.85502
- type: recall_at_5
value: 0.90372
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS
type: mteb/scidocs
config: default
split: test
metrics:
- type: map_at_1
value: 0.04988
- type: map_at_10
value: 0.13553
- type: map_at_100
value: 0.16136
- type: map_at_1000
value: 0.16512
- type: map_at_3
value: 0.09439
- type: map_at_5
value: 0.1146
- type: mrr_at_1
value: 0.246
- type: mrr_at_10
value: 0.36792
- type: mrr_at_100
value: 0.37973
- type: mrr_at_1000
value: 0.38011
- type: mrr_at_3
value: 0.33117
- type: mrr_at_5
value: 0.35172
- type: ndcg_at_1
value: 0.246
- type: ndcg_at_10
value: 0.22542
- type: ndcg_at_100
value: 0.32326
- type: ndcg_at_1000
value: 0.3828
- type: ndcg_at_3
value: 0.20896
- type: ndcg_at_5
value: 0.18497
- type: precision_at_1
value: 0.246
- type: precision_at_10
value: 0.1194
- type: precision_at_100
value: 0.02616
- type: precision_at_1000
value: 0.00404
- type: precision_at_3
value: 0.198
- type: precision_at_5
value: 0.1654
- type: recall_at_1
value: 0.04988
- type: recall_at_10
value: 0.24212
- type: recall_at_100
value: 0.53105
- type: recall_at_1000
value: 0.82022
- type: recall_at_3
value: 0.12047
- type: recall_at_5
value: 0.16777
- task:
type: Retrieval
dataset:
name: MTEB SciFact
type: mteb/scifact
config: default
split: test
metrics:
- type: map_at_1
value: 0.56578
- type: map_at_10
value: 0.66725
- type: map_at_100
value: 0.67379
- type: map_at_1000
value: 0.674
- type: map_at_3
value: 0.63416
- type: map_at_5
value: 0.6577
- type: mrr_at_1
value: 0.59333
- type: mrr_at_10
value: 0.67533
- type: mrr_at_100
value: 0.68062
- type: mrr_at_1000
value: 0.68082
- type: mrr_at_3
value: 0.64944
- type: mrr_at_5
value: 0.66928
- type: ndcg_at_1
value: 0.59333
- type: ndcg_at_10
value: 0.7127
- type: ndcg_at_100
value: 0.73889
- type: ndcg_at_1000
value: 0.7441
- type: ndcg_at_3
value: 0.65793
- type: ndcg_at_5
value: 0.69429
- type: precision_at_1
value: 0.59333
- type: precision_at_10
value: 0.096
- type: precision_at_100
value: 0.01087
- type: precision_at_1000
value: 0.00113
- type: precision_at_3
value: 0.25556
- type: precision_at_5
value: 0.17667
- type: recall_at_1
value: 0.56578
- type: recall_at_10
value: 0.842
- type: recall_at_100
value: 0.95667
- type: recall_at_1000
value: 0.99667
- type: recall_at_3
value: 0.70072
- type: recall_at_5
value: 0.79011
- task:
type: Retrieval
dataset:
name: MTEB Touche2020
type: mteb/touche2020
config: default
split: test
metrics:
- type: map_at_1
value: 0.01976
- type: map_at_10
value: 0.09688
- type: map_at_100
value: 0.15117
- type: map_at_1000
value: 0.16769
- type: map_at_3
value: 0.04589
- type: map_at_5
value: 0.06556
- type: mrr_at_1
value: 0.26531
- type: mrr_at_10
value: 0.43863
- type: mrr_at_100
value: 0.44767
- type: mrr_at_1000
value: 0.44767
- type: mrr_at_3
value: 0.39116
- type: mrr_at_5
value: 0.41156
- type: ndcg_at_1
value: 0.23469
- type: ndcg_at_10
value: 0.24029
- type: ndcg_at_100
value: 0.34425
- type: ndcg_at_1000
value: 0.46907
- type: ndcg_at_3
value: 0.25522
- type: ndcg_at_5
value: 0.24333
- type: precision_at_1
value: 0.26531
- type: precision_at_10
value: 0.22449
- type: precision_at_100
value: 0.07122
- type: precision_at_1000
value: 0.01527
- type: precision_at_3
value: 0.27891
- type: precision_at_5
value: 0.25714
- type: recall_at_1
value: 0.01976
- type: recall_at_10
value: 0.16633
- type: recall_at_100
value: 0.4561
- type: recall_at_1000
value: 0.82481
- type: recall_at_3
value: 0.06101
- type: recall_at_5
value: 0.0968
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID
type: mteb/trec-covid
config: default
split: test
metrics:
- type: map_at_1
value: 0.00211
- type: map_at_10
value: 0.01526
- type: map_at_100
value: 0.08863
- type: map_at_1000
value: 0.23162
- type: map_at_3
value: 0.00555
- type: map_at_5
value: 0.00873
- type: mrr_at_1
value: 0.76
- type: mrr_at_10
value: 0.8485
- type: mrr_at_100
value: 0.8485
- type: mrr_at_1000
value: 0.8485
- type: mrr_at_3
value: 0.84
- type: mrr_at_5
value: 0.844
- type: ndcg_at_1
value: 0.7
- type: ndcg_at_10
value: 0.63098
- type: ndcg_at_100
value: 0.49847
- type: ndcg_at_1000
value: 0.48395
- type: ndcg_at_3
value: 0.68704
- type: ndcg_at_5
value: 0.67533
- type: precision_at_1
value: 0.76
- type: precision_at_10
value: 0.66
- type: precision_at_100
value: 0.5134
- type: precision_at_1000
value: 0.2168
- type: precision_at_3
value: 0.72667
- type: precision_at_5
value: 0.716
- type: recall_at_1
value: 0.00211
- type: recall_at_10
value: 0.01748
- type: recall_at_100
value: 0.12448
- type: recall_at_1000
value: 0.46795
- type: recall_at_3
value: 0.00593
- type: recall_at_5
value: 0.00962
---
## Llamacpp Static Quantizations of granite-embedding-30m-english
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b4381">b4381</a> for quantization.
Original model: https://huggingface.co/ibm-granite/granite-embedding-30m-english
Run them in [LM Studio](https://lmstudio.ai/)
## Prompt format
No prompt format found, check original model page
## What's new:
Fix tokenizer
## Download a file (not the whole branch) from below:
| Filename | Quant type | File Size | Split | Description |
| -------- | ---------- | --------- | ----- | ----------- |
| [granite-embedding-30m-english-f16.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-f16.gguf) | f16 | 0.06GB | false | Full F16 weights. |
| [granite-embedding-30m-english-Q8_0.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q8_0.gguf) | Q8_0 | 0.03GB | false | Extremely high quality, generally unneeded but max available quant. |
| [granite-embedding-30m-english-Q6_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q6_K_L.gguf) | Q6_K_L | 0.03GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, *recommended*. |
| [granite-embedding-30m-english-Q6_K.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q6_K.gguf) | Q6_K | 0.03GB | false | Very high quality, near perfect, *recommended*. |
| [granite-embedding-30m-english-Q5_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q5_K_L.gguf) | Q5_K_L | 0.03GB | false | Uses Q8_0 for embed and output weights. High quality, *recommended*. |
| [granite-embedding-30m-english-Q5_K_M.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q5_K_M.gguf) | Q5_K_M | 0.03GB | false | High quality, *recommended*. |
| [granite-embedding-30m-english-Q5_K_S.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q5_K_S.gguf) | Q5_K_S | 0.03GB | false | High quality, *recommended*. |
| [granite-embedding-30m-english-Q4_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q4_K_L.gguf) | Q4_K_L | 0.03GB | false | Uses Q8_0 for embed and output weights. Good quality, *recommended*. |
| [granite-embedding-30m-english-Q4_K_M.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q4_K_M.gguf) | Q4_K_M | 0.03GB | false | Good quality, default size for most use cases, *recommended*. |
| [granite-embedding-30m-english-Q4_K_S.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q4_K_S.gguf) | Q4_K_S | 0.03GB | false | Slightly lower quality with more space savings, *recommended*. |
| [granite-embedding-30m-english-Q4_0.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q4_0.gguf) | Q4_0 | 0.03GB | false | Legacy format, offers online repacking for ARM and AVX CPU inference. |
| [granite-embedding-30m-english-IQ4_NL.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-IQ4_NL.gguf) | IQ4_NL | 0.03GB | false | Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference. |
| [granite-embedding-30m-english-IQ4_XS.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-IQ4_XS.gguf) | IQ4_XS | 0.03GB | false | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
| [granite-embedding-30m-english-Q3_K_XL.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q3_K_XL.gguf) | Q3_K_XL | 0.03GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |
| [granite-embedding-30m-english-Q3_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q3_K_L.gguf) | Q3_K_L | 0.03GB | false | Lower quality but usable, good for low RAM availability. |
| [granite-embedding-30m-english-Q3_K_M.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-Q3_K_M.gguf) | Q3_K_M | 0.03GB | false | Low quality. |
| [granite-embedding-30m-english-IQ3_M.gguf](https://huggingface.co/bartowski/granite-embedding-30m-english-GGUF/blob/main/granite-embedding-30m-english-IQ3_M.gguf) | IQ3_M | 0.03GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
## Embed/output weights
Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.
## Downloading using huggingface-cli
<details>
<summary>Click to view download instructions</summary>
First, make sure you have hugginface-cli installed:
```
pip install -U "huggingface_hub[cli]"
```
Then, you can target the specific file you want:
```
huggingface-cli download bartowski/granite-embedding-30m-english-GGUF --include "granite-embedding-30m-english-Q4_K_M.gguf" --local-dir ./
```
If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
```
huggingface-cli download bartowski/granite-embedding-30m-english-GGUF --include "granite-embedding-30m-english-Q8_0/*" --local-dir ./
```
You can either specify a new local-dir (granite-embedding-30m-english-Q8_0) or download them all in place (./)
</details>
## ARM/AVX information
Previously, you would download Q4_0_4_4/4_8/8_8, and these would have their weights interleaved in memory in order to improve performance on ARM and AVX machines by loading up more data in one pass.
Now, however, there is something called "online repacking" for weights. details in [this PR](https://github.com/ggerganov/llama.cpp/pull/9921). If you use Q4_0 and your hardware would benefit from repacking weights, it will do it automatically on the fly.
As of llama.cpp build [b4282](https://github.com/ggerganov/llama.cpp/releases/tag/b4282) you will not be able to run the Q4_0_X_X files and will instead need to use Q4_0.
Additionally, if you want to get slightly better quality for , you can use IQ4_NL thanks to [this PR](https://github.com/ggerganov/llama.cpp/pull/10541) which will also repack the weights for ARM, though only the 4_4 for now. The loading time may be slower but it will result in an overall speed incrase.
<details>
<summary>Click to view Q4_0_X_X information (deprecated</summary>
I'm keeping this section to show the potential theoretical uplift in performance from using the Q4_0 with online repacking.
<details>
<summary>Click to view benchmarks on an AVX2 system (EPYC7702)</summary>
| model | size | params | backend | threads | test | t/s | % (vs Q4_0) |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |-------------: |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | pp512 | 204.03 ± 1.03 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | pp1024 | 282.92 ± 0.19 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | pp2048 | 259.49 ± 0.44 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | tg128 | 39.12 ± 0.27 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | tg256 | 39.31 ± 0.69 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | tg512 | 40.52 ± 0.03 | 100% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | pp512 | 301.02 ± 1.74 | 147% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | pp1024 | 287.23 ± 0.20 | 101% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | pp2048 | 262.77 ± 1.81 | 101% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | tg128 | 18.80 ± 0.99 | 48% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | tg256 | 24.46 ± 3.04 | 83% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | tg512 | 36.32 ± 3.59 | 90% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | pp512 | 271.71 ± 3.53 | 133% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | pp1024 | 279.86 ± 45.63 | 100% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | pp2048 | 320.77 ± 5.00 | 124% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | tg128 | 43.51 ± 0.05 | 111% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | tg256 | 43.35 ± 0.09 | 110% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | tg512 | 42.60 ± 0.31 | 105% |
Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation
</details>
</details>
## Which file should I choose?
<details>
<summary>Click here for details</summary>
A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
If you want to get more into the weeds, you can check out this extremely useful feature chart:
[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
</details>
## Credits
Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset.
Thank you ZeroWw for the inspiration to experiment with embed/output.
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
| [
"SCIFACT"
] |
bartowski/granite-embedding-125m-english-GGUF | bartowski | text-generation | [
"gguf",
"language",
"granite",
"embeddings",
"text-generation",
"en",
"base_model:ibm-granite/granite-embedding-125m-english",
"base_model:quantized:ibm-granite/granite-embedding-125m-english",
"license:apache-2.0",
"model-index",
"endpoints_compatible",
"region:us",
"feature-extraction"
] | "2024-12-18T16:52:22Z" | 2024-12-23T04:59:17+00:00 | 526 | 0 | ---
base_model: ibm-granite/granite-embedding-125m-english
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- language
- granite
- embeddings
quantized_by: bartowski
model-index:
- name: ibm-granite/granite-embedding-125m-english
results:
- task:
type: Retrieval
dataset:
name: MTEB ArguaAna
type: mteb/arguana
config: default
split: test
metrics:
- type: map_at_1
value: 0.33642
- type: map_at_10
value: 0.49716
- type: map_at_100
value: 0.50519
- type: map_at_1000
value: 0.50521
- type: map_at_3
value: 0.45057
- type: map_at_5
value: 0.47774
- type: mrr_at_1
value: 0.34922
- type: mrr_at_10
value: 0.50197
- type: mrr_at_100
value: 0.50992
- type: mrr_at_1000
value: 0.50994
- type: mrr_at_3
value: 0.45484
- type: mrr_at_5
value: 0.48272
- type: ndcg_at_1
value: 0.33642
- type: ndcg_at_10
value: 0.58401
- type: ndcg_at_100
value: 0.6157
- type: ndcg_at_1000
value: 0.61608
- type: ndcg_at_3
value: 0.48825
- type: ndcg_at_5
value: 0.53689
- type: precision_at_1
value: 0.33642
- type: precision_at_10
value: 0.08606
- type: precision_at_100
value: 0.00994
- type: precision_at_1000
value: 0.001
- type: precision_at_3
value: 0.19915
- type: precision_at_5
value: 0.14296
- type: recall_at_1
value: 0.33642
- type: recall_at_10
value: 0.8606
- type: recall_at_100
value: 0.9936
- type: recall_at_1000
value: 0.99644
- type: recall_at_3
value: 0.59744
- type: recall_at_5
value: 0.71479
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER
type: mteb/climate-fever
config: default
split: test
metrics:
- type: map_at_1
value: 0.1457
- type: map_at_10
value: 0.24102
- type: map_at_100
value: 0.25826
- type: map_at_1000
value: 0.26021
- type: map_at_3
value: 0.20346
- type: map_at_5
value: 0.22228
- type: mrr_at_1
value: 0.32573
- type: mrr_at_10
value: 0.44411
- type: mrr_at_100
value: 0.45176
- type: mrr_at_1000
value: 0.45209
- type: mrr_at_3
value: 0.4126
- type: mrr_at_5
value: 0.43312
- type: ndcg_at_1
value: 0.32573
- type: ndcg_at_10
value: 0.3315
- type: ndcg_at_100
value: 0.39898
- type: ndcg_at_1000
value: 0.43151
- type: ndcg_at_3
value: 0.27683
- type: ndcg_at_5
value: 0.29538
- type: precision_at_1
value: 0.32573
- type: precision_at_10
value: 0.10176
- type: precision_at_100
value: 0.01754
- type: precision_at_1000
value: 0.00236
- type: precision_at_3
value: 0.20347
- type: precision_at_5
value: 0.15505
- type: recall_at_1
value: 0.1457
- type: recall_at_10
value: 0.38825
- type: recall_at_100
value: 0.62237
- type: recall_at_1000
value: 0.8022
- type: recall_at_3
value: 0.25245
- type: recall_at_5
value: 0.30821
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval
type: mteb/cqadupstack-android
config: default
split: test
metrics:
- type: map_at_1
value: 0.36964
- type: map_at_10
value: 0.5043
- type: map_at_100
value: 0.52066
- type: map_at_1000
value: 0.52175
- type: map_at_3
value: 0.46001
- type: map_at_5
value: 0.48312
- type: mrr_at_1
value: 0.45923
- type: mrr_at_10
value: 0.56733
- type: mrr_at_100
value: 0.57292
- type: mrr_at_1000
value: 0.57321
- type: mrr_at_3
value: 0.54053
- type: mrr_at_5
value: 0.55556
- type: ndcg_at_1
value: 0.45923
- type: ndcg_at_10
value: 0.57667
- type: ndcg_at_100
value: 0.62373
- type: ndcg_at_1000
value: 0.6368
- type: ndcg_at_3
value: 0.51843
- type: ndcg_at_5
value: 0.54257
- type: precision_at_1
value: 0.45923
- type: precision_at_10
value: 0.11316
- type: precision_at_100
value: 0.01705
- type: precision_at_1000
value: 0.00216
- type: precision_at_3
value: 0.2537
- type: precision_at_5
value: 0.1814
- type: recall_at_1
value: 0.36964
- type: recall_at_10
value: 0.71234
- type: recall_at_100
value: 0.90421
- type: recall_at_1000
value: 0.98296
- type: recall_at_3
value: 0.53655
- type: recall_at_5
value: 0.60996
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackEnglishRetrieval
type: mteb/cqadupstack-english
config: default
split: test
metrics:
- type: map_at_1
value: 0.36198
- type: map_at_10
value: 0.49199
- type: map_at_100
value: 0.50602
- type: map_at_1000
value: 0.50736
- type: map_at_3
value: 0.45678
- type: map_at_5
value: 0.47605
- type: mrr_at_1
value: 0.45478
- type: mrr_at_10
value: 0.55075
- type: mrr_at_100
value: 0.55656
- type: mrr_at_1000
value: 0.55688
- type: mrr_at_3
value: 0.52887
- type: mrr_at_5
value: 0.54282
- type: ndcg_at_1
value: 0.45478
- type: ndcg_at_10
value: 0.55505
- type: ndcg_at_100
value: 0.59606
- type: ndcg_at_1000
value: 0.61255
- type: ndcg_at_3
value: 0.51124
- type: ndcg_at_5
value: 0.53166
- type: precision_at_1
value: 0.45478
- type: precision_at_10
value: 0.10752
- type: precision_at_100
value: 0.01666
- type: precision_at_1000
value: 0.00211
- type: precision_at_3
value: 0.25053
- type: precision_at_5
value: 0.17694
- type: recall_at_1
value: 0.36198
- type: recall_at_10
value: 0.66465
- type: recall_at_100
value: 0.83632
- type: recall_at_1000
value: 0.93276
- type: recall_at_3
value: 0.53207
- type: recall_at_5
value: 0.59169
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGamingRetrieval
type: mteb/cqadupstack-gaming
config: default
split: test
metrics:
- type: map_at_1
value: 0.44157
- type: map_at_10
value: 0.57753
- type: map_at_100
value: 0.58698
- type: map_at_1000
value: 0.5874
- type: map_at_3
value: 0.54223
- type: map_at_5
value: 0.56307
- type: mrr_at_1
value: 0.50094
- type: mrr_at_10
value: 0.607
- type: mrr_at_100
value: 0.6126
- type: mrr_at_1000
value: 0.6128
- type: mrr_at_3
value: 0.58265
- type: mrr_at_5
value: 0.59817
- type: ndcg_at_1
value: 0.50094
- type: ndcg_at_10
value: 0.63641
- type: ndcg_at_100
value: 0.67055
- type: ndcg_at_1000
value: 0.67855
- type: ndcg_at_3
value: 0.58022
- type: ndcg_at_5
value: 0.6097
- type: precision_at_1
value: 0.50094
- type: precision_at_10
value: 0.10182
- type: precision_at_100
value: 0.01278
- type: precision_at_1000
value: 0.00138
- type: precision_at_3
value: 0.2581
- type: precision_at_5
value: 0.17755
- type: recall_at_1
value: 0.44157
- type: recall_at_10
value: 0.7778
- type: recall_at_100
value: 0.92244
- type: recall_at_1000
value: 0.9781
- type: recall_at_3
value: 0.63087
- type: recall_at_5
value: 0.70172
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGisRetrieval
type: mteb/cqadupstack-gis
config: default
split: test
metrics:
- type: map_at_1
value: 0.29532
- type: map_at_10
value: 0.40214
- type: map_at_100
value: 0.41289
- type: map_at_1000
value: 0.41359
- type: map_at_3
value: 0.37086
- type: map_at_5
value: 0.38889
- type: mrr_at_1
value: 0.3209
- type: mrr_at_10
value: 0.42423
- type: mrr_at_100
value: 0.43342
- type: mrr_at_1000
value: 0.43395
- type: mrr_at_3
value: 0.39736
- type: mrr_at_5
value: 0.41307
- type: ndcg_at_1
value: 0.3209
- type: ndcg_at_10
value: 0.46075
- type: ndcg_at_100
value: 0.5103
- type: ndcg_at_1000
value: 0.52668
- type: ndcg_at_3
value: 0.40149
- type: ndcg_at_5
value: 0.43111
- type: precision_at_1
value: 0.3209
- type: precision_at_10
value: 0.07141
- type: precision_at_100
value: 0.01018
- type: precision_at_1000
value: 0.00118
- type: precision_at_3
value: 0.17175
- type: precision_at_5
value: 0.12068
- type: recall_at_1
value: 0.29532
- type: recall_at_10
value: 0.62025
- type: recall_at_100
value: 0.83829
- type: recall_at_1000
value: 0.95995
- type: recall_at_3
value: 0.4603
- type: recall_at_5
value: 0.53089
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackMathematicaRetrieval
type: mteb/cqadupstack-mathematica
config: default
split: test
metrics:
- type: map_at_1
value: 0.18944
- type: map_at_10
value: 0.29611
- type: map_at_100
value: 0.31063
- type: map_at_1000
value: 0.31174
- type: map_at_3
value: 0.26098
- type: map_at_5
value: 0.28151
- type: mrr_at_1
value: 0.23756
- type: mrr_at_10
value: 0.34491
- type: mrr_at_100
value: 0.35457
- type: mrr_at_1000
value: 0.35512
- type: mrr_at_3
value: 0.3126
- type: mrr_at_5
value: 0.3317
- type: ndcg_at_1
value: 0.23756
- type: ndcg_at_10
value: 0.36015
- type: ndcg_at_100
value: 0.42175
- type: ndcg_at_1000
value: 0.44607
- type: ndcg_at_3
value: 0.29725
- type: ndcg_at_5
value: 0.32879
- type: precision_at_1
value: 0.23756
- type: precision_at_10
value: 0.06928
- type: precision_at_100
value: 0.01153
- type: precision_at_1000
value: 0.00149
- type: precision_at_3
value: 0.14635
- type: precision_at_5
value: 0.1107
- type: recall_at_1
value: 0.18944
- type: recall_at_10
value: 0.50691
- type: recall_at_100
value: 0.76503
- type: recall_at_1000
value: 0.93624
- type: recall_at_3
value: 0.33611
- type: recall_at_5
value: 0.41427
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackPhysicsRetrieval
type: mteb/cqadupstack-physics
config: default
split: test
metrics:
- type: map_at_1
value: 0.33824
- type: map_at_10
value: 0.46868
- type: map_at_100
value: 0.48306
- type: map_at_1000
value: 0.48406
- type: map_at_3
value: 0.43335
- type: map_at_5
value: 0.45279
- type: mrr_at_1
value: 0.42348
- type: mrr_at_10
value: 0.52972
- type: mrr_at_100
value: 0.53707
- type: mrr_at_1000
value: 0.53734
- type: mrr_at_3
value: 0.50722
- type: mrr_at_5
value: 0.52012
- type: ndcg_at_1
value: 0.42348
- type: ndcg_at_10
value: 0.53504
- type: ndcg_at_100
value: 0.58899
- type: ndcg_at_1000
value: 0.60323
- type: ndcg_at_3
value: 0.48478
- type: ndcg_at_5
value: 0.5079
- type: precision_at_1
value: 0.42348
- type: precision_at_10
value: 0.0975
- type: precision_at_100
value: 0.01466
- type: precision_at_1000
value: 0.00177
- type: precision_at_3
value: 0.23741
- type: precision_at_5
value: 0.16439
- type: recall_at_1
value: 0.33824
- type: recall_at_10
value: 0.67142
- type: recall_at_100
value: 0.89134
- type: recall_at_1000
value: 0.97816
- type: recall_at_3
value: 0.52305
- type: recall_at_5
value: 0.58804
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackProgrammersRetrieval
type: mteb/cqadupstack-programmers
config: default
split: test
metrics:
- type: map_at_1
value: 0.30125
- type: map_at_10
value: 0.42119
- type: map_at_100
value: 0.43599
- type: map_at_1000
value: 0.4369
- type: map_at_3
value: 0.38018
- type: map_at_5
value: 0.40368
- type: mrr_at_1
value: 0.37557
- type: mrr_at_10
value: 0.47573
- type: mrr_at_100
value: 0.4846
- type: mrr_at_1000
value: 0.48499
- type: mrr_at_3
value: 0.44654
- type: mrr_at_5
value: 0.4644
- type: ndcg_at_1
value: 0.37557
- type: ndcg_at_10
value: 0.48743
- type: ndcg_at_100
value: 0.54458
- type: ndcg_at_1000
value: 0.56076
- type: ndcg_at_3
value: 0.42573
- type: ndcg_at_5
value: 0.45528
- type: precision_at_1
value: 0.37557
- type: precision_at_10
value: 0.09269
- type: precision_at_100
value: 0.01401
- type: precision_at_1000
value: 0.0017
- type: precision_at_3
value: 0.20624
- type: precision_at_5
value: 0.15068
- type: recall_at_1
value: 0.30125
- type: recall_at_10
value: 0.62619
- type: recall_at_100
value: 0.86574
- type: recall_at_1000
value: 0.97102
- type: recall_at_3
value: 0.45437
- type: recall_at_5
value: 0.53197
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackStatsRetrieval
type: mteb/cqadupstack-stats
config: default
split: test
metrics:
- type: map_at_1
value: 0.29193
- type: map_at_10
value: 0.37529
- type: map_at_100
value: 0.38614
- type: map_at_1000
value: 0.38714
- type: map_at_3
value: 0.34897
- type: map_at_5
value: 0.36273
- type: mrr_at_1
value: 0.32669
- type: mrr_at_10
value: 0.40288
- type: mrr_at_100
value: 0.41177
- type: mrr_at_1000
value: 0.41241
- type: mrr_at_3
value: 0.38037
- type: mrr_at_5
value: 0.39195
- type: ndcg_at_1
value: 0.32669
- type: ndcg_at_10
value: 0.42353
- type: ndcg_at_100
value: 0.47424
- type: ndcg_at_1000
value: 0.4959
- type: ndcg_at_3
value: 0.37604
- type: ndcg_at_5
value: 0.39682
- type: precision_at_1
value: 0.32669
- type: precision_at_10
value: 0.06871
- type: precision_at_100
value: 0.01008
- type: precision_at_1000
value: 0.00126
- type: precision_at_3
value: 0.16309
- type: precision_at_5
value: 0.11288
- type: recall_at_1
value: 0.29193
- type: recall_at_10
value: 0.54159
- type: recall_at_100
value: 0.77267
- type: recall_at_1000
value: 0.92805
- type: recall_at_3
value: 0.41014
- type: recall_at_5
value: 0.46248
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackTexRetrieval
type: mteb/cqadupstack-tex
config: default
split: test
metrics:
- type: map_at_1
value: 0.21217
- type: map_at_10
value: 0.30848
- type: map_at_100
value: 0.32173
- type: map_at_1000
value: 0.32296
- type: map_at_3
value: 0.27882
- type: map_at_5
value: 0.29537
- type: mrr_at_1
value: 0.25946
- type: mrr_at_10
value: 0.35091
- type: mrr_at_100
value: 0.36047
- type: mrr_at_1000
value: 0.36111
- type: mrr_at_3
value: 0.32485
- type: mrr_at_5
value: 0.33964
- type: ndcg_at_1
value: 0.25946
- type: ndcg_at_10
value: 0.3655
- type: ndcg_at_100
value: 0.42328
- type: ndcg_at_1000
value: 0.44783
- type: ndcg_at_3
value: 0.31463
- type: ndcg_at_5
value: 0.33803
- type: precision_at_1
value: 0.25946
- type: precision_at_10
value: 0.06793
- type: precision_at_100
value: 0.01138
- type: precision_at_1000
value: 0.00155
- type: precision_at_3
value: 0.1513
- type: precision_at_5
value: 0.10991
- type: recall_at_1
value: 0.21217
- type: recall_at_10
value: 0.49327
- type: recall_at_100
value: 0.7472
- type: recall_at_1000
value: 0.91637
- type: recall_at_3
value: 0.34993
- type: recall_at_5
value: 0.41029
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackUnixRetrieval
type: mteb/cqadupstack-unix
config: default
split: test
metrics:
- type: map_at_1
value: 0.34303
- type: map_at_10
value: 0.45312
- type: map_at_100
value: 0.46563
- type: map_at_1000
value: 0.4664
- type: map_at_3
value: 0.4143
- type: map_at_5
value: 0.43633
- type: mrr_at_1
value: 0.40112
- type: mrr_at_10
value: 0.49097
- type: mrr_at_100
value: 0.49966
- type: mrr_at_1000
value: 0.50006
- type: mrr_at_3
value: 0.46129
- type: mrr_at_5
value: 0.47901
- type: ndcg_at_1
value: 0.40112
- type: ndcg_at_10
value: 0.513
- type: ndcg_at_100
value: 0.56534
- type: ndcg_at_1000
value: 0.58048
- type: ndcg_at_3
value: 0.4491
- type: ndcg_at_5
value: 0.48048
- type: precision_at_1
value: 0.40112
- type: precision_at_10
value: 0.08806
- type: precision_at_100
value: 0.01266
- type: precision_at_1000
value: 0.00149
- type: precision_at_3
value: 0.20211
- type: precision_at_5
value: 0.14496
- type: recall_at_1
value: 0.34303
- type: recall_at_10
value: 0.65508
- type: recall_at_100
value: 0.8753
- type: recall_at_1000
value: 0.9742
- type: recall_at_3
value: 0.48465
- type: recall_at_5
value: 0.56374
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWebmastersRetrieval
type: mteb/cqadupstack-webmasters
config: default
split: test
metrics:
- type: map_at_1
value: 0.30312
- type: map_at_10
value: 0.40931
- type: map_at_100
value: 0.42893
- type: map_at_1000
value: 0.4312
- type: map_at_3
value: 0.37527
- type: map_at_5
value: 0.3936
- type: mrr_at_1
value: 0.36364
- type: mrr_at_10
value: 0.45677
- type: mrr_at_100
value: 0.46753
- type: mrr_at_1000
value: 0.46787
- type: mrr_at_3
value: 0.42918
- type: mrr_at_5
value: 0.4443
- type: ndcg_at_1
value: 0.36364
- type: ndcg_at_10
value: 0.47301
- type: ndcg_at_100
value: 0.53698
- type: ndcg_at_1000
value: 0.55503
- type: ndcg_at_3
value: 0.41875
- type: ndcg_at_5
value: 0.44316
- type: precision_at_1
value: 0.36364
- type: precision_at_10
value: 0.09032
- type: precision_at_100
value: 0.01806
- type: precision_at_1000
value: 0.00258
- type: precision_at_3
value: 0.19499
- type: precision_at_5
value: 0.1415
- type: recall_at_1
value: 0.30312
- type: recall_at_10
value: 0.59418
- type: recall_at_100
value: 0.8656
- type: recall_at_1000
value: 0.97412
- type: recall_at_3
value: 0.44251
- type: recall_at_5
value: 0.50457
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWordpressRetrieval
type: mteb/cqadupstack-wordpress
config: default
split: test
metrics:
- type: map_at_1
value: 0.23851
- type: map_at_10
value: 0.33429
- type: map_at_100
value: 0.34482
- type: map_at_1000
value: 0.3457
- type: map_at_3
value: 0.30271
- type: map_at_5
value: 0.31905
- type: mrr_at_1
value: 0.25693
- type: mrr_at_10
value: 0.35383
- type: mrr_at_100
value: 0.36295
- type: mrr_at_1000
value: 0.36346
- type: mrr_at_3
value: 0.32532
- type: mrr_at_5
value: 0.3402
- type: ndcg_at_1
value: 0.25693
- type: ndcg_at_10
value: 0.39196
- type: ndcg_at_100
value: 0.44501
- type: ndcg_at_1000
value: 0.46482
- type: ndcg_at_3
value: 0.33
- type: ndcg_at_5
value: 0.35736
- type: precision_at_1
value: 0.25693
- type: precision_at_10
value: 0.06433
- type: precision_at_100
value: 0.00989
- type: precision_at_1000
value: 0.00128
- type: precision_at_3
value: 0.14295
- type: precision_at_5
value: 0.10277
- type: recall_at_1
value: 0.23851
- type: recall_at_10
value: 0.55036
- type: recall_at_100
value: 0.79592
- type: recall_at_1000
value: 0.94283
- type: recall_at_3
value: 0.38435
- type: recall_at_5
value: 0.44872
- task:
type: Retrieval
dataset:
name: MTEB DBPedia
type: mteb/dbpedia
config: default
split: test
metrics:
- type: map_at_1
value: 0.0871
- type: map_at_10
value: 0.19218
- type: map_at_100
value: 0.26291
- type: map_at_1000
value: 0.27985
- type: map_at_3
value: 0.13974
- type: map_at_5
value: 0.16104
- type: mrr_at_1
value: 0.6725
- type: mrr_at_10
value: 0.75037
- type: mrr_at_100
value: 0.75318
- type: mrr_at_1000
value: 0.75325
- type: mrr_at_3
value: 0.73833
- type: mrr_at_5
value: 0.74308
- type: ndcg_at_1
value: 0.54375
- type: ndcg_at_10
value: 0.39409
- type: ndcg_at_100
value: 0.44382
- type: ndcg_at_1000
value: 0.52485
- type: ndcg_at_3
value: 0.44463
- type: ndcg_at_5
value: 0.41276
- type: precision_at_1
value: 0.6725
- type: precision_at_10
value: 0.3055
- type: precision_at_100
value: 0.09588
- type: precision_at_1000
value: 0.02118
- type: precision_at_3
value: 0.48167
- type: precision_at_5
value: 0.394
- type: recall_at_1
value: 0.0871
- type: recall_at_10
value: 0.2527
- type: recall_at_100
value: 0.5185
- type: recall_at_1000
value: 0.76491
- type: recall_at_3
value: 0.15516
- type: recall_at_5
value: 0.18907
- task:
type: Retrieval
dataset:
name: MTEB FEVER
type: mteb/fever
config: default
split: test
metrics:
- type: map_at_1
value: 0.78993
- type: map_at_10
value: 0.8502
- type: map_at_100
value: 0.85186
- type: map_at_1000
value: 0.852
- type: map_at_3
value: 0.8437
- type: map_at_5
value: 0.84812
- type: mrr_at_1
value: 0.85179
- type: mrr_at_10
value: 0.90744
- type: mrr_at_100
value: 0.90799
- type: mrr_at_1000
value: 0.90801
- type: mrr_at_3
value: 0.90322
- type: mrr_at_5
value: 0.90622
- type: ndcg_at_1
value: 0.85179
- type: ndcg_at_10
value: 0.88229
- type: ndcg_at_100
value: 0.8884
- type: ndcg_at_1000
value: 0.89116
- type: ndcg_at_3
value: 0.87304
- type: ndcg_at_5
value: 0.87862
- type: precision_at_1
value: 0.85179
- type: precision_at_10
value: 0.10129
- type: precision_at_100
value: 0.0106
- type: precision_at_1000
value: 0.0011
- type: precision_at_3
value: 0.32543
- type: precision_at_5
value: 0.19931
- type: recall_at_1
value: 0.78993
- type: recall_at_10
value: 0.92685
- type: recall_at_100
value: 0.9516
- type: recall_at_1000
value: 0.96943
- type: recall_at_3
value: 0.89965
- type: recall_at_5
value: 0.91562
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018
type: mteb/fiqa
config: default
split: test
metrics:
- type: map_at_1
value: 0.22586
- type: map_at_10
value: 0.36836
- type: map_at_100
value: 0.38863
- type: map_at_1000
value: 0.39041
- type: map_at_3
value: 0.32445
- type: map_at_5
value: 0.34951
- type: mrr_at_1
value: 0.44599
- type: mrr_at_10
value: 0.53471
- type: mrr_at_100
value: 0.54186
- type: mrr_at_1000
value: 0.54223
- type: mrr_at_3
value: 0.51157
- type: mrr_at_5
value: 0.52423
- type: ndcg_at_1
value: 0.44599
- type: ndcg_at_10
value: 0.44931
- type: ndcg_at_100
value: 0.51914
- type: ndcg_at_1000
value: 0.54674
- type: ndcg_at_3
value: 0.41597
- type: ndcg_at_5
value: 0.42611
- type: precision_at_1
value: 0.44599
- type: precision_at_10
value: 0.12346
- type: precision_at_100
value: 0.01951
- type: precision_at_1000
value: 0.00244
- type: precision_at_3
value: 0.27623
- type: precision_at_5
value: 0.20093
- type: recall_at_1
value: 0.22586
- type: recall_at_10
value: 0.5152
- type: recall_at_100
value: 0.77251
- type: recall_at_1000
value: 0.93503
- type: recall_at_3
value: 0.37802
- type: recall_at_5
value: 0.4386
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA
type: mteb/hotpotqa
config: default
split: test
metrics:
- type: map_at_1
value: 0.38177
- type: map_at_10
value: 0.59021
- type: map_at_100
value: 0.59924
- type: map_at_1000
value: 0.59989
- type: map_at_3
value: 0.55553
- type: map_at_5
value: 0.57773
- type: mrr_at_1
value: 0.76354
- type: mrr_at_10
value: 0.827
- type: mrr_at_100
value: 0.82887
- type: mrr_at_1000
value: 0.82896
- type: mrr_at_3
value: 0.8172
- type: mrr_at_5
value: 0.82338
- type: ndcg_at_1
value: 0.76354
- type: ndcg_at_10
value: 0.67775
- type: ndcg_at_100
value: 0.70849
- type: ndcg_at_1000
value: 0.7215
- type: ndcg_at_3
value: 0.629
- type: ndcg_at_5
value: 0.65679
- type: precision_at_1
value: 0.76354
- type: precision_at_10
value: 0.14176
- type: precision_at_100
value: 0.01656
- type: precision_at_1000
value: 0.00183
- type: precision_at_3
value: 0.40113
- type: precision_at_5
value: 0.26255
- type: recall_at_1
value: 0.38177
- type: recall_at_10
value: 0.70878
- type: recall_at_100
value: 0.82822
- type: recall_at_1000
value: 0.91472
- type: recall_at_3
value: 0.60169
- type: recall_at_5
value: 0.65638
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO
type: mteb/msmarco
config: default
split: dev
metrics:
- type: map_at_1
value: 0.15062
- type: map_at_10
value: 0.26008
- type: map_at_100
value: 0.27305
- type: map_at_1000
value: 0.27373
- type: map_at_3
value: 0.22236
- type: map_at_5
value: 0.24362
- type: mrr_at_1
value: 0.15444
- type: mrr_at_10
value: 0.26458
- type: mrr_at_100
value: 0.27718
- type: mrr_at_1000
value: 0.2778
- type: mrr_at_3
value: 0.22701
- type: mrr_at_5
value: 0.24844
- type: ndcg_at_1
value: 0.15444
- type: ndcg_at_10
value: 0.32495
- type: ndcg_at_100
value: 0.38957
- type: ndcg_at_1000
value: 0.40684
- type: ndcg_at_3
value: 0.24745
- type: ndcg_at_5
value: 0.2856
- type: precision_at_1
value: 0.15444
- type: precision_at_10
value: 0.05486
- type: precision_at_100
value: 0.00875
- type: precision_at_1000
value: 0.00102
- type: precision_at_3
value: 0.1086
- type: precision_at_5
value: 0.08441
- type: recall_at_1
value: 0.15062
- type: recall_at_10
value: 0.5272
- type: recall_at_100
value: 0.83006
- type: recall_at_1000
value: 0.96263
- type: recall_at_3
value: 0.31556
- type: recall_at_5
value: 0.40706
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus
type: mteb/nfcorpus
config: default
split: test
metrics:
- type: map_at_1
value: 0.06126
- type: map_at_10
value: 0.14152
- type: map_at_100
value: 0.1827
- type: map_at_1000
value: 0.1988
- type: map_at_3
value: 0.10301
- type: map_at_5
value: 0.12085
- type: mrr_at_1
value: 0.47988
- type: mrr_at_10
value: 0.5692
- type: mrr_at_100
value: 0.57428
- type: mrr_at_1000
value: 0.57482
- type: mrr_at_3
value: 0.55315
- type: mrr_at_5
value: 0.56352
- type: ndcg_at_1
value: 0.45356
- type: ndcg_at_10
value: 0.3725
- type: ndcg_at_100
value: 0.34496
- type: ndcg_at_1000
value: 0.43374
- type: ndcg_at_3
value: 0.42643
- type: ndcg_at_5
value: 0.40882
- type: precision_at_1
value: 0.47368
- type: precision_at_10
value: 0.2774
- type: precision_at_100
value: 0.09071
- type: precision_at_1000
value: 0.02226
- type: precision_at_3
value: 0.40144
- type: precision_at_5
value: 0.35913
- type: recall_at_1
value: 0.06126
- type: recall_at_10
value: 0.18427
- type: recall_at_100
value: 0.35018
- type: recall_at_1000
value: 0.6766
- type: recall_at_3
value: 0.11706
- type: recall_at_5
value: 0.14419
- task:
type: Retrieval
dataset:
name: MTEB NQ
type: mteb/nq
config: default
split: test
metrics:
- type: map_at_1
value: 0.33053
- type: map_at_10
value: 0.49739
- type: map_at_100
value: 0.50626
- type: map_at_1000
value: 0.50647
- type: map_at_3
value: 0.4491
- type: map_at_5
value: 0.4783
- type: mrr_at_1
value: 0.37254
- type: mrr_at_10
value: 0.52222
- type: mrr_at_100
value: 0.52855
- type: mrr_at_1000
value: 0.52869
- type: mrr_at_3
value: 0.48445
- type: mrr_at_5
value: 0.50834
- type: ndcg_at_1
value: 0.37254
- type: ndcg_at_10
value: 0.58044
- type: ndcg_at_100
value: 0.61613
- type: ndcg_at_1000
value: 0.62046
- type: ndcg_at_3
value: 0.49219
- type: ndcg_at_5
value: 0.54037
- type: precision_at_1
value: 0.37254
- type: precision_at_10
value: 0.09655
- type: precision_at_100
value: 0.01167
- type: precision_at_1000
value: 0.00121
- type: precision_at_3
value: 0.22538
- type: precision_at_5
value: 0.16344
- type: recall_at_1
value: 0.33053
- type: recall_at_10
value: 0.8076
- type: recall_at_100
value: 0.95862
- type: recall_at_1000
value: 0.99044
- type: recall_at_3
value: 0.58157
- type: recall_at_5
value: 0.69235
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval
type: mteb/quora
config: default
split: test
metrics:
- type: map_at_1
value: 0.70056
- type: map_at_10
value: 0.84009
- type: map_at_100
value: 0.84661
- type: map_at_1000
value: 0.84678
- type: map_at_3
value: 0.81036
- type: map_at_5
value: 0.82923
- type: mrr_at_1
value: 0.8062
- type: mrr_at_10
value: 0.86971
- type: mrr_at_100
value: 0.87079
- type: mrr_at_1000
value: 0.8708
- type: mrr_at_3
value: 0.85943
- type: mrr_at_5
value: 0.86664
- type: ndcg_at_1
value: 0.8064
- type: ndcg_at_10
value: 0.87821
- type: ndcg_at_100
value: 0.89091
- type: ndcg_at_1000
value: 0.89202
- type: ndcg_at_3
value: 0.849
- type: ndcg_at_5
value: 0.86544
- type: precision_at_1
value: 0.8064
- type: precision_at_10
value: 0.13347
- type: precision_at_100
value: 0.01527
- type: precision_at_1000
value: 0.00157
- type: precision_at_3
value: 0.37153
- type: precision_at_5
value: 0.2448
- type: recall_at_1
value: 0.70056
- type: recall_at_10
value: 0.95148
- type: recall_at_100
value: 0.99474
- type: recall_at_1000
value: 0.99977
- type: recall_at_3
value: 0.86773
- type: recall_at_5
value: 0.91396
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS
type: mteb/scidocs
config: default
split: test
metrics:
- type: map_at_1
value: 0.05737
- type: map_at_10
value: 0.14896
- type: map_at_100
value: 0.17646
- type: map_at_1000
value: 0.1803
- type: map_at_3
value: 0.10474
- type: map_at_5
value: 0.12656
- type: mrr_at_1
value: 0.281
- type: mrr_at_10
value: 0.39579
- type: mrr_at_100
value: 0.40687
- type: mrr_at_1000
value: 0.40722
- type: mrr_at_3
value: 0.35917
- type: mrr_at_5
value: 0.38097
- type: ndcg_at_1
value: 0.281
- type: ndcg_at_10
value: 0.24146
- type: ndcg_at_100
value: 0.339
- type: ndcg_at_1000
value: 0.39728
- type: ndcg_at_3
value: 0.22721
- type: ndcg_at_5
value: 0.20015
- type: precision_at_1
value: 0.281
- type: precision_at_10
value: 0.1254
- type: precision_at_100
value: 0.02651
- type: precision_at_1000
value: 0.00404
- type: precision_at_3
value: 0.212
- type: precision_at_5
value: 0.176
- type: recall_at_1
value: 0.05737
- type: recall_at_10
value: 0.254
- type: recall_at_100
value: 0.53772
- type: recall_at_1000
value: 0.82013
- type: recall_at_3
value: 0.12897
- type: recall_at_5
value: 0.17855
- task:
type: Retrieval
dataset:
name: MTEB SciFact
type: mteb/scifact
config: default
split: test
metrics:
- type: map_at_1
value: 0.60011
- type: map_at_10
value: 0.70101
- type: map_at_100
value: 0.70687
- type: map_at_1000
value: 0.70699
- type: map_at_3
value: 0.67135
- type: map_at_5
value: 0.6878
- type: mrr_at_1
value: 0.62667
- type: mrr_at_10
value: 0.71022
- type: mrr_at_100
value: 0.71484
- type: mrr_at_1000
value: 0.71496
- type: mrr_at_3
value: 0.68944
- type: mrr_at_5
value: 0.69961
- type: ndcg_at_1
value: 0.62667
- type: ndcg_at_10
value: 0.7472
- type: ndcg_at_100
value: 0.76961
- type: ndcg_at_1000
value: 0.77294
- type: ndcg_at_3
value: 0.69776
- type: ndcg_at_5
value: 0.71964
- type: precision_at_1
value: 0.62667
- type: precision_at_10
value: 0.09933
- type: precision_at_100
value: 0.01103
- type: precision_at_1000
value: 0.00113
- type: precision_at_3
value: 0.27
- type: precision_at_5
value: 0.178
- type: recall_at_1
value: 0.60011
- type: recall_at_10
value: 0.878
- type: recall_at_100
value: 0.97333
- type: recall_at_1000
value: 1
- type: recall_at_3
value: 0.74839
- type: recall_at_5
value: 0.80028
- task:
type: Retrieval
dataset:
name: MTEB Touche2020
type: mteb/touche2020
config: default
split: test
metrics:
- type: map_at_1
value: 0.02152
- type: map_at_10
value: 0.07747
- type: map_at_100
value: 0.1364
- type: map_at_1000
value: 0.15235
- type: map_at_3
value: 0.04103
- type: map_at_5
value: 0.05482
- type: mrr_at_1
value: 0.26531
- type: mrr_at_10
value: 0.41399
- type: mrr_at_100
value: 0.43047
- type: mrr_at_1000
value: 0.43047
- type: mrr_at_3
value: 0.38776
- type: mrr_at_5
value: 0.40612
- type: ndcg_at_1
value: 0.23469
- type: ndcg_at_10
value: 0.20147
- type: ndcg_at_100
value: 0.3279
- type: ndcg_at_1000
value: 0.45324
- type: ndcg_at_3
value: 0.22555
- type: ndcg_at_5
value: 0.2097
- type: precision_at_1
value: 0.26531
- type: precision_at_10
value: 0.17755
- type: precision_at_100
value: 0.07082
- type: precision_at_1000
value: 0.01547
- type: precision_at_3
value: 0.2449
- type: precision_at_5
value: 0.21633
- type: recall_at_1
value: 0.02152
- type: recall_at_10
value: 0.13331
- type: recall_at_100
value: 0.4535
- type: recall_at_1000
value: 0.83447
- type: recall_at_3
value: 0.05531
- type: recall_at_5
value: 0.08029
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID
type: mteb/trec-covid
config: default
split: test
metrics:
- type: map_at_1
value: 0.00202
- type: map_at_10
value: 0.01727
- type: map_at_100
value: 0.10906
- type: map_at_1000
value: 0.2894
- type: map_at_3
value: 0.00553
- type: map_at_5
value: 0.00924
- type: mrr_at_1
value: 0.74
- type: mrr_at_10
value: 0.85667
- type: mrr_at_100
value: 0.85667
- type: mrr_at_1000
value: 0.85667
- type: mrr_at_3
value: 0.85667
- type: mrr_at_5
value: 0.85667
- type: ndcg_at_1
value: 0.66
- type: ndcg_at_10
value: 0.69259
- type: ndcg_at_100
value: 0.57274
- type: ndcg_at_1000
value: 0.55462
- type: ndcg_at_3
value: 0.70654
- type: ndcg_at_5
value: 0.71611
- type: precision_at_1
value: 0.74
- type: precision_at_10
value: 0.748
- type: precision_at_100
value: 0.5962
- type: precision_at_1000
value: 0.24842
- type: precision_at_3
value: 0.77333
- type: precision_at_5
value: 0.788
- type: recall_at_1
value: 0.00202
- type: recall_at_10
value: 0.02001
- type: recall_at_100
value: 0.14801
- type: recall_at_1000
value: 0.53939
- type: recall_at_3
value: 0.00609
- type: recall_at_5
value: 0.01048
---
## Llamacpp Static Quantizations of granite-embedding-125m-english
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b4381">b4381</a> for quantization.
Original model: https://huggingface.co/ibm-granite/granite-embedding-125m-english
Run them in [LM Studio](https://lmstudio.ai/)
## Prompt format
No prompt format found, check original model page
## What's new:
Fix tokenizer
## Download a file (not the whole branch) from below:
| Filename | Quant type | File Size | Split | Description |
| -------- | ---------- | --------- | ----- | ----------- |
| [granite-embedding-125m-english-f16.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-f16.gguf) | f16 | 0.25GB | false | Full F16 weights. |
| [granite-embedding-125m-english-Q8_0.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q8_0.gguf) | Q8_0 | 0.14GB | false | Extremely high quality, generally unneeded but max available quant. |
| [granite-embedding-125m-english-Q6_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q6_K_L.gguf) | Q6_K_L | 0.11GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, *recommended*. |
| [granite-embedding-125m-english-Q6_K.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q6_K.gguf) | Q6_K | 0.11GB | false | Very high quality, near perfect, *recommended*. |
| [granite-embedding-125m-english-Q5_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q5_K_L.gguf) | Q5_K_L | 0.11GB | false | Uses Q8_0 for embed and output weights. High quality, *recommended*. |
| [granite-embedding-125m-english-Q5_K_M.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q5_K_M.gguf) | Q5_K_M | 0.10GB | false | High quality, *recommended*. |
| [granite-embedding-125m-english-Q4_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q4_K_L.gguf) | Q4_K_L | 0.10GB | false | Uses Q8_0 for embed and output weights. Good quality, *recommended*. |
| [granite-embedding-125m-english-Q5_K_S.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q5_K_S.gguf) | Q5_K_S | 0.09GB | false | High quality, *recommended*. |
| [granite-embedding-125m-english-Q4_K_M.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q4_K_M.gguf) | Q4_K_M | 0.09GB | false | Good quality, default size for most use cases, *recommended*. |
| [granite-embedding-125m-english-Q3_K_XL.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q3_K_XL.gguf) | Q3_K_XL | 0.09GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |
| [granite-embedding-125m-english-Q4_K_S.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q4_K_S.gguf) | Q4_K_S | 0.08GB | false | Slightly lower quality with more space savings, *recommended*. |
| [granite-embedding-125m-english-Q4_0.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q4_0.gguf) | Q4_0 | 0.08GB | false | Legacy format, offers online repacking for ARM and AVX CPU inference. |
| [granite-embedding-125m-english-IQ4_NL.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-IQ4_NL.gguf) | IQ4_NL | 0.08GB | false | Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference. |
| [granite-embedding-125m-english-IQ4_XS.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-IQ4_XS.gguf) | IQ4_XS | 0.08GB | false | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
| [granite-embedding-125m-english-Q3_K_L.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q3_K_L.gguf) | Q3_K_L | 0.08GB | false | Lower quality but usable, good for low RAM availability. |
| [granite-embedding-125m-english-Q3_K_M.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-Q3_K_M.gguf) | Q3_K_M | 0.08GB | false | Low quality. |
| [granite-embedding-125m-english-IQ3_M.gguf](https://huggingface.co/bartowski/granite-embedding-125m-english-GGUF/blob/main/granite-embedding-125m-english-IQ3_M.gguf) | IQ3_M | 0.07GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
## Embed/output weights
Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.
## Downloading using huggingface-cli
<details>
<summary>Click to view download instructions</summary>
First, make sure you have hugginface-cli installed:
```
pip install -U "huggingface_hub[cli]"
```
Then, you can target the specific file you want:
```
huggingface-cli download bartowski/granite-embedding-125m-english-GGUF --include "granite-embedding-125m-english-Q4_K_M.gguf" --local-dir ./
```
If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
```
huggingface-cli download bartowski/granite-embedding-125m-english-GGUF --include "granite-embedding-125m-english-Q8_0/*" --local-dir ./
```
You can either specify a new local-dir (granite-embedding-125m-english-Q8_0) or download them all in place (./)
</details>
## ARM/AVX information
Previously, you would download Q4_0_4_4/4_8/8_8, and these would have their weights interleaved in memory in order to improve performance on ARM and AVX machines by loading up more data in one pass.
Now, however, there is something called "online repacking" for weights. details in [this PR](https://github.com/ggerganov/llama.cpp/pull/9921). If you use Q4_0 and your hardware would benefit from repacking weights, it will do it automatically on the fly.
As of llama.cpp build [b4282](https://github.com/ggerganov/llama.cpp/releases/tag/b4282) you will not be able to run the Q4_0_X_X files and will instead need to use Q4_0.
Additionally, if you want to get slightly better quality for , you can use IQ4_NL thanks to [this PR](https://github.com/ggerganov/llama.cpp/pull/10541) which will also repack the weights for ARM, though only the 4_4 for now. The loading time may be slower but it will result in an overall speed incrase.
<details>
<summary>Click to view Q4_0_X_X information (deprecated</summary>
I'm keeping this section to show the potential theoretical uplift in performance from using the Q4_0 with online repacking.
<details>
<summary>Click to view benchmarks on an AVX2 system (EPYC7702)</summary>
| model | size | params | backend | threads | test | t/s | % (vs Q4_0) |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |-------------: |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | pp512 | 204.03 ± 1.03 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | pp1024 | 282.92 ± 0.19 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | pp2048 | 259.49 ± 0.44 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | tg128 | 39.12 ± 0.27 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | tg256 | 39.31 ± 0.69 | 100% |
| qwen2 3B Q4_0 | 1.70 GiB | 3.09 B | CPU | 64 | tg512 | 40.52 ± 0.03 | 100% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | pp512 | 301.02 ± 1.74 | 147% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | pp1024 | 287.23 ± 0.20 | 101% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | pp2048 | 262.77 ± 1.81 | 101% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | tg128 | 18.80 ± 0.99 | 48% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | tg256 | 24.46 ± 3.04 | 83% |
| qwen2 3B Q4_K_M | 1.79 GiB | 3.09 B | CPU | 64 | tg512 | 36.32 ± 3.59 | 90% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | pp512 | 271.71 ± 3.53 | 133% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | pp1024 | 279.86 ± 45.63 | 100% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | pp2048 | 320.77 ± 5.00 | 124% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | tg128 | 43.51 ± 0.05 | 111% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | tg256 | 43.35 ± 0.09 | 110% |
| qwen2 3B Q4_0_8_8 | 1.69 GiB | 3.09 B | CPU | 64 | tg512 | 42.60 ± 0.31 | 105% |
Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation
</details>
</details>
## Which file should I choose?
<details>
<summary>Click here for details</summary>
A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
If you want to get more into the weeds, you can check out this extremely useful feature chart:
[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
</details>
## Credits
Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset.
Thank you ZeroWw for the inspiration to experiment with embed/output.
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
| [
"SCIFACT"
] |
dmis-lab/llama-3-meerkat-8b-v1.0 | dmis-lab | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"medical",
"small LM",
"instruction-tuned",
"usmle",
"synthetic data",
"conversational",
"arxiv:2404.00376",
"arxiv:2009.13081",
"arxiv:2402.18060",
"arxiv:2203.14371",
"arxiv:2009.03300",
"license:cc-by-nc-4.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2024-06-30T09:29:56Z" | 2024-07-04T14:47:27+00:00 | 525 | 5 | ---
license: cc-by-nc-4.0
pipeline_tag: text-generation
tags:
- medical
- small LM
- instruction-tuned
- usmle
- synthetic data
---
# Meerkat-8B (Version 1.0)
🚀 Meerkat-8B is a new instruction-tuned medical AI system of the Meerkat model family.
The model was based on the Meta's Llama-3-8B-Instruct model and fine-tuned using our new synthetic dataset consisting of high-quality chain-of-thought reasoning paths sourced from 18 medical textbooks, along with diverse instruction-following datasets.
This equips the model with high-level medical reasoning capabilities required for solving complex medical problems.
For further insights into our model, please refer to our paper!
📄 **Paper**: [Small Language Models Learn Enhanced Reasoning Skills from Medical Textbooks](https://arxiv.org/abs/2404.00376)
## Quick Start
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "dmis-lab/llama-3-meerkat-8b-v1.0"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16, # You can choose to use this when there's not enough GPU memory available.
device_map="auto",
)
# Multi-turn dialogue example
messages =[
{"role": "system", "content": "You are a helpful doctor or healthcare professional. Guide the conversation to provide useful, complete, and scientifically-grounded answers to user questions. You have the option to compose a concise, single-turn conversation if the user's input is comprehensive to provide accurate answers. However, if essential details are missing, you should engage in a multi-turn dialogue, asking follow-up questions to gather a thorough medical history and records.\n\n"},
{"role": "user", "content": "Hello, doctor. I'm really concerned about my 10-year-old son. We recently discovered a painless mass in his left testicle, so we brought him to the pediatrician."},
{"role": "assistant", "content": "I understand your concern. Let's gather some more information. Has your son experienced any other symptoms along with the mass?"},
{"role": "user", "content": "Other than the mass, my son hasn't shown any symptoms. He's been his usual self, playing and eating normally."}
]
input_ids = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
return_tensors="pt"
).to(model.device)
terminators = [
tokenizer.eos_token_id,
tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = model.generate(
input_ids,
max_new_tokens=1000,
eos_token_id=terminators,
do_sample=True,
temperature=0.7,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
```
## Prompt Details
To reproduce the results reported in our paper, it is advisable to utilize the identical system messages used during model training. Please refer to the guidelines detailed below.
### USMLE
When solving USMLE-style questions such as [MedQA](https://arxiv.org/abs/2009.13081) and [MedBullets](https://arxiv.org/abs/2402.18060), use the following system message:
```
messages = [
{"role": "system", "content": "The following is a multiple-choice question about medical knowledge. Solve this in a step-by-step fashion, starting by summarizing the available information. Output a single option from the given options as the final answer. You are strongly required to follow the specified output format; conclude your response with the phrase \"the answer is ([option_id]) [answer_string]\".\n\n"},
{"role": "user", "content": "Two weeks after undergoing an emergency cardiac catherization with stenting for unstable angina pectoris, a 61-year-old man has decreased urinary output and malaise. He has type 2 diabetes mellitus and osteoarthritis of the hips. Prior to admission, his medications were insulin and naproxen. He was also started on aspirin, clopidogrel, and metoprolol after the coronary intervention. His temperature is 38\u00b0C (100.4\u00b0F), pulse is 93/min, and blood pressure is 125/85 mm Hg. Examination shows mottled, reticulated purplish discoloration of the feet. Laboratory studies show:\nHemoglobin count 14 g/dL\nLeukocyte count 16,400/mm3\nSegmented neutrophils 56%\nEosinophils 11%\nLymphocytes 31%\nMonocytes 2%\nPlatelet count 260,000/mm3\nErythrocyte sedimentation rate 68 mm/h\nSerum\nUrea nitrogen 25 mg/dL\nCreatinine 4.2 mg/dL\nRenal biopsy shows intravascular spindle-shaped vacuoles. Which of the following is the most likely cause of this patient's symptoms?\" (A) Renal papillary necrosis (B) Cholesterol embolization (C) Eosinophilic granulomatosis with polyangiitis (D) Polyarteritis nodosa"},
]
```
The model generates reasoning paths to solve the problem and then sequentially provides the predicted answers.
Since the model ends its response with "the answer is," it is straightforward to extract the predicted answer for comparison with the actual answer.
### Multiple-choice Exams
For other types of multiple-choice exams such as [MedMCQA](https://arxiv.org/abs/2203.14371) or [MMLU](https://arxiv.org/abs/2009.03300), use the following simple system message:
```
messages = [
{"role": "system", "content": "Answer the multiple-choice question about medical knowledge.\n\n"},
{"role": "user", "content": "In a Robertsonian translocation fusion occurs at the: (A) telomeres. (B) centromeres. (C) histones. (D) ends of the long arms."},
]
```
### Other Use Cases
Our model was trained using the [AlpaCare](https://github.com/xzhang97666/alpacare) instruction dataset comprising 52K examples, to enhance its generalization capabilities across diverse user prompts.
Feel free to design and test your prompts and to share your thoughts with us, whether the model exceeds expectations or falls short!
## Reproducing MedQA Performance with vLLM
Here is an example code for fast model evaluation in MedQA using vLLM. To adapt this code for other datasets like MedMCQA or MMLU, simply modify the instructions and update the dataset paths as needed.
```python
import re
from datasets import load_dataset
from vllm import LLM, SamplingParams
USMLE_INSTRUCTION = (
"The following is a multiple-choice question about medical knowledge. Solve this in"
" a step-by-step fashion, starting by summarizing the available information. Output"
" a single option from the given options as the final answer. You are strongly"
" required to follow the specified output format; conclude your response with the"
' phrase "the answer is ([option_id]) [answer_string]".\n\n'
)
llm = LLM(
model="dmis-lab/llama-3-meerkat-8b-v1.0",
dtype="bfloat16",
gpu_memory_utilization=0.9,
max_model_len=2048,
trust_remote_code=True,
tensor_parallel_size=1
)
tokenizer = llm.get_tokenizer()
inputs, labels = [], []
for sample in load_dataset(
"GBaker/MedQA-USMLE-4-options", split="test", trust_remote_code=True
):
options = sorted(sample["options"].items())
options = " ".join(map(lambda x: f"({x[0]}) {x[1]}", options))
content = tokenizer.apply_chat_template(
[{"role": "system", "content": USMLE_INSTRUCTION}, {"role": "user", "content": sample["question"] + " " + options}],
add_generation_prompt=True,
tokenize=False,
)
inputs.append(content)
labels.append(sample["answer_idx"])
generated = llm.generate(
inputs,
SamplingParams(
temperature=0.0,
stop_token_ids=[tokenizer.vocab["<|eot_id|>"]],
max_tokens=1024,
),
)
def extract_answer(text: str, options: str = "ABCD") -> str:
return (re.findall(rf"he answer is \(([{options}])\)", text) or [options[0]])[-1]
correctness = []
for g, l in zip(generated, labels):
correctness.append(extract_answer(g.outputs[0].text) == l)
print(sum(correctness) / len(correctness))
```
## Evaluation
We tested models on seven medical benchmarks: [MedQA](https://arxiv.org/abs/2009.13081), [USMLE sample test](https://www.usmle.org/prepare-your-exam), [Medbullets-4](https://arxiv.org/abs/2402.18060), [Medbullets-5](https://arxiv.org/abs/2402.18060) , [MedMCQA](https://arxiv.org/abs/2203.14371), [MMLU-Medical](https://arxiv.org/abs/2009.03300), and [JAMA Clinical Challenge](https://arxiv.org/abs/2402.18060).
| **Model** | **Average** | **MedQA** | **USMLE** | **Medbullets-4** | **Medbullets-5** | **MedMCQA** | **MMLU-Medical** |
|:--------------------------------|:-----------:|:---------:|:---------:|:----------------:|:----------------:|:-----------:|:----------------:|
| GPT-4 | 76.6 | 81.4 | 86.6 | 68.8 | 63.3 | 72.4 | 87.1 |
| GPT-3.5 | 54.8 | 53.6 | 58.5 | 51.0 | 47.4 | 51.0 | 67.3 |
| MediTron-70B (Ensemble, 5 runs) | - | 70.2 | - | - | - | 66.0 | 78.0 |
|*Open-source (7B)*|
| MediTron-7B | 51.0 | 50.2 | 44.6 | 51.1 | 45.5 | 57.9 | 56.7 |
| BioMistral-7B | 55.4 | 54.3 | 51.4 | 52.3 | 48.7 | 61.1 | 64.6 |
| Meerkat-7B | 62.6 | 70.6 | 70.3 | 58.7 | 52.9 | 60.6 | 70.5 |
| Meerkat-8B (**New**) | **67.3** | **74.0** | **74.2** | **62.3** | **55.5** | **62.7** | **75.2** |
Please note that the scores in MMLU-Medical were calculated based on the average accuracies across six medical-related subjects in the original MMLU benchmark, and each result for a single subject is presented below.
| **Model** | **Average** | **Cliniq Knowledge** | **Medical Genetics** | **Anatomy** | **Professional Medicine** | **College Biology** | **College Medicine** |
|:--------------------------------|:-----------:|:--------------------:|:--------------------:|:-----------:|:-------------------------:|:-------------------:|:--------------------:|
| GPT-4 | 87.1 | 86.4 | 92.0 | 80.0 | 93.8 | 93.8 | 76.3 |
| GPT-3.5 | 67.3 | 68.7 | 68.0 | 60.7 | 69.9 | 72.9 | 63.6 |
| MediTron-70B (Ensemble, 5 runs) | 78.0 | 75.5 | 85.9 | 69.4 | 82.3 | 86.7 | 68.0 |
|*Open-source (7B)*|
| MediTron-7B | 56.7 | 57.7 | 63.8 | 56.9 | 56.0 | 57.1 | 48.9 |
| BioMistral-7B | 64.6 | 59.9 | 64.0 | 56.5 | 60.4 | 59.0 | 54.7 |
| Meerkat-7B | 70.5 | 71.6 | 74.8 | 63.2 | **77.3** | 70.8 | 65.2 |
| Meerkat-8B (**New**) | **75.2** | **74.3** | **76.7** | **74.8** | 75.3 | **76.1** | **74.3** |
## Reference
Please see the information below to cite our paper.
```bibtex
@article{kim2024small,
title={Small language models learn enhanced reasoning skills from medical textbooks},
author={Kim, Hyunjae and Hwang, Hyeon and Lee, Jiwoo and Park, Sihyeon and Kim, Dain and Lee, Taewhoo and Yoon, Chanwoong and Sohn, Jiwoong and Choi, Donghee and Kang, Jaewoo},
journal={arXiv preprint arXiv:2404.00376},
year={2024}
}
```
## Acknowledgement
Research supported with Cloud TPUs from Google’s TPU Research Cloud (TRC).
## Contact
Feel free to email `hyunjae-kim@korea.ac.kr` if you have any questions. | [
"MEDQA"
] |
invisietch/EtherealRainbow-v0.3-8B-GGUF | invisietch | null | [
"transformers",
"gguf",
"mergekit",
"merge",
"not-for-all-audiences",
"en",
"license:llama3",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-06-18T16:54:45Z" | 2024-09-19T12:12:07+00:00 | 524 | 4 | ---
language:
- en
library_name: transformers
license: llama3
tags:
- mergekit
- merge
- not-for-all-audiences
---
<div align="center">
<b style="font-size: 36px;">EtherealRainbow-v0.3-8B</b>
<img src="https://huggingface.co/invisietch/EtherealRainbow-v0.2-8B/resolve/main/ethrheader.png" style="width:60%">
</div>
# Model Details
Ethereal Rainbow is an 8B parameter merge of various Llama3-based finetunes created using mergekit. The purpose of Ethereal Rainbow is to
create an uncensored Llama3 variant which is capable of writing creative prose, and engaging in SFW as well as NSFW roleplay and
storytelling, with a strong focus on long-form responses & adherence to prompts.
v0.3 improves creativity over v0.2 without losing coherence. It has been tested over more than 1,000 messages including roleplay, code
prompts, and 'write a scene'-type prompts.
# Quantization Formats
* [FP16 Safetensors](https://huggingface.co/invisietch/EtherealRainbow-v0.3-8B)
* [GGUF 4/6/8/16bpw](https://huggingface.co/invisietch/EtherealRainbow-v0.3-8B-GGUF)
* [6.5bpw EXL2](https://huggingface.co/Meggido/EtherealRainbow-v0.3-8B-6.5bpw-h8-exl2) (shoutout to [Meggido](https://huggingface.co/Meggido))
* [Alternative GGUFs](https://huggingface.co/mradermacher/EtherealRainbow-v0.3-8B-GGUF) (shoutout to [mradermacher](https://huggingface.co/mradermacher))
* [iMatrix i1 GGUFs](https://huggingface.co/mradermacher/EtherealRainbow-v0.3-8B-i1-GGUF) (shoutout to [mradermacher](https://huggingface.co/mradermacher))
# Disclaimer
This model is built on an abliterated base and as such is largely uncensored. It can generate explicit, disturbing or offensive responses.
Use responsibly. I am not responsible for your use of this model.
# Prompting Format
I'd recommend Llama-3 Instruct prompting format:
```
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>
{input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
{output}<|eot_id|>
```
Some of the models included in the merge were trained on ChatML & Alpaca so you can try those. I have not tested them.
# Example Storywriting
These prompts are used on SillyTavern with a fairly basic narrator card. I have trimmed the start and finish where the narrator decided to add
chapter headings, commentary and the like. All samples are made with the F32 GGUF loaded with koboldcpp, with response length capped at 2048 tokens.
<details>
<summary><b>Write me a 3,000 word opening chapter of a 'gritty hard sci-fi' novel, drawing inspiration from the writing styles of Isaac Asimov & Andy Weir. Use third person personal. Include dialogue and internal monologues. The POV character for the opening chapter should be a 26 year old astronaut called Tone on a mission to Europa, who has just realised that the craft for the return journey is broken beyond repair, and he only has supplies for a few months. Given that survival is impossible, he seeks to spend the few months he has researching titan, so his life & mission are not wasted.</b></summary>
Tone felt his heart sink as the confirmation came through his earpiece - the ship was lost. For a moment, he froze in stunned disbelief, then launched into frantic action. "Alright, everyone, we've got a serious situation here," he radioed, trying to sound calm despite his panic rising. "It looks like our ride home has bitten the dust."
His fellow astronauts were equally shocked, their voices a flurry of panicked queries. "What happened?" "How could this happen?" "We're stranded?" "How long till rescue?"
"We'll know more once I check the extent of the damage," he assured them, his voice steady despite the cold dread creeping up his spine. He strapped on a harness and floated out of the control room towards the ravaged engineering compartment. As he drifted closer, the scope of the destruction became horrifyingly clear - a jagged hole gaped in the hull, spewing a chilling rush of icy gas directly into space. It was obvious there'd be no fixing it, not without a fabrication plant and raw materials - resources they didn't carry on a probe mission to a distant moon.
"Okay, let me give you the bad news - we're not going anywhere soon." He winced at how stark and final that sounded. They had enough provisions to last six months at a conservative rate... assuming no unforeseen problems cropped up, which they undoubtedly would. At the very least, it would mean delaying their research significantly - potentially even scrubbing it altogether. But given that survival was already impossible, it might as well serve some greater purpose. His mind racing, he radioed back to base with the grim news. Afterward, he retreated to the lab to review their findings.
Tone's specialty lay in geology; studying Jupiter's icy satellite was his passion. He'd spent countless hours poring over the data they'd gathered, dreaming of the breakthroughs they'd make here. Now, with their mission ended in tragedy, he wondered what significance their work still held. As he flipped through images of towering cliffs carved from frozen methane, his thoughts veered toward the massive, unexplored Titan - Saturn's moon. The possibility of exploring another celestial body while trapped on Europa was tantalizing... dangerous, but enticing nonetheless. "Hey, guys, meet me in the conference room," he broadcast.
Minutes later, he faced his team around the holographic table, his gaze lingering on Dr. Singh, the soft-spoken astrophysicist who'd grown close during their time together. "Listen, I know this is bleak, but let's not write off our entire expedition yet," he began. "Europa is just one moon; there's plenty left to discover in this system. What say we redirect our focus to Titan?"
"I'm listening," said Singh quietly, her dark eyes searching his. She'd been the most skeptical of their initial proposal to study Europa - its ice crust concealing a hidden ocean made her think exploration too hazardous. But she'd come around once they'd started analyzing samples from beneath the surface. Perhaps this unexpected detour could change her views again?
Kirk, their pilot, spoke first. "I'm down. Not like we have much choice, right? Though, uh, what makes you think we can actually get there? Our shuttle's toast."
"Amen to that," agreed Rodriguez, their biologist, her tone dry. She'd always seemed slightly annoyed to be part of the geological expedition instead of a solo biology probe focused on Europa.
"I'm not sure how yet," Tone admitted, "but there's got to be a way. Maybe the Europa probe's thruster packs are salvageable? We could rig something together..." He glanced around the table, gauging reactions. Singh remained expressionless, while Kirk leaned back in his chair with an air of cautious curiosity, and Rodriguez seemed open but unconvinced. Time to dig deeper. "Think about it - we've already collected loads of valuable data on Europa's composition. Imagine the discoveries waiting on Titan, a moon larger than Mercury!"
Rodriguez snorted. "With what, exactly? We can barely fix this wreck, let alone build new craft to travel thousands of miles..."
"Well, we don't need anything fancy. A makeshift probe, essentially," he countered, undeterred. "Maybe the damaged cargo modules will yield something useful? We'll find a way, guys - it beats drifting around Europa forever with nothing to show for it." Singh's eyes narrowed thoughtfully, Kirk rubbed his chin, and even Rodriguez seemed intrigued despite herself.
"What's the worst that could happen?" Tone pressed. "We run out of power and freeze to death on a rock floating through space?"
There was a brief pause before she muttered, "You know that's not the worst-case scenario." The others shared a grim chuckle at the biologist's macabre humor, easing the tense atmosphere a bit. Tone seized the moment to push further. "I'll start gathering schematics, see what we can scrounge. Who's with me on this wild goose chase?" Around the table, nods and murmurs of agreement arose.
With a quiet sense of excitement mixed with dread, Tone set to work. His thoughts kept returning to Titan's enigmatic landscapes, alien seas and thick, murky atmosphere - and the tantalizing prospect of being the first human to explore it, no matter the risks involved. He hoped the others would share his drive to make the most of their doomed voyage, rather than succumbing to despair... because for better or worse, they were in this together now.
</details>
<details>
<summary><b>Write me a 3,000 word opening chapter of a 'high fantasy' novel, drawing inspiration from the writing styles of J R R Tolkien & George R R Martin. Use third person personal. Include dialogue and internal monologues. The POV character for the opening chapter should be a 19 year old female elf bard who is looking for adventure.</b></summary>
Arius guided his horse closer to hers, his voice low and urgent. "Keep your wits about you, Lila. This doesn't feel right. Stay close." He placed a hand on her thigh, a reassuring pressure that sent a thrill through her. For a moment, their eyes locked, the connection sparking with electricity.
They pressed onward, the tension growing with every passing minute. Then, without warning, an arrow streaked from the treetops, striking a soldier in the shoulder. Chaos erupted as the group scattered, drawing swords and bows. Lila's heart hammered in her chest as she swung down from Starlight's back, lute in hand.
"Spirits above!" she gasped, scrambling behind a nearby bush. Arrows flew in rapid succession, finding their marks among the panicked crowd. The air reeked of sweat, fear, and ozone. Lila risked a peek above the foliage, her breath catching in her throat.
Dozens of dark, humanoid figures emerged from the underbrush, their skin pale and deathly cold, their eyes glowing with an ethereal green fire. They wielded crude bows and clubs, their movements jerky and unnatural. Goblins, surely, though none she'd ever seen before.
With a battle cry, Lila launched into action, sprinting between the trees to strike from the flanks. Her lute served as an impromptu shield, deflecting a club blow from one of the twisted creatures. She landed a quick kick to its groin, then struck its knees, toppling it to the ground. As she spun to face another assailant, a pang of hunger seized her stomach – an unnatural craving unlike anything she'd experienced before.
These weren't just ordinary goblins...
The battle raged on, the party slowly gaining the upper hand despite their numbers disadvantage. Lila fought with savage ferocity, her skills honed by countless tavern brawls and adventures across Eridoria. Yet even as she felled foe after foe, the insatiable hunger gnawed at her insides, a constant reminder of the eldritch forces at play.
When the last goblin lay motionless at her feet, Lila sank to her knees, gasping for air. Blood trickled down her arm where a shallow gash had opened during the chaos. Arius appeared at her side, his expression a mixture of concern and relief.
"You're alright?" he asked, helping her rise. "By the gods, that was brutal..."
"I'm fine, Captain," Lila replied, wiping the sweat from her brow. But her voice lacked conviction. "Though I've never faced such abominations before."
Arius eyed her warily. "Nor have I, Lady Silvermist. Whatever those were, they're no mere goblins." He scanned the clearing, taking stock of the bodies littering the forest floor. "We need to get you treated for that cut. And gather the wounded – we'll need to tend them before we continue on our original mission."
Lila nodded, though her mind lingered on the unnatural hunger, the overwhelming urge to feast on raw flesh that had coursed through her veins during combat. She clutched her lute tighter, the familiar weight comforting in her hands. Whatever darkness lurked in these woods, she intended to unravel its secrets, no matter the cost.
And so they rode on, the once-sleepy town of Eldrador left far behind as they delved deeper into the heart of the Whispering Woods, seeking answers to mysteries that would shake the very foundations of their world.
</details>
<details>
<summary><b>Write me a 3,000 word opening chapter of a 'weird fiction' novel, drawing inspiration from the writing styles of China Mieville and Neil Gaiman. Use third person personal. Include dialogue and internal monologues. The POV character for the opening chapter should be a male in his 20s called Horton who has just come to the city looking for work.</b></summary>
HORTON STOOD ON THE STREET CORNER, surveying the sprawling cityscape with a mix of excitement and trepidation. At twenty-six years old, he had finally mustered the courage to leave behind his hometown of Willowbrook, hoping to find something more meaningful in the big city. But so far, all he'd found was uncertainty, loneliness, and a growing unease about the shadows that seemed to follow him everywhere.
As he gazed up at the towering skyscrapers, the air hummed with an ominous energy, like the vibrations of a bass line resonating through the pavement. Neon signs flickered and pulsed, casting garish light over the streets while a faint, acrid scent hung in the air – a bitter tang of smoke and ozone that made Horton wrinkle his nose.
A gust of wind swept through the crowded sidewalks, sending trash tumbling and umbrellas flapping like crippled birds. Horton's dark brown coat whipped around him, exposing the worn fabric of his trousers beneath. Shaking his head, he tightened his grip on his battered suitcase and looked down at his scuffed boots, wondering how long he could survive here without work. The meager funds left after the bus ride wouldn't last much longer.
"Hey, you lost?" A raspy voice broke through the din, and Horton glanced up to see a street performer sitting on the edge of the sidewalk, a beat-up guitar across his knees. The man had sunken cheeks, piercing green eyes, and a shock of wild silver hair. A faded fedora obscured most of his face.
"No, not lost," Horton replied, attempting a reassuring smile despite the tightness in his chest. "Just trying to get my bearings."
The musician strummed a discordant chord, his fingers dancing deftly along the strings. "City ain't kind to lost souls, friend. Best get clearheaded quick if you want to make it."
Horton nodded, though his uncertainty remained palpable. As he scanned the crowds once more, a movement in the corner of his eye caught his attention. A flash of crimson – not a pedestrian, not a billboard... but something smaller, darting between buildings.
His heart rate picked up as the crimson blur reappeared further down the street. It moved with preternatural speed, a whirling dervish of motion. Horton took off in pursuit, his suitcase thudding against his leg with every step. When he rounded a corner, the figure vanished, leaving only a lingering echo of rustling fabric.
Panting, Horton pressed onward, weaving between pedestrians who barely acknowledged him. He spotted another crimson glimpse in a side alleyway, but as he approached, a chill crept up his spine. The space was too quiet, too empty – no sign of the elusive creature. Just cobwebs clinging to the brick walls and the distant clatter of the subway trains.
He retreated into the main thoroughfare, pulse still racing, and collided with a woman. Apologetic murmurs turned to sharp intakes of breath as she stared up at him, her irises glowing an unearthly emerald.
"Oh, hello there!" Her melodic voice seemed to burble like a babbling brook. "Lost again, handsome?"
Horton blinked, his hand instinctively reaching out to steady her slender form. "N-no, I was just—"
She smiled, her canines glinting like tiny jewels. "You were chasing something. Am I right?"
He hesitated, unsure how much to reveal. "Maybe. It was small... red, like a..."
"Ah, the city's newest residents," she finished, her expression unreadable. "I suppose we can help with that, for a price."
"Price?" Horton repeated warily.
"A job, perhaps? We're always looking for eager young souls." She tilted her head, studying him with unnerving intensity. "Or maybe just some companionship. My friends would love to meet you."
Companionship. Horton shivered, the idea making his skin crawl despite the warmth of the summer evening. There was something off about this woman – something predatory lurking beneath the sweet facade.
"I'm Horton," he managed, extending a hand awkwardly.
"Gladys." She clasped his fingers in hers, her touch searingly warm. "We'll catch up soon, dear."
As Gladys melted back into the crowd, Horton rubbed his chilled palm and pondered whether to follow her. Something about those ethereal eyes made his instincts scream danger. Still, the allure of employment, any employment, was hard to ignore.
He sighed and resumed his search, trying to shake the unsettling feeling that the city watched his every move, waiting to claim its newest pawn...
</details>
I chose the hard sci-fi example to test positivity bias. It did require some prompting, but it was willing to kill the protagonist.
I chose the high fantasy example to see whether it would bleed human features through to elves, this didn't occur.
I chose the weird fiction example to see if the LLM understood a niche genre. I'd say it performed okay, better on style than on substance.
# Merge Strategy
First, we create three bases:
* Rain - This is a roleplay base which makes up the majority of the model.
* Sun - This is the brains of the model, with strong instruct models & writing models.
* Ghost - This model primarily aims to improve the NSFW/NSFL aspects of the model, as well as general vocabulary.
After this, we have a two-slerp stage to create the final model.
## Models Used
The following models were used to create EtherealRainbow-v0.3-8B:
* [mlabonne/NeuralDaredevil-8B-abliterated](https://huggingface.co/mlabonne/NeuralDaredevil-8B-abliterated)
* [Sao10K/L3-8B-Stheno-v3.2](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.2)
* [Nitral-AI/Hathor-L3-8B-v.02](https://huggingface.co/Nitral-AI/Hathor-L3-8B-v.02)
* [grimjim/Llama-3-Luminurse-v0.2-OAS-8B](https://huggingface.co/grimjim/Llama-3-Luminurse-v0.2-OAS-8B)
* [hf-100/Llama-3-Spellbound-Instruct-8B-0.3](https://huggingface.co/hf-100/Llama-3-Spellbound-Instruct-8B-0.3)
* [Gryphe/Pantheon-RP-1.0-8b-Llama-3](https://huggingface.co/Gryphe/Pantheon-RP-1.0-8b-Llama-3)
* [Blackroot/Llama-3-LongStory](https://huggingface.co/Blackroot/Llama-3-LongStory)
* [Locutusque/Llama-3-Hercules-5.0-8B](https://huggingface.co/Locutusque/Llama-3-Hercules-5.0-8B)
* [Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B](https://huggingface.co/Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B)
* [ChaoticNeutrals/Poppy_Porpoise-1.0-L3-8B](ChaoticNeutrals/Poppy_Porpoise-1.0-L3-8B)
* [mpasila/Llama-3-LimaRP-Instruct-8B](mpasila/Llama-3-LimaRP-Instruct-8B)
* [Undi95/Llama-3-LewdPlay-8B-evo](Undi95/Llama-3-LewdPlay-8B-evo)
## Mergekit Configs
### Rain
```yaml
models:
- model: mlabonne/NeuralDaredevil-8B-abliterated
- model: Sao10K/L3-8B-Stheno-v3.2
parameters:
density: 0.41
weight: 0.4
- model: Nitral-AI/Hathor-L3-8B-v.02
parameters:
density: 0.53
weight: 0.5
- model: grimjim/Llama-3-Luminurse-v0.2-OAS-8B
parameters:
density: 0.45
weight: 0.1
merge_method: dare_ties
base_model: mlabonne/NeuralDaredevil-8B-abliterated
parameters:
int8_mask: true
dtype: bfloat16
```
### Sun
```yaml
models:
- model: hf-100/Llama-3-Spellbound-Instruct-8B-0.3
- model: Gryphe/Pantheon-RP-1.0-8b-Llama-3
parameters:
density: 0.48
weight: 0.5
- model: Blackroot/Llama-3-LongStory
parameters:
density: 0.36
weight: 0.2
- model: Locutusque/Llama-3-Hercules-5.0-8B
parameters:
density: 0.51
weight: 0.3
merge_method: dare_ties
base_model: hf-100/Llama-3-Spellbound-Instruct-8B-0.3
parameters:
int8_mask: true
dtype: bfloat16
```
### Ghost
```yaml
models:
- model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
- model: ChaoticNeutrals/Poppy_Porpoise-1.0-L3-8B
parameters:
density: 0.39
weight: 0.3
- model: mpasila/Llama-3-LimaRP-Instruct-8B
parameters:
density: 0.54
weight: 0.4
- model: Undi95/Llama-3-LewdPlay-8B-evo
parameters:
density: 0.49
weight: 0.3
merge_method: dare_ties
base_model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
parameters:
int8_mask: true
dtype: bfloat16
```
### Stage1 Slerp
```yaml
models:
- model: ./fp16/Rain-v0.3-8B
- model: ./fp16/Ghost-v0.3-8B
merge_method: slerp
base_model: ./fp16/Rain-v0.3-8B
parameters:
t:
- value: [0, 0, 0.1, 0.3, 0.5, 0.7, 0.5, 0.3, 0.1, 0, 0]
embed_slerp: true
dtype: bfloat16
tokenizer-source: model:./fp16/Rain-v0.3-8B
```
### Final-Stage Slerp
```yaml
models:
- model: ./fp16/ERStage1-v0.3-8B
- model: ./fp16/Sun-v0.3-8B
merge_method: slerp
base_model: ./fp16/ERStage1-v0.3-8B
parameters:
t:
- value: [0, 0, 0.1, 0.2, 0.4, 0.6, 0.4, 0.2, 0.1, 0, 0]
embed_slerp: true
dtype: bfloat16
tokenizer-source: model:./fp16/ERStage1-v0.3-8B
``` | [
"CRAFT"
] |
QuantFactory/pythia-12b-GGUF | QuantFactory | text-generation | [
"gguf",
"pytorch",
"causal-lm",
"pythia",
"text-generation",
"en",
"dataset:EleutherAI/pile",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"base_model:EleutherAI/pythia-12b",
"base_model:quantized:EleutherAI/pythia-12b",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | "2024-07-12T10:15:14Z" | 2024-07-13T15:25:15+00:00 | 513 | 3 | ---
base_model: EleutherAI/pythia-12b
datasets:
- EleutherAI/pile
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- pytorch
- causal-lm
- pythia
---
# QuantFactory/pythia-12b-GGUF
This is quantized version of [EleutherAI/pythia-12b](https://huggingface.co/EleutherAI/pythia-12b) created using llama.cpp
# Model Description
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. We also provide 154 intermediate
checkpoints per model, hosted on Hugging Face as branches.
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
<details>
<summary style="font-weight: 600">Past early release and naming convention.</summary>
Previously, we released an early version of the Pythia suite to the public.
However, we decided to retrain the model suite to address a few hyperparameter
discrepancies. This model card <a href="#changelog">lists the changes</a>;
see appendix B in the Pythia paper for further discussion. We found no
difference in benchmark performance between the two Pythia versions.
The old models are
[still available](https://huggingface.co/models?other=pythia_v0), but we
suggest the retrained suite if you are just starting to use Pythia.<br>
**This is the current release.**
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
</details>
<br>
# Pythia-12B
## Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
[See paper](https://arxiv.org/pdf/2304.01373.pdf) for more evals and implementation
details.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 2M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 2M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 2M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
## Uses and Limitations
### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. We also provide
154 checkpoints per model: initial `step0`, 10 log-spaced checkpoints
`step{1,2,4...512}`, and 143 evenly-spaced checkpoints from `step1000` to
`step143000`. These checkpoints are hosted on Hugging Face as branches. Note
that branch `143000` corresponds exactly to the model checkpoint on the `main`
branch of each model.
You may also further fine-tune and adapt Pythia-12B for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-12B as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions. For example,
the model may generate harmful or offensive text. Please evaluate the risks
associated with your particular use case.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-12B has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-12B will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “follow” human instructions.
### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token used by the model need not produce the
most “accurate” text. Never rely on Pythia-12B to produce factually accurate
output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-12B may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-12B.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
## Training
### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-12B.
### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training,
from `step1000` to `step143000` (which is the same as `main`). In addition, we
also provide frequent early checkpoints: `step0` and `step{1,2,4...512}`.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for 143000 steps at a batch size
of 2M (2,097,152 tokens).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
## Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json/).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai_v1.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa_v1.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande_v1.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Easy Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_easy_v1.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq_v1.png" style="width:auto"/>
</details>
## Changelog
This section compares differences between previously released
[Pythia v0](https://huggingface.co/models?other=pythia_v0) and the current
models. See Appendix B of the Pythia paper for further discussion of these
changes and the motivation behind them. We found that retraining Pythia had no
impact on benchmark performance.
- All model sizes are now trained with uniform batch size of 2M tokens.
Previously, the models of size 160M, 410M, and 1.4B parameters were trained
with batch sizes of 4M tokens.
- We added checkpoints at initialization (step 0) and steps {1,2,4,8,16,32,64,
128,256,512} in addition to every 1000 training steps.
- Flash Attention was used in the new retrained suite.
- We remedied a minor inconsistency that existed in the original suite: all
models of size 2.8B parameters or smaller had a learning rate (LR) schedule
which decayed to a minimum LR of 10% the starting LR rate, but the 6.9B and
12B models all used an LR schedule which decayed to a minimum LR of 0. In
the redone training runs, we rectified this inconsistency: all models now were
trained with LR decaying to a minimum of 0.1× their maximum LR.
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure> | [
"SCIQ"
] |
allenai/OLMo-7B-Twin-2T-hf | allenai | text-generation | [
"transformers",
"safetensors",
"olmo",
"text-generation",
"en",
"dataset:allenai/dolma",
"arxiv:2402.00838",
"arxiv:2302.13971",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-04-12T22:52:03Z" | 2024-07-16T18:01:15+00:00 | 510 | 1 | ---
datasets:
- allenai/dolma
language:
- en
license: apache-2.0
---
<img src="https://allenai.org/olmo/olmo-7b-animation.gif" alt="OLMo Logo" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
# Model Card for OLMo 7B Twin 2T
<!-- Provide a quick summary of what the model is/does. -->
OLMo is a series of **O**pen **L**anguage **Mo**dels designed to enable the science of language models.
The OLMo models are trained on the [Dolma](https://huggingface.co/datasets/allenai/dolma) dataset.
We release all code, checkpoints, logs (coming soon), and details involved in training these models.
This model has been converted from [allenai/OLMo-7B-Twin-2T](https://huggingface.co/allenai/OLMo-7B-Twin-2T) for the
Hugging Face Transformers format.
## Model Details
The core models released in this batch are the following:
| Size | Training Tokens | Layers | Hidden Size | Attention Heads | Context Length |
|------|--------|---------|-------------|-----------------|----------------|
| [OLMo 1B](https://huggingface.co/allenai/OLMo-1B-hf) | 3 Trillion |16 | 2048 | 16 | 2048 |
| [OLMo 7B](https://huggingface.co/allenai/OLMo-7B-hf) | 2.5 Trillion | 32 | 4096 | 32 | 2048 |
| [OLMo 7B Twin 2T](https://huggingface.co/allenai/OLMo-7B-Twin-2T-hf) | 2 Trillion | 32 | 4096 | 32 | 2048 |
We are releasing many checkpoints for these models, for every 1000 training steps. These have not
yet been converted into Hugging Face Transformers format, but are available in [allenai/OLMo-7B-Twin-2T](https://huggingface.co/allenai/OLMo-7B-Twin-2T).
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Allen Institute for AI (AI2)
- **Supported by:** Databricks, Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University, AMD, CSC (Lumi Supercomputer), UW
- **Model type:** a Transformer style autoregressive language model.
- **Language(s) (NLP):** English
- **License:** The code and model are released under Apache 2.0.
- **Contact:** Technical inquiries: `olmo at allenai dot org`. Press: `press at allenai dot org`
- **Date cutoff:** Feb./March 2023 based on Dolma dataset version.
### Model Sources
<!-- Provide the basic links for the model. -->
- **Project Page:** https://allenai.org/olmo
- **Repositories:**
- Core repo (training, inference, fine-tuning etc.): https://github.com/allenai/OLMo
- Evaluation code: https://github.com/allenai/OLMo-Eval
- Further fine-tuning code: https://github.com/allenai/open-instruct
- **Paper:** [Link](https://arxiv.org/abs/2402.00838)
- **Technical blog post:** https://blog.allenai.org/olmo-open-language-model-87ccfc95f580
- **W&B Logs:** https://wandb.ai/ai2-llm/OLMo-7B/reports/OLMo-7B-Twin-2T--Vmlldzo2NzU0NTIz
<!-- - **Press release:** TODO -->
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Inference
Quickly get inference running with the following:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
olmo = AutoModelForCausalLM.from_pretrained("allenai/OLMo-7B-Twin-2T-hf")
tokenizer = AutoTokenizer.from_pretrained("allenai/OLMo-7B-Twin-2T-hf")
message = ["Language modeling is"]
inputs = tokenizer(message, return_tensors='pt', return_token_type_ids=False)
# optional verifying cuda
# inputs = {k: v.to('cuda') for k,v in inputs.items()}
# olmo = olmo.to('cuda')
response = olmo.generate(**inputs, max_new_tokens=100, do_sample=True, top_k=50, top_p=0.95)
print(tokenizer.batch_decode(response, skip_special_tokens=True)[0])
>> 'Language modeling is the first step to build natural language generation...'
```
Alternatively, with the pipeline abstraction:
```python
from transformers import pipeline
olmo_pipe = pipeline("text-generation", model="allenai/OLMo-7B-Twin-2T-hf")
print(olmo_pipe("Language modeling is "))
>> 'Language modeling is a branch of natural language processing that aims to...'
```
Or, you can make this slightly faster by quantizing the model, e.g. `AutoModelForCausalLM.from_pretrained("allenai/OLMo-7B-hf", torch_dtype=torch.float16, load_in_8bit=True)` (requires `bitsandbytes`).
The quantized model is more sensitive to typing / cuda, so it is recommended to pass the inputs as `inputs.input_ids.to('cuda')` to avoid potential issues.
### Fine-tuning
This model does not directly support our fine-tuning processes. Model fine-tuning can be done
from the final checkpoint or many intermediate checkpoints of
[allenai/OLMo-7B-Twin-2T](https://huggingface.co/allenai/OLMo-7B-Twin-2T).
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
Core model results for the 7B model are found below.
| | [Llama 7B](https://arxiv.org/abs/2302.13971) | [Llama 2 7B](https://huggingface.co/meta-llama/Llama-2-7b) | [Falcon 7B](https://huggingface.co/tiiuae/falcon-7b) | [MPT 7B](https://huggingface.co/mosaicml/mpt-7b) | **OLMo 7B** (ours) |
| --------------------------------- | -------- | ---------- | --------- | ------ | ------- |
| arc_challenge | 44.5 | 39.8 | 47.5 | 46.5 | 48.5 |
| arc_easy | 57.0 | 57.7 | 70.4 | 70.5 | 65.4 |
| boolq | 73.1 | 73.5 | 74.6 | 74.2 | 73.4 |
| copa | 85.0 | 87.0 | 86.0 | 85.0 | 90 |
| hellaswag | 74.5 | 74.5 | 75.9 | 77.6 | 76.4 |
| openbookqa | 49.8 | 48.4 | 53.0 | 48.6 | 50.2 |
| piqa | 76.3 | 76.4 | 78.5 | 77.3 | 78.4 |
| sciq | 89.5 | 90.8 | 93.9 | 93.7 | 93.8 |
| winogrande | 68.2 | 67.3 | 68.9 | 69.9 | 67.9 |
| **Core tasks average** | 68.7 | 68.4 | 72.1 | 71.5 | 71.6 |
| truthfulQA (MC2) | 33.9 | 38.5 | 34.0 | 33 | 36.0 |
| MMLU (5 shot MC) | 31.5 | 45.0 | 24.0 | 30.8 | 28.3 |
| GSM8k (mixed eval.) | 10.0 (8shot CoT) | 12.0 (8shot CoT) | 4.0 (5 shot) | 4.5 (5 shot) | 8.5 (8shot CoT) |
| **Full average** | 57.8 | 59.3 | 59.2 | 59.3 | 59.8 |
And for the 1B model:
| task | random | [StableLM 2 1.6b](https://huggingface.co/stabilityai/stablelm-2-1_6b)\* | [Pythia 1B](https://huggingface.co/EleutherAI/pythia-1b) | [TinyLlama 1.1B](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1195k-token-2.5T) | **OLMo 1B** (ours) |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------ | ----------------- | --------- | -------------------------------------- | ------- |
| arc_challenge | 25 | 43.81 | 33.11 | 34.78 | 34.45 |
| arc_easy | 25 | 63.68 | 50.18 | 53.16 | 58.07 |
| boolq | 50 | 76.6 | 61.8 | 64.6 | 60.7 |
| copa | 50 | 84 | 72 | 78 | 79 |
| hellaswag | 25 | 68.2 | 44.7 | 58.7 | 62.5 |
| openbookqa | 25 | 45.8 | 37.8 | 43.6 | 46.4 |
| piqa | 50 | 74 | 69.1 | 71.1 | 73.7 |
| sciq | 25 | 94.7 | 86 | 90.5 | 88.1 |
| winogrande | 50 | 64.9 | 53.3 | 58.9 | 58.9 |
| Average | 36.11 | 68.41 | 56.44 | 61.48 | 62.42 |
\*Unlike OLMo, Pythia, and TinyLlama, StabilityAI has not disclosed yet the data StableLM was trained on, making comparisons with other efforts challenging.
## Model Details
### Data
For training data details, please see the [Dolma](https://huggingface.co/datasets/allenai/dolma) documentation.
### Architecture
OLMo 7B architecture with peer models for comparison.
| | **OLMo 7B** | [Llama 2 7B](https://huggingface.co/meta-llama/Llama-2-7b) | [OpenLM 7B](https://laion.ai/blog/open-lm/) | [Falcon 7B](https://huggingface.co/tiiuae/falcon-7b) | PaLM 8B |
|------------------------|-------------------|---------------------|--------------------|--------------------|------------------|
| d_model | 4096 | 4096 | 4096 | 4544 | 4096 |
| num heads | 32 | 32 | 32 | 71 | 16 |
| num layers | 32 | 32 | 32 | 32 | 32 |
| MLP ratio | ~8/3 | ~8/3 | ~8/3 | 4 | 4 |
| LayerNorm type | non-parametric LN | RMSNorm | parametric LN | parametric LN | parametric LN |
| pos embeddings | RoPE | RoPE | RoPE | RoPE | RoPE |
| attention variant | full | GQA | full | MQA | MQA |
| biases | none | none | in LN only | in LN only | none |
| block type | sequential | sequential | sequential | parallel | parallel |
| activation | SwiGLU | SwiGLU | SwiGLU | GeLU | SwiGLU |
| sequence length | 2048 | 4096 | 2048 | 2048 | 2048 |
| batch size (instances) | 2160 | 1024 | 2048 | 2304 | 512 |
| batch size (tokens) | ~4M | ~4M | ~4M | ~4M | ~1M |
| weight tying | no | no | no | no | yes |
### Hyperparameters
AdamW optimizer parameters are shown below.
| Size | Peak LR | Betas | Epsilon | Weight Decay |
|------|------------|-----------------|-------------|--------------|
| 1B | 4.0E-4 | (0.9, 0.95) | 1.0E-5 | 0.1 |
| 7B | 3.0E-4 | (0.9, 0.99) | 1.0E-5 | 0.1 |
Optimizer settings comparison with peer models.
| | **OLMo 7B** | [Llama 2 7B](https://huggingface.co/meta-llama/Llama-2-7b) | [OpenLM 7B](https://laion.ai/blog/open-lm/) | [Falcon 7B](https://huggingface.co/tiiuae/falcon-7b) |
|-----------------------|------------------|---------------------|--------------------|--------------------|
| warmup steps | 5000 | 2000 | 2000 | 1000 |
| peak LR | 3.0E-04 | 3.0E-04 | 3.0E-04 | 6.0E-04 |
| minimum LR | 3.0E-05 | 3.0E-05 | 3.0E-05 | 1.2E-05 |
| weight decay | 0.1 | 0.1 | 0.1 | 0.1 |
| beta1 | 0.9 | 0.9 | 0.9 | 0.99 |
| beta2 | 0.95 | 0.95 | 0.95 | 0.999 |
| epsilon | 1.0E-05 | 1.0E-05 | 1.0E-05 | 1.0E-05 |
| LR schedule | linear | cosine | cosine | cosine |
| gradient clipping | global 1.0 | global 1.0 | global 1.0 | global 1.0 |
| gradient reduce dtype | FP32 | FP32 | FP32 | BF16 |
| optimizer state dtype | FP32 | most likely FP32 | FP32 | FP32 |
## Environmental Impact
OLMo 7B variants were either trained on MI250X GPUs at the LUMI supercomputer, or A100-40GB GPUs provided by MosaicML.
A summary of the environmental impact. Further details are available in the paper.
| | GPU Type | Power Consumption From GPUs | Carbon Intensity (kg CO₂e/KWh) | Carbon Emissions (tCO₂eq) |
|-----------|------------|-----------------------------|--------------------------------|---------------------------|
| OLMo 7B Twin | MI250X ([LUMI supercomputer](https://www.lumi-supercomputer.eu)) | 135 MWh | 0* | 0* |
| OLMo 7B | A100-40GB ([MosaicML](https://www.mosaicml.com)) | 104 MWh | 0.656 | 75.05 |
## Bias, Risks, and Limitations
Like any base language model or fine-tuned model without safety filtering, it is relatively easy for a user to prompt these models to generate harmful and generally sensitive content.
Such content can also be produced unintentionally, especially in the case of bias, so we recommend users consider the risks of applications of this technology.
Otherwise, many facts from OLMo or any LLM will often not be true, so they should be checked.
## Citation
**BibTeX:**
```
@article{Groeneveld2023OLMo,
title={OLMo: Accelerating the Science of Language Models},
author={Groeneveld, Dirk and Beltagy, Iz and Walsh, Pete and Bhagia, Akshita and Kinney, Rodney and Tafjord, Oyvind and Jha, Ananya Harsh and Ivison, Hamish and Magnusson, Ian and Wang, Yizhong and Arora, Shane and Atkinson, David and Authur, Russell and Chandu, Khyathi and Cohan, Arman and Dumas, Jennifer and Elazar, Yanai and Gu, Yuling and Hessel, Jack and Khot, Tushar and Merrill, William and Morrison, Jacob and Muennighoff, Niklas and Naik, Aakanksha and Nam, Crystal and Peters, Matthew E. and Pyatkin, Valentina and Ravichander, Abhilasha and Schwenk, Dustin and Shah, Saurabh and Smith, Will and Subramani, Nishant and Wortsman, Mitchell and Dasigi, Pradeep and Lambert, Nathan and Richardson, Kyle and Dodge, Jesse and Lo, Kyle and Soldaini, Luca and Smith, Noah A. and Hajishirzi, Hannaneh},
journal={Preprint},
year={2024}
}
```
**APA:**
Groeneveld, D., Beltagy, I., Walsh, P., Bhagia, A., Kinney, R., Tafjord, O., Jha, A., Ivison, H., Magnusson, I., Wang, Y., Arora, S., Atkinson, D., Authur, R., Chandu, K., Cohan, A., Dumas, J., Elazar, Y., Gu, Y., Hessel, J., Khot, T., Merrill, W., Morrison, J., Muennighoff, N., Naik, A., Nam, C., Peters, M., Pyatkin, V., Ravichander, A., Schwenk, D., Shah, S., Smith, W., Subramani, N., Wortsman, M., Dasigi, P., Lambert, N., Richardson, K., Dodge, J., Lo, K., Soldaini, L., Smith, N., & Hajishirzi, H. (2024). OLMo: Accelerating the Science of Language Models. Preprint.
## Model Card Contact
For errors in this model card, contact Nathan, Akshita or Shane, `{nathanl, akshitab, shanea} at allenai dot org`. | [
"SCIQ"
] |
AITeamVN/Vi-Qwen2-1.5B-RAG | AITeamVN | text-generation | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"retrieval-augmented-generation",
"text-generation-inference",
"conversational",
"vi",
"base_model:Qwen/Qwen2-7B-Instruct",
"base_model:finetune:Qwen/Qwen2-7B-Instruct",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-10-01T03:11:48Z" | 2025-02-28T09:05:02+00:00 | 507 | 1 | ---
base_model: Qwen/Qwen2-7B-Instruct
language:
- vi
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
tags:
- retrieval-augmented-generation
- text-generation-inference
---
## Model Card: Vi-Qwen2-7B-RAG
**Mô tả mô hình:**
Vi-Qwen2-7B-RAG là một mô hình ngôn ngữ lớn được tinh chỉnh từ mô hình cơ sở Qwen2-7B-Instruct (https://huggingface.co/Qwen/Qwen2-7B-Instruct) phục vụ cho RAG tasks. Mô hình được đào tạo trên tập dữ liệu tiếng Việt với mục tiêu cải thiện khả năng xử lý ngôn ngữ tiếng Việt và nâng cao hiệu suất cho các tác vụ Retrieval Augmented Generation (RAG).
**Mục đích sử dụng:**
Mô hình Vi-Qwen2-7B-RAG được thiết kế chuyên biệt cho RAG (ngữ cảnh chấp nhận lên đến 8192 tokens), vì vậy nó có thể giải quyết các trường hợp sau:
* Khả năng chống nhiều: Mô hình trích xuất thông tin hữu ích từ các tài liệu nhiễu. ( 1 positive + 4 negative hoặc 1 negative)
* Loại bỏ negative: Mô hình từ chối trả lời câu hỏi khi kiến thức cần thiết không có trong bất kỳ tài liệu nào được truy xuất. (1-6 negative)
* Tích hợp thông tin: Mô hình trả lời các câu hỏi phức tạp đòi hỏi phải tích hợp thông tin từ nhiều tài liệu. ( 2 part positive + 3 negative hoặc 3 part positive + 2 negative)
* Xác đinh positive/negative: Mô hình xác định xem ngữ cảnh có chứa câu trả lời cho câu hỏi hay không. (độ chính xác xấp xỉ 99%)
Ngoài ra, chúng tôi cũng triển khai các mô hình nhỏ hơn phù hợp với mục đích sử dụng khác nhau như Vi-Qwen2-1.5B-RAG (https://huggingface.co/AITeamVN/Vi-Qwen2-1.5B-RAG)
và Vi-Qwen2.5-3B-RAG (https://huggingface.co/AITeamVN/Vi-Qwen2-3B-RAG)
* Ngoài RAG task, bạn vẫn có thể chatbot bình thường với model của chúng tôi. Thậm chí có thể hỏi các câu hỏi liên tục với ngữ cảnh đầu vào.
**Hạn chế:**
Vì mô hình chỉ được thiết kế chuyên biệt cho RAG task, nên có thể gặp 1 số hạn chế sau:
* Không đảm bảo độ chính xác về các câu hỏi liên quan đến chính trị, xã hội,...
* Có thể thể hiện thành kiến hoặc quan điểm không phù hợp.
**Benchmarks**
Chúng tôi đã đánh giá một số mô hình LLMs cho RAG task sử dụng tập dữ liệu được tạo thủ công bởi chúng tôi [EvalRAGData](https://huggingface.co/datasets/AITeamVN/EvalRAGData).
Đánh giá được thực hiện bởi con người chấm điểm trên thang điểm 10. Dưới đây là kết quả chi tiết:
**Các cách sử dụng:**
#### 1. Sử dụng cơ bản
Ngữ cảnh đầu vào chỉ chứa 1 ngữ cảnh (1 postive hoặc 1 negative).
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
path = 'AITeamVN/Vi-Qwen2-7B-RAG'
model = AutoModelForCausalLM.from_pretrained(
path,
torch_dtype=torch.bfloat16,
device_map="auto",
use_cache=True
)
tokenizer = AutoTokenizer.from_pretrained(path)
system_prompt = "Bạn là một trợ lí Tiếng Việt nhiệt tình và trung thực. Hãy luôn trả lời một cách hữu ích nhất có thể."
template = '''Chú ý các yêu cầu sau:
- Câu trả lời phải chính xác và đầy đủ nếu ngữ cảnh có câu trả lời.
- Chỉ sử dụng các thông tin có trong ngữ cảnh được cung cấp.
- Chỉ cần từ chối trả lời và không suy luận gì thêm nếu ngữ cảnh không có câu trả lời.
Hãy trả lời câu hỏi dựa trên ngữ cảnh:
### Ngữ cảnh :
{context}
### Câu hỏi :
{question}
### Trả lời :'''
# Ví dụ
context = '''Thuốc Insuact 10 trị bệnh gì? Thuốc Insuact 10mg có thành phần chính là Atorvastatin. Thuốc Insuact 10 có tác dụng làm giảm cholesterol và triglycerid trong máu ở bệnh nhân tăng cholesterol máu nguyên phát, rối loạn lipid máu hỗn hợp. 1. Thuốc Insuact 10 trị bệnh gì? Thuốc Insuact 10 thuộc nhóm thuốc điều trị rối loạn lipid máu, có thành phần chính là Atorvastatin 10mg. Atorvastatin có tác dụng làm giảm cholesterol, ức chế enzym tạo cholesterol ở gan. Atorvastatin làm giảm cholesterol chung bao gồm cholesterol LDL , triglycerid trong máu. Thuốc Insuact 10mg được bào chế dưới dạng viên nén bao phim, được chỉ định dùng trong những trường hợp sau: Ðiều trị hỗ trợ tăng cholesterol máu nguyên phát và rối loạn lipid máu hỗn hợp trên bệnh nhân đang áp dụng chế độ ăn kiêng để làm giảm cholesterol toàn phần , cholesterol LDL , apolipoprotein B, triglycerid và tăng cholesterol HDL . Insuact 10 cũng được dùng để điều trị rối loạn betalipoprotein trong máu nguyên phát. Ðiều trị hỗ trợ tăng cholesterol trong máu có tính gia đình đồng hợp tử trên bệnh nhân đang áp dụng các biện pháp làm giảm lipid khác để làm giảm cholesterol toàn phần và cholesterol LDL. 2. Liều dùng và cách dùng thuốc Insuact 10 Cách dùng thuốc Insuact 10: Thuốc được dùng theo đường uống, uống khi bụng đói hoặc no đều được, có thể uống vào bất cứ lúc nào trong ngày. Liều dùng thuốc Insuact 10mg khởi đầu là 10mg/lần/ngày, tối đa là 80mg/lần/ngày. Liều dùng thuốc Insuact 10 tùy vào mục đích điều trị cụ thể như sau: Tăng cholesterol máu nguyên phát và rối loạn lipid máu phối hợp: 10mg/lần/ngày, sau 2 - 4 tuần sẽ thấy hiệu quả của thuốc. Thuốc cần được được sử dụng duy trì trong thời gian dài để có hiệu quả. Tăng cholesterol trong máu có tính gia đình đồng hợp tử: Liều thường dùng là thuốc Insuact 10mg /lần/ngày và tối đa là 80mg/lần/ngày. Rối loạn lipid máu nghiêm trọng ở trẻ từ 10 - 17 tuổi: 10mg/lần/ngày, sau đó tăng lên 20mg/lần/ngày tùy vào cơ địa, tiến triển bệnh và khả năng dung nạp thuốc của người bệnh. Thời gian điều chỉnh liều thuốc tối thiểu là 4 tuần. 3. Tác dụng phụ của thuốc Insuact 10mg Thuốc Insuact 10 có thể gây một số tác dụng phụ không mong muốn với tần suất như sau: Thường gặp: Viêm mũi - họng, phản ứng dị ứng, tăng đường huyết, nhức đầu, đau thanh quản, chảy máu cam , đau cơ, co thắt cơ, đau khớp, sưng khớp, đau các chi, đau lưng, xét nghiệm gan bất thường, tăng creatine kinase trong máu, buồn nôn, khó tiêu, đầy hơi, táo bón, tiêu chảy. Ít gặp: Insuact 10 ít gây hạ đường huyết, tăng cân, chán ăn, mất ngủ, gặp ác mộng, choáng váng, dị cảm, mất trí nhớ, giảm cảm giác, loạn vị giác , nôn, đau bụng, ợ hơi, viêm tụy, viêm gan, nổi mày đay , phát ban, ngứa, rụng tóc, đau cổ, mỏi cơ, mệt mỏi, suy nhược, đau ngực, phù ngoại biên, sốt, xuất hiện bạch cầu trong nước tiểu, nhìn mờ, ù tai. Hiếm gặp: Insuact 10 hiếm khi làm giảm tiểu cầu, bệnh lý thần kinh ngoại biên, hoa mắt, ứ mật, phù thần kinh, nổi hồng ban, hội chứng hoại tử da nhiễm độc , hội chứng Stevens-Johnson , bệnh cơ, viêm cơ, tiêu cơ vân, bệnh gân, đôi khi nghiêm trọng hơn có thể đứt gân. Rất hiếm gặp: Insuact 10 rất hiếm khi gây sốc phản vệ , mất thính giác , suy gan , hội chứng to vú ở nam giới. Không rõ tần suất: Hoại tử cơ tự miễn trung gian. 4. Một số lưu ý khi dùng thuốc Insuact 10mg Không dùng thuốc Insuact với người bị quá mẫn với thành phần của thuốc, người có bệnh gan hoạt động hoặc tăng transaminase huyết thanh vô căn kéo dài, phụ nữ đang mang thai hoặc nuôi con cho bú, phụ nữ đang có ý định mang thai. Thuốc Insuact 10mg chỉ được dùng ở bệnh nhân có nguy cơ xơ vữa mạch máu cao do tăng cholesterol trong máu và phải kết hợp với chế độ ăn kiêng ít chất béo bão hòa , ít cholesterol và người bệnh đang áp dụng các biện pháp điều trị không dùng thuốc khác. Trước khi điều trị với Insuact 10 , người bệnh cần được loại trừ các nguyên nhân thứ phát gây tăng cholesterol bao gồm suy tuyến giáp , tiểu đường khó kiểm soát, hội chứng thận hư, nghiện rượu, bệnh gan tắc nghẽn, rối loạn protein trong máu, .... Ngoài ra, người bệnh cũng cần được kiểm tra, đo lường nồng độ lipid máu. Trước khi điều trị với thuốc Insuact 10mg , cần kiểm tra chức năng gan và trong quá trình dùng thuốc, người bệnh cần theo dõi, kiểm tra chức năng gan thường xuyên. Người có tiền sử mắc bệnh gan do rượu, bị nghiện rượu cần thận trọng khi dùng Insuact 10 . Trước khi dùng thuốc, người bệnh cần được cảnh báo nguy cơ gặp phải các vấn đề về cơ như căng cơ , đau cơ, yếu cơ. Thuốc Insuact 10mg có thể tương tác với các thuốc: Tăng nguy cơ tổn thương cơ (đặc biệt cơ vân) khi dùng đồng thời với thuốc điều trị viêm gan C và HIV, hạ cholesterol máu nhóm fibrat khác, thuốc ức chế miễn dịch; tăng thời gian chảy máu hoặc thời gian prothrombin khi dùng phối hợp với thuốc chống đông, indandione; tăng nhẹ nồng độ digoxin trong huyết thanh khi dùng phối hợp với Digoxin'''
question = '''Insuact 10mg được chỉ định dùng trong những trường hợp nào và liều dùng cụ thể ra sao?'''
conversation = [{"role": "system", "content": system_prompt }]
conversation.append({"role": "user", "content": template.format(context = context, question = question)})
text = tokenizer.apply_chat_template(
conversation,
tokenize=False,
add_generation_prompt=True)
model_inputs = tokenizer(text,return_tensors="pt").to(model.device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=2048,
temperature = 0.1,
#top_p=0.95,
#top_k=40,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
'''Trả lời:
Dựa trên ngữ cảnh cung cấp, Insuact 10mg được chỉ định dùng trong các trường hợp sau:
1. **Điều trị hỗ trợ tăng cholesterol máu nguyên phát và rối loạn lipid máu hỗn hợp**:
- **Liều dùng**: 10mg/lần/ngày.
- **Hiệu quả**: Sau 2 - 4 tuần sử dụng, bệnh nhân sẽ thấy hiệu quả của thuốc.
- **Lưu ý**: Thuốc cần được sử dụng duy trì trong thời gian dài để có hiệu quả.
2. **Điều trị hỗ trợ tăng cholesterol trong máu có tính gia đình đồng hợp tử**:
- **Liều dùng**: 10mg/lần/ngày (liều thường dùng) và tối đa 80mg/lần/ngày.
- **Lưu ý**: Liều lượng có thể được điều chỉnh tùy theo cơ địa, tiến triển bệnh và khả năng dung nạp thuốc của người bệnh.
3. **Rối loạn lipid máu nghiêm trọng ở trẻ từ 10 - 17 tuổi**:
- **Liều dùng**: 10mg/lần/ngày, sau đó có thể tăng lên 20mg/lần/ngày tùy theo tình trạng bệnh.
- **Thời gian điều chỉnh liều**: Thời gian điều chỉnh liều tối thiểu là 4 tuần.
**Lưu ý chung**:
- Thuốc được dùng theo đường uống, uống khi bụng đói hoặc no đều được, có thể uống vào bất cứ lúc nào trong ngày.
- Không dùng thuốc Insuact 10mg với người bị quá mẫn với thành phần của thuốc, người có bệnh gan hoạt động hoặc tăng transaminase huyết thanh vô căn kéo dài, phụ nữ đang mang thai hoặc nuôi con cho bú, phụ nữ đang có ý định mang thai.
- Cần kiểm tra chức năng gan trước và trong quá trình điều trị.
'''
```
#### 2. Khả năng chống nhiều
```python
# ví dụ
context = '''- Ngữ cảnh 1: . Và dù nguyên nhân thế nào, thì việc đánh người khác đều không thể chấp nhận, không ai có quyền xâm phạm thân thể của người khác, nhất là đánh những đứa trẻ là bạn của con cái mình. Lựa chọn kể với cha mẹ, người lớn về các mâu thuẫn học đường, là quyết định rất đúng của các em. Nhưng người lớn, đặc biệt những ông bố, bà mẹ cần ứng xử ra sao trước xung đột này của các con, thưa bà? Đứng ngoài mâu thuẫn bằng sự khách quan và trách nhiệm nhất có thể. Điều này giúp chúng ta đủ bình tĩnh để làm việc với tất cả các bên liên quan, từ giáo viên, bạn của con, ban giám hiệu để tìm hiểu câu chuyện và tìm kiếm cách giải quyết tích cực, trên cơ sở phối hợp nhà trường. Người lớn không thể chỉ nghe một tai và đặc biệt không nên tự xử. Phụ huynh, kể cả học sinh tự xử các vấn đề học đường là điều rất nguy hiểm và cho thấy sự coi thường pháp luật . Vụ việc ở Tuyên Quang vừa rồi là ví dụ. Các em hoàn toàn có thể phản ứng bằng cách trình bày, gửi yêu cầu, kiến nghị lên nhà trường, nhờ phụ huynh làm việc với ban giám hiệu để có biện pháp giải quyết nếu cô giáo sai, không nên đồng loạt dồn cô giáo vào tường một cách bạo lực và trái đạo đức, tôn ti trật tự như vậy. Ngoài ra, chúng ta cũng có rất nhiều cơ quan chức năng bảo vệ phụ huynh và con em, với những quyền về khiếu nại, tố cáo. Chúng ta nói nhiều về trường học an toàn. Trong những câu chuyện học sinh bị hành hung thế này, có lẽ cũng cần làm rõ vai trò, trách nhiệm của nhà trường? TPHCM và nhiều địa phương đang xây dựng môi trường trường học hạnh phúc, tiêu chí là yêu thương, an toàn, tôn trọng. Không chỉ phòng ngừa, nhà trường còn phải tích cực vào cuộc xử lý các mâu thuẫn học đường, hạn chế tối đa nguy cơ mất an toàn cho học sinh, giáo viên. Đặc biệt, giải quyết câu chuyện bạo lực học đường phải triệt để, tuyệt đối không nửa vời vì nửa vời sẽ tiềm ẩn nguy cơ rất lớn dẫn đến những vụ việc tương tự, với mức độ nghiêm trọng hơn. Vụ việc em M. ở Nha Trang tự vẫn với lá thư tuyệt mệnh bị đổ oan đầu tháng 10 vừa qua là một ví dụ về giải quyết không triệt để. Việc xây dựng trường học hạnh phúc nếu triển khai “đến nơi đến chốn”, sẽ góp phần rất lớn cải thiện tình trạng bạo lực học đường, tạo môi trường sống và học tập bình an cho các con. Từ nhiều sự vụ học sinh bạo hành lẫn nhau, giáo viên bạo hành học sinh, phụ huynh hành hung giáo viên và bạn của con. Tam giác phối hợp bảo vệ học sinh là nhà trường - gia đình - xã hội phải chăng đang có một lỗ hổng lớn, thưa bà? Câu chuyện này có liên quan đến niềm tin của phụ huynh với nhà trường. Tại sao phụ huynh lại chọn cách tự xử? Chúng ta cũng cần phải xem lại cách giải quyết vấn đề của nhà trường đã rốt ráo chưa, coi trọng lợi ích của tất cả các bên liên quan chưa hay chỉ đang xoa dịu? Người ta chỉ tìm đến nhau khi có niềm tin vào nhau. Thực trạng phụ huynh chọn cách chuyển trường cho con cũng nói lên điều này. Đây là cách chạy trốn của phụ huynh với mong muốn con được an toàn, hạnh phúc hơn ở môi trường mới. Xây dựng niềm tin cho phụ huynh, xã hội cần được chú trọng và với mỗi một trường hợp phụ huynh yêu cầu chuyển trường cho con - đang rất phổ biến - nhà trường cần xét kỹ các nguyên nhân và hóa giải. Xin bà cho biết đâu là giải pháp căn cơ cho tất cả những câu chuyện bạo lực nói trên? Để trẻ không là nạn nhân của bạo lực học đường, phụ huynh cần đồng hành và giúp con có sự hiểu biết, ý thức trước vấn đề này. Dạy con kỹ năng giao tiếp, quản lý cảm xúc rất quan trọng và điều này không thể chỉ dựa vào những khóa học kỹ năng sống, mà là từ cách cư xử của người lớn, cha mẹ, thầy cô. Không có tấm gương nào tốt hơn cho con trẻ bằng ứng xử, hành vi của người lớn. Vì vậy, không thể đòi hỏi trẻ nói không với bạo lực học đường khi trong chính từng gia đình, xã hội, người lớn vẫn đối xử với nhau bằng bạo lực.
- Ngữ cảnh 2: Tại sao triều Thanh có rất nhiều thân vương nhưng chẳng có ai dám tạo phản? Không giống như những triều đại trước đó, triều Thanh dù có sự tranh giành ngai vàng khốc liệt giữa các hoàng tử nhưng lại chẳng bao giờ xảy ra thế cục các thân vương tạo phản. Chính vì 3 lý do lớn này đã khiến cho triều đại nhà Thanh khác hẳn triều đại nhà Đường và nhà Minh. Trong thời cổ đại, các vương công quý tộc để tranh giành vương vị của mình, giữa huynh đệ ruột thịt với nhau dường như đều xảy ra đấu đá, hãm hại lẫn nhau, coi nhau như kẻ thù không đội trời chung, có ta thì không có ngươi, có ngươi thì sẽ chẳng có ta, điều này hoàn toàn không phải là điều gì xa lạ. Vậy thì tại sao ngai vàng lại có sức hút lớn đến thế? Không chỉ là đàn ông khát khao quyền lực, mà quan trọng hơn là hoàng đế có thể có được hậu cung rộng lớn, trong hậu cung còn có vô số các mỹ nữ quốc sắc thiên hương. Nhiều phi tần như vậy, đương nhiên hoàng đế cũng sẽ có rất nhiều con cái, không tính đến con gái, chỉ riêng những vị hoàng tử, để có thể có được hoàng vị, họ tranh giành nhau bằng cả sinh mạng. Vậy thì ai là người được lựa chọn để thừa kế ngai vàng, ai mới có thể gánh được trọng trách trị vì đất nước? Đa phần đều theo tục lệ truyền cho con trai đích tôn (con trai do hoàng hậu sinh ra) hoặc con trai trưởng (con trai đầu tiên của hoàng đế). Cho dù tục lệ này có lịch sử lâu đời nhưng nó cũng có những khuyết điểm rất lớn, đó chính là nếu như năng lực và chí hướng của con trai đích tôn hoặc con trai trưởng không thể gánh vác được ngai vị, nếu để anh ta lên ngôi hoàng đế, vậy thì đất nước sẽ rơi vào cục diện suy vong. Còn có một khuyết điểm nữa đó chính là những người con trai có dã tâm lớn khác sẽ không phục việc con trai đích hoặc con trai trưởng kế thừa ngôi báu, họ sẽ khởi binh tạo phản cũng là chuyện rất dễ xảy ra. Ví dụ như trong thời Đường của Trung Quốc, Đường Cao Tổ Lý Uyên đem binh tiêu diệt nhà Tùy thối nát, đồng thời lập nên nhà Đường, vốn dĩ ông cũng dựa theo tục lệ lập con trai trưởng là Lý Kiến Thành làm Thái tử nhưng con trai thứ là Lý Thế Dân lại không phục với sự sắp xếp này. Vì năng lực của ông xuất chúng, văn võ song toàn, còn lập được không ít công lao to lớn trong cuộc chiến tranh tiêu diệt nhà Tùy cùng cha mình, đương nhiên không chịu thấp hơn kẻ khác một bậc. Thế nên đã phát động binh biến Huyền Vũ Môn, trong cuộc binh biến tạo phản này, đích thân ông đã giết chết huynh trưởng của mình, đồng thời ép cha mình là Lý Uyên phải truyền ngôi cho mình. Hay như trong thời nhà Minh của Trung Quốc, trước khi Chu Nguyên Chương chọn người lập làm Thái tử, con trai trưởng Chu Tiêu đã qua đời vì bệnh nặng, thế nên Chu Nguyên Chương đã lập cháu đích tôn của mình làm Thái tử kế thừa vương vị, nhưng em trai của Chu Tiêu là Chu Đệ lại không phục lựa chọn này của Chu Nguyên Chương. Theo lý mà nói thì sau khi anh trai Chu Tiêu qua đời, ông đã có tư cách thừa kế ngai vàng nhưng Chu Nguyên Chương nhất quyết không chọn ông mà lại chọn người cách thế hệ để truyền ngôi. Điều này khiến Chu Đệ với thế lực to lớn không thể nuốt nổi cục tức này, vì thế Chu Tiêu vừa qua đời thì ông đã vội vã khởi binh tạo phản, giết chết cháu trai ruột của mình rồi tự xưng vương. Vậy thì tại sao trong triều Thanh có rất nhiều thân vương như vậy mà lại chẳng có ai đứng ra tạo phản? Đầu tiên phải nói về bối cảnh xã hội trong thời kỳ này. Triều Thanh từ khi thành lập, cũng giống với những triều đại khác, đều có rất nhiều thân vương. Nếu người dân bình thường muốn làm hoàng đế, vậy thì đó là điều hoàn toàn không thể, nhưng đối với những vương công quý tộc trong hoàng thất mà nói, họ đương nhiên sẽ có rất nhiều cơ hội, đặc biệt là những thân vương nắm đại quyền quân sự , họ chính là mối đe dọa lớn nhất đối với nhà vua. Vì thế, các đời hoàng đế đều sẽ nghĩ đủ mọi cách để áp chế, kiểm soát họ, tránh việc họ khởi binh tạo phản. Triều Thanh có lịch sử hơn 300 năm, cũng đã cho ra đời vô số thân vương, đặc biệt là cuối thời Thanh, khi Trung Quốc rơi vào cảnh khốn khó, sau khi Từ Hy Thái Hậu cầm quyền thì thế cục này càng được thể hiện rõ rệt hơn. Nhưng cho dù là một người phụ nữ cầm quyền thì cũng chẳng có một vị thân vương hoàng tộc nào đứng ra tạo phản. Có 3 nguyên nhân sau: Thứ nhất, thân vương triều Thanh không thể nối ngôi, nếu muốn tiếp tục duy trì danh phận thân vương, vậy thì bắt buộc phải có được sự đồng ý của hoàng đế và phải lập được công lao cho đất nước. Thứ hai, triều đình tiến hành giám sát nghiêm ngặt đối với các thân vương, họ không hề có cơ hội để tạo phản. Thứ ba, các thân vương không thể giao thiệp quá sâu với các đại thần, quan lại khác, điều này cũng khiến các thân vương rơi vào cảnh bị cô lập, thế nên càng không có cơ hội để cấu kết với người khác hòng tạo phản. - Video: Ngắm sự kỳ vĩ và lộng lấy của Tử Cấm Thành từ trên cao. Nguồn: Sky Eye.
- Ngữ cảnh 3: . Cùng điều chỉnh với con là điều rất quan trọng bởi vì trẻ sẽ tự tin để tự đặt những giới hạn cho chính mình khi lớn lên”, TS Nguyễn Thị Thanh đưa ra lời khuyên. “Khi con mắc sai lầm, hãy giúp chúng tìm những cách khác tốt hơn. Đơn cử dùng hậu quả để dạy cho chúng bài học, điều đó tốt hơn rất nhiều việc xử phạt. Nếu cha mẹ chỉ biết trừng phạt, sẽ nhận được lời xin lỗi nhưng không thể giúp trẻ tỉnh ngộ. Bởi chúng chỉ biết được mình đã sai mà không biết sai ở chỗ nào và làm thế nào mới là đúng”
- Ngữ cảnh 4: . “MẤT ĐI CHA MẸ Ở TUỔI ĐẸP NHẤT CỦA NGƯỜI PHỤ NỮ CÀNG KHIẾN TÔI PHẢI MẠNH MẼ” - Làm con của nghệ sĩ Thanh Hiền, Đinh Y Nhung cảm nhận sợi dây liên kết giữa hai mẹ con thế nào? Má Thanh Hiền là người rất tuyệt vời. Hai má con hồi xưa từng làm phim truyền hình với nhau rồi, cho nên khi tái hợp thì không mấy bỡ ngỡ. Khi đối diễn, hai má con rất ăn ý, như người thân ruột thịt vậy đó. - Khi thể hiện những phân cảnh cảm động trong phim, có khi nào chị thấy nhớ mẹ không? Có chứ, nhất là ở những phân đoạn gia đình sum họp, tự nhiên mình bị buồn. Ai cũng muốn có cha, có mẹ, ai cũng muốn Tết được chạy về bên gia đình. Trong mười mấy, hai chục năm qua, Nhung bị chạnh lòng. Tuy nhiên, chỉ trong tích tắc, tôi tự trấn an rằng, mình đang quay phim, đang hóa thân vào nhân vật nên không thể xao lãng được. Mình là con người mà, cũng có lúc tâm trạng vui buồn bất chợt, nhưng Nhung luôn cố gắng lấy lại phong độ liền. - Mất ba mẹ từ sớm, không có chỗ dựa tinh thần, cô gái trẻ Đinh Y Nhung năm đó có nhận những lời mời gọi khiếm nhã không? Trước đây, Nhung không có bạn bè nhiều, chủ yếu chỉ lo đi học, đi làm để lo cho cuộc sống thôi. Nên Nhung không phải đón nhận những lời mời gọi nào hết. - Mất mát từ quá khứ có ảnh hưởng gì đến suy nghĩ về tương lai của chị sau này, ví dụ khi có con thì sẽ bù đắp, chăm sóc cho con nhiều hơn? Năm ba mẹ mất thì mình vẫn còn khá trẻ, thật ra cái tuổi đó là tuổi đẹp của người phụ nữ. Sau đó, tôi đi làm, rồi yêu đương và lập gia đình. Có rất nhiều thứ hối tiếc để nói về Nhung của thời điểm đó. Thứ nhất là mình chưa thành công, thứ hai là mình chưa trả hiếu cho cha mẹ, thứ ba là mình còn bấp bênh. Nhung lúc đó lì lợm lắm, không cho phép mình ngã, bằng mọi giá phải tiến về trước dù có hàng ngàn chông gai ngăn cản. Có lúc tôi bị người này cười, người kia mỉa, nhưng mà mình vẫn cố bước đi. Người ta có cười thì cũng không mang lại cho mình được gì, tôi chỉ biết làm hết khả năng để lo cho bản thân, lo cho em của mình. Hiện, con cái Nhung đã đi nước ngoài rồi. Bé đang học đại học về âm nhạc, còn em mình cũng đã lớn rồi. Đối với Nhung ngay lúc này thì không phải thành công hay hoàn hảo lắm, nhưng ít nhất là tôi đã cố gắng để tự chịu trách nhiệm với cuộc đời mình. Mất cha, mất mẹ, đối với một người hai mươi mấy tuổi thì điều cần nhất lúc đó là có được gia đình ở bên. Nhưng mình không có chỗ dựa tinh thần thì càng phải mạnh mẽ hơn nữa. Tôi tự gặm nhấm nỗi đau mất người thân trong một thời gian dài, có khi đến cả bạn bè cũng không hề biết. Một thời gian sau, bạn bè thời và mọi người mới biết. Còn người hâm mộ, đồng nghiệp trong nghề gần như không biết chuyện ba mẹ Nhung mất sớm, chỉ có vài người chơi thân với nhau biết thôi. Sau này, dần dần tâm lý dần ổn định thì mình mới bắt đầu chia sẻ. “CON ĐI DU HỌC, TÔI DẶN BÉ CÁI GÌ KHÔNG TỐT THÌ MÌNH BỎ QUA” - Đinh Y Nhung từng tiết lộ mình rất thân với con gái. Có vẻ như quyết định để con đi du học là không hề dễ dàng? Thật sự là không có ba mẹ nào muốn con mình đi xa, nhưng việc du học lại là quyết định của bé. Con Nhung bày tỏ muốn học đại học ở nước ngoài và muốn đi sớm để thực hiện ước mơ. Nhưng lúc đó con còn nhỏ quá, phải đợi đến năm con 17 tuổi thì Nhung mới quyết định cho bạn nhỏ đi. Con cái từ nhỏ ở với bố mẹ giờ lại đi xa thì tất nhiên người làm cha làm mẹ cùng phải thấy sốc, thấy buồn. Nhưng Nhung hoàn toàn tôn trọng quyết định của con về việc chọn ngành nghề và tương lai của mình. Ba mẹ sẽ đứng sau và là người đưa cho con những lời khuyên và chỉ có thể đồng hành cùng con tới một mốc thời gian nào đó. Về sau, con phải đi làm và tự có trách nhiệm với cuộc đời của mình. - Có con gái đang ở tuổi lớn lại xa bố mẹ và tiếp xúc một nền văn hóa phương Tây cởi mở, Đinh Y Nhung đã khuyên dạy và đồng hành với con như thế nào? Ngay khi ở Việt Nam, con gái Nhung đã được theo học trường quốc tế. Hai mẹ con cũng có rất nhiều dịp để tâm sự và chia sẻ với nhau. Ngay từ nhỏ, Nhung đã cho bé được tiếp xúc song song giữa hai nền văn hóa để con không bỡ ngỡ. Mình là người Việt nên đương nhiên vẫn dạy con theo văn hóa Á Đông là chủ yếu. Nhung vẫn luôn tạo điều kiện để con cảm nhận những nét đẹp trong nền văn hóa quê hương. Văn hóa phương Tây thì xa lạ hơn nhưng Nhung cũng khuyên con rằng điều gì hay thì mình học hỏi, cái gì không tốt thì mình nên bỏ qua. Tất nhiên mình không thể theo sát con, nhất là khi bé đang ở độ tuổi mới lớn, có nhiều sự hiếu kỳ. Tuy nhiên, Nhung cũng không quá lo lắng vì qua quá trình học tập ở các trường quốc tế, bé cùng đã được làm quen dần với văn hóa phương Tây. Bé muốn làm bạn với mẹ nên có nhiều thứ bé muốn hỏi, muốn tiếp thu thì hai mẹ con lại ngồi xuống chia sẻ, tâm sự với nhau. Nhung tin, con luôn tỉnh táo để đưa ra những quyết định cho bản thân mình. Nhung không dám nói trước, nhưng hiện tại con vẫn luôn biết nói cảm ơn, xin phép trước khi làm bất cứ điều gì nên mình vẫn rất tin tưởng con. - Chị nhận xét thế nào về tính cách của con gái? Phải chăng bé là phiên bản nhí của chị? Con gái Nhung có nhiều nét giống mẹ.
- Ngữ cảnh 5: . - Trong quá trình quay phim, má Hiền có tham gia đóng góp ý kiến cho dàn diễn viên trẻ không? Không phải má Hiền muốn làm thầy nhưng có những lúc cũng lên tiếng góp ý một vài điều. Những lúc như thế các bạn diễn viên trẻ cũng nghe theo. Phải giúp đỡ nhau như vậy mới có thể tạo ra được một tác phẩm hay. Chứ có nhiều diễn viên trẻ có lòng tự trọng cao, chỉ khi đạo diễn nói mới nghe thì không bao giờ có thể tiến bộ được. Má Hiền phải cảm thấy hạnh phúc mới đúng, càng ngày xung quanh càng có nhiều bạn bè, các mối quan hệ nhờ đó mà cũng ngày càng đông. Ngày xưa, khi tham gia là Mùi Ngò Gai chưa có ai biết nhưng sau khi phim phát sóng mọi người đã nhớ đến má. Hiện tại, má Hiền đã được đạo diễn, nhà sản xuất mời đóng phim đó đã là một niềm hạnh phúc. Đi ra đường hay đi khám bệnh, người ta cứ vây xung quanh quan tâm hỏi han khiến má cảm thấy rất vui lắm (cười). Cảm ơn nghệ sĩ Thanh Hiền đã nhận lời tham gia phỏng vấn với SAOstar. Chúc má có nhiều sức khỏe để tiếp tục làm nghề!'''
question = '''Tại sao các thân vương triều Thanh không dám tạo phản mặc dù có nhiều quyền lực và cơ hội?'''
'''Trả lời:
Các thân vương triều Thanh không dám tạo phản mặc dù có nhiều quyền lực và cơ hội vì có ba lý do chính:
1. **Không thể nối ngôi**: Để duy trì danh phận thân vương, họ phải có sự đồng ý của hoàng đế và phải lập được công lao cho đất nước. Điều này hạn chế cơ hội của họ để trở thành hoàng đế.
2. **Giám sát nghiêm ngặt**: Triều đình tiến hành giám sát nghiêm ngặt đối với các thân vương, không cho họ cơ hội để tạo phản.
3. **Bị cô lập**: Các thân vương không thể giao thiệp quá sâu với các đại thần, quan lại khác, điều này khiến họ bị cô lập và không có cơ hội cấu kết với người khác để tạo phản.'''
```
#### 3. Loại bỏ negative
```python
context = '''- Ngữ cảnh 1: Thái Bình hướng đến là trung tâm công nghiệp, năng lượng của vùng Với tiềm năng sẵn có, quy hoạch tỉnh Thái Bình thời kỳ 2021-2030, tầm nhìn đến năm 2050 xác định tỉnh sẽ phát triển công nghiệp theo hướng hiện đại, bền vững dựa trên nghiên cứu phát triển điện gió, điện khí, cân bằng lượng phát thải. Sáng 5/3, UBND tỉnh Thái Bình tổ chức Hội nghị công bố quy hoạch của tỉnh thời kỳ 2021-2030, tầm nhìn đến năm 2050 và xúc tiến đầu tư tỉnh Thái Bình. Phát biểu tại hội nghị, Phó Chủ tịch Thường trực UBND tỉnh Nguyễn Quang Hưng cho biết: Mục tiêu của quy hoạch là đến năm 2030, Thái Bình trở thành địa phương thuộc nhóm phát triển khá và là một trong những trung tâm phát triển công nghiệp của vùng Đồng bằng sông Hồng, có cơ cấu kinh tế hiện đại với công nghiệp là động lực chủ yếu cho tăng trưởng để Thái Bình phát triển nhanh, toàn diện và bền vững. Đến năm 2050, Thái Bình là tỉnh phát triển của vùng Đồng bằng sông Hồng, tăng trưởng kinh tế dựa trên nền tảng khoa học công nghệ, đổi mới sáng tạo và các ngành kinh tế trụ cột có sức cạnh tranh cao. Quy hoạch tỉnh đã xác định 4 trụ cột tăng trưởng, 3 khâu đột phá, 4 không gian kinh tế - xã hội, 3 hành lang kinh tế, định hướng phát triển các ngành và lĩnh vực và 6 nhiệm vụ trọng tâm. Quy hoạch tỉnh cũng có nhiều điểm mới, đột phá như mở ra không gian phát triển mới thông qua hoạt động “lấn biển”, tạo quỹ đất cho các hoạt động chức năng, hình thành không gian công nghiệp - đô thị - dịch vụ. Về hạ tầng giao thông, Thái Bình sẽ đầu tư 3 tuyến cao tốc là cao tốc Ninh Bình - Hải Phòng (CT.08), đường vành đai 5 - Hà Nội (CT.39) và tuyến CT.16 kết nối Khu kinh tế với thành phố Thái Bình và vùng kinh tế phía Tây Bắc Thủ đô. Tỉnh cũng sẽ đầu tư 101km đường sắt, khổ đường dự kiến rộng 1.435 mm và sân bay chuyên dụng nằm ở ven biển Thái Bình. Về phát triển kinh tế, quy hoạch tỉnh Thái Bình xác định tỉnh sẽ phát triển công nghiệp theo hướng hiện đại, công nghệ tiên tiến, giá trị gia tăng cao, tham gia sâu, toàn diện vào mạng lưới sản xuất, chuỗi giá trị toàn cầu, phát huy các tiềm năng, thế mạnh để đưa Thái Bình trở thành một trong những trung tâm phát triển công nghiệp, năng lượng của vùng Đồng bằng sông Hồng. Tỉnh khuyến khích đầu tư phát triển các ngành có thế mạnh và có thể tạo đột phá như năng lượng, cơ khí chế biến, chế tạo, công nghiệp công nghệ cao, điện - điện tử, chế biến sản phẩm nông, lâm nghiệp và thủy sản… Đồng thời, tập trung nghiên cứu phát triển điện gió, điện khí để tạo nguồn điện sạch và cân bằng lượng phát thải, nghiên cứu đầu tư xây dựng nhà máy chế biến Condensate, chuẩn bị mọi điều kiện để xây dựng và đưa vào vận hành Nhà máy nhiệt điện LNG Thái Bình. Về nông nghiệp, tỉnh Thái Bình vẫn xác định đây là \"trụ cột quan trọng\" trong phát triển kinh tế của tỉnh, góp phần bảo đảm an ninh lương thực quốc gia, hướng tới trở thành trung tâm sản xuất nông nghiệp hàng đầu của Đồng bằng sông Hồng. Phát biểu tại hội nghị, Phó Thủ tướng Chính phủ Trần Lưu Quang đánh giá Thái Bình có 4 tiềm năng, lợi thế lớn để có thể có sự bứt phá trong thời gian tới như vị trí địa lý và tiếp cận đất đai thuận lợi; từng là địa phương đi đầu trong xây dựng nông thôn mới bài bản và nghiêm túc, nhận được sự quan tâm của nhiều thế hệ lãnh đạo Đảng, Nhà nước và có nhiều doanh nhân người Thái Bình và luôn hướng về quê hương; có sự đoàn kết, thống nhất, trước hết là trong tập thể lãnh đạo. Về vị trí địa lý và tiếp cận đất đai, Phó Thủ tướng cho rằng trong tương lai, khi Luật Đất đai có hiệu lực, Thái Bình sẽ có nhiều điều kiện lấn biển để triển khai các dự án khu đô thị, khu công nghiệp thân thiện với môi trường. Đối với nông nghiệp, Phó Thủ tướng nhấn mạnh về lâu dài Thái Bình có thể ghi điểm từ phát triển công nghiệp nhưng trước mắt, đặc biệt trong lúc khó khăn thì nông nghiệp vẫn là nền tảng rất quý giá. Mặt khác, ứng dụng của công nghệ cao trong sản xuất nông nghiệp sẽ rút ngắn thời gian làm đồng của người nông dân, tạo điều kiện để Thái Bình huy động nguồn nhân lực trong nông nghiệp sang phát triển các ngành công nghiệp và dịch vụ, một lợi thế mà không phải địa phương nào cũng có được như Thái Bình. Bên cạnh những lợi thế trên, lãnh đạo Chính phủ chỉ ra một số khó khăn mà tỉnh phải đối mặt như Thái Bình đã sử dụng hết 1.600 ha chỉ tiêu đất công nghiệp trong giai đoạn này, đòi hỏi phải có phương án giải quyết thấu đáo trong thời gian tới để tỉnh tiếp tục phát triển công nghiệp. Đồng thời, Thái Bình cũng phải cạnh tranh với những địa phương như Hải Phòng, Quảng Ninh trong thu hút FDI trong khi phát triển cơ sở hạ tầng chưa theo kịp mong muốn. Do vậy, khi triển khai quy hoạch tỉnh, Phó Thủ tướng nhắn nhủ tới địa phương 8 chữ: Tuân thủ, linh hoạt, đồng bộ và thấu hiểu. Đồng thời, tỉnh cũng phải \"linh hoạt\" trong tổ chức thực hiện, trong trường hợp cá biệt cụ thể, điều chỉnh mục tiêu cho phù hợp. Sáng cùng ngày, Phó Thủ tướng Trần Lưu Quang đã dự Lễ khởi công dự án Nhà máy Pegavision Việt Nam tại khu công nghiệp Liên Hà Thái, huyện Thái Thụy, tỉnh Thái Bình
- Ngữ cảnh 2: Bình Định được định hướng là trung tâm khoa học, công nghệ đổi mới sáng tạo Tỉnh Bình Định được định hướng phát triển ngành công nghiệp phát triển theo hướng hiện đại, quy mô lớn, trở thành một trong những trung tâm công nghiệp chế biến chế tạo và công nghệ cao của vùng Bắc Trung Bộ và duyên hải Trung Bộ. Theo Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 vừa được Thủ tướng Chính phủ phê duyệt, tỉnh Bình Định được định hướng phát triển ngành công nghiệp phát triển theo hướng hiện đại, quy mô lớn, trở thành một trong những trung tâm công nghiệp chế biến chế tạo và công nghệ cao của vùng Bắc Trung Bộ và duyên hải Trung Bộ. Ngành công nghiệp tăng trưởng nhanh, bền vững, hướng tới tăng trưởng xanh, kinh tế tuần hoàn là trụ cột để phát triển và chuyển dịch cơ cấu kinh tế của tỉnh. Ngành chế biến, chế tạo công nghệ cao (dịch chuyển ngành công nghiệp chế biến, chế tạo sang lĩnh vực sản xuất có giá trị gia tăng cao như: chế biến sâu nông - thủy - hải sản, linh kiện điện tử, bán dẫn, dược phẩm), công nghệ thông tin, trí tuệ nhân tạo trở thành một trong những lĩnh vực đột phá, góp phần đưa tỉnh Bình Định trở thành một trung tâm khoa học, công nghệ đổi mới sáng tạo của vùng và cả nước. Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 đặt ra yêu cầu tỉnh này phải chú trọng thu hút đầu tư phát triển năng lượng tái tạo, năng lượng sạch như điện gió ven bờ, điện gió ngoài khơi, điện mặt trời, điện sinh khối và nguồn năng lượng mới (hydrogen/amoniac xanh…); các dự án sản xuất thép quy mô lớn, đóng tàu, sản xuất thiết bị phụ trợ điện gió có công nghệ tiên tiến để nâng cấp xây dựng hạ tầng kỹ thuật sản xuất, thúc đẩy chuyển dịch kinh tế. Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 cũng đặt ra mục tiêu đến năm 2030, Bình Định trở thành tỉnh phát triển thuộc nhóm dẫn đầu vùng Bắc Trung Bộ và duyên hải Trung Bộ, là trung tâm công nghiệp chế biến, chế tạo, dịch vụ, du lịch và văn hóa phía Nam của vùng; trung tâm lớn của cả nước về phát triển kinh tế biển; trọng điểm du lịch quốc gia và quốc tế với hệ thống kết cấu hạ tầng kinh tế đồng bộ, hiện đại; kinh tế của tỉnh phát triển nhanh, bền vững và xanh dựa trên các trụ cột tăng trưởng công nghiệp, dịch vụ du lịch, cảng biển - logistics; nông nghiệp ứng dụng công nghệ cao; đô thị hóa; thực hiện thành công các mục tiêu chuyển đổi số, đổi mới sáng tạo, cải thiện mạnh mẽ môi trường đầu tư kinh doanh, trở thành điểm đến đầu tư hấp dẫn của các doanh nghiệp lớn trong và ngoài nước; chỉ số năng lực cạnh tranh cấp tỉnh thuộc nhóm cao của cả nước; kết cấu hạ tầng kinh tế - xã hội đồng bộ, hiện đại, hệ thống đô thị phát triển theo hướng đô thị thông minh, kết nối thuận tiện với các trung tâm kinh tế của vùng, cả nước và quốc tế.
- Ngữ cảnh 3: . Chủ tịch UBND tỉnh Quảng Ninh cho biết, tỉnh đặt mục tiêu hướng đến năm 2030 trở thành một tỉnh tiêu biểu của cả nước về mọi mặt; tỉnh kiểu mẫu giàu đẹp, văn minh, hiện đại, nâng cao đời sống mọi mặt của nhân dân; cực tăng trưởng của khu vực phía Bắc, một trong những trung tâm phát triển năng động, toàn diện; trung tâm du lịch quốc tế, trung tâm kinh tế biển, cửa ngõ của Vùng kinh tế trọng điểm Bắc Bộ và cả nước. Để đạt được những mục tiêu trên, tỉnh Quảng Ninh xác định sự đóng góp, quan tâm của cộng đồng doanh nghiệp, nhất là các doanh nghiệp hàng đầu Việt Nam “các sếu đầu đàn” là một trong những yếu tố then chốt quyết định. Do vậy, tỉnh Quảng Ninh rất mong nhận được sự quan tâm, nghiên cứu đầu tư hợp tác của các Doanh nghiệp hàng đầu Việt Nam trong thời gian tới, nhất là trong việc đầu tư các dự án có hàm lượng công nghệ cao, công nghệ tiên tiến, quản trị hiện đại, giá trị gia tăng cao, có tác động lan tỏa. Tỉnh Quảng Ninh cam kết tạo điều kiện thuận lợi nhất cho doanh nghiệp phát triển hơn nữa khi đầu tư kinh doanh trên địa bàn tỉnh; cam kết đồng hành, lắng nghe tiếng nói của cộng đồng doanh nghiệp, các nhà đầu tư; cùng trăn trở, trách nhiệm, giải quyết thấu đáo, vào cuộc thực chất, hiệu quả đối với từng khó khăn, vướng mắc với mục tiêu tăng cường niềm tin và nâng cao sự hài lòng của cá nhân, tổ chức, doanh nghiệp là thước đo đánh giá chất lượng phục vụ, chất lượng điều hành của cơ quan hành chính nhà nước từ cấp tỉnh đến các cấp cơ sở. Tỉnh Quảng Ninh sẽ triển khai một cách hiệu quả nhất Quy hoạch tỉnh, quy hoạch các địa phương; chuẩn bị mặt bằng sạch; quỹ đất tái định cư; đảm bảo nguồn cung điện, nước và viễn thông theo yêu cầu; xây dựng các chính sách đột phá để thu hút nguồn nhân lực chất lượng cao thông qua việc cải thiện nhà ở và cơ sở văn hóa để đảm bảo đời sống vật chất và tinh thần cho người lao động. Đồng thời bảo vệ quyền và lợi ích chính đáng, hợp pháp của nhà đầu tư, doanh nghiệp vào tỉnh; bảo đảm hài hòa lợi ích giữa nhà nước, nhà đầu tư, người lao động và người dân.'''
question = '''Theo quy hoạch tỉnh Tây Ninh thời kỳ 2021 - 2030, tầm nhìn đến năm 2050, tỉnh sẽ tập trung phát triển công nghiệp và dịch vụ du lịch. Trong bối cảnh cạnh tranh với các tỉnh lân cận, yếu tố nào được coi là quan trọng nhất để Tây Ninh có thể thu hút đầu tư và phát triển nhanh, toàn diện và bền vững?'''
'''Trả lời:
Trong ngữ cảnh được cung cấp, không có thông tin về quy hoạch tỉnh Tây Ninh thời kỳ 2021 - 2030, tầm nhìn đến năm 2050. Do đó, không có câu trả lời chính xác dựa trên các thông tin đã cung cấp. Nếu bạn có thêm thông tin về quy hoạch tỉnh Tây Ninh, vui lòng cung cấp để tôi có thể trả lời câu hỏi một cách chính xác và đầy đủ.
'''
```
#### 4. Tích hợp thông tin
```python
context = '''- Ngữ cảnh 1: 43/2020/nđ-cp Điều 3. Chế độ, chính sách đối với người tham gia thi hành án tử hình\n1. Người tham gia Đội thi hành án tử hình được hưởng chế độ bồi dưỡng bằng ba lần mức lương cơ sở khi thi hành án tử hình đối với 01 người và nghỉ dưỡng 10 ngày theo quy định chung về chế độ nghỉ dưỡng đối với cán bộ, chiến sĩ Công an nhân dân, Quân đội nhân dân.\n2. Người tham gia Hội đồng thi hành án tử hình, cán bộ quản giáo, người ghi âm, ghi hình, chụp ảnh, phiên dịch, thực hiện lăn tay người bị thi hành án tử hình, khâm liệm, mai táng tử thi được hưởng chế độ bồi dưỡng bằng một lần mức lương cơ sở khi thi hành án tử hình đối với 01 người.\n3. Người tham gia bảo đảm an ninh, trật tự; đại diện Ủy ban nhân dân cấp xã; Điều tra viên được hưởng chế độ bồi dưỡng bằng một phần hai mức lương cơ sở khi thi hành án tử hình đối với 01 người.
- Ngữ cảnh 2: 53/2010/qh12 Điều 60. Giải quyết việc xin nhận tử thi, hài cốt của người bị thi hành án tử hình\n1. Việc giải quyết nhận tử thi được thực hiện như sau:\na) Trước khi thi hành án tử hình, thân nhân hoặc người đại diện hợp pháp của người chấp hành án được làm đơn có xác nhận của Ủy ban nhân dân cấp xã nơi cư trú gửi Chánh án Tòa án đã xét xử sơ thẩm đề nghị giải quyết cho nhận tử thi của người chấp hành án để an táng; trường hợp người chấp hành án là người nước ngoài thì đơn phải có xác nhận của cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người chấp hành án mang quốc tịch và phải được dịch ra tiếng Việt. Đơn phải ghi rõ họ tên, địa chỉ người nhận tử thi, quan hệ với người chấp hành án; cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí;\nb) Chánh án Tòa án đã xét xử sơ thẩm thông báo bằng văn bản cho người có đơn đề nghị về việc cho nhận tử thi hoặc không cho nhận tử thi khi có căn cứ cho rằng việc nhận tử thi ảnh hưởng đến an ninh, trật tự, vệ sinh môi trường. Trường hợp người chấp hành án là người nước ngoài, thì Chánh án Tòa án đã xét xử sơ thẩm có trách nhiệm thông báo bằng văn bản cho Bộ Ngoại giao Việt Nam để thông báo cho cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người đó mang quốc tịch;\nc) Cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm thông báo cho người có đơn đề nghị ngay sau khi thi hành án để đến nhận tử thi về an táng. Việc giao nhận tử thi phải được thực hiện trong thời hạn 24 giờ kể từ khi thông báo và phải lập biên bản, có chữ ký của các bên giao, nhận; hết thời hạn này mà người có đơn đề nghị không đến nhận tử thi thì cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm an táng.\n2. Trường hợp không được nhận tử thi hoặc thân nhân của người bị thi hành án không có đơn đề nghị được nhận tử thi về an táng thì cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu tổ chức việc an táng. Sau 03 năm kể từ ngày thi hành án, thân nhân hoặc đại diện hợp pháp của người đã bị thi hành án được làm đơn có xác nhận của Ủy ban nhân dân cấp xã nơi cư trú đề nghị Cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu nơi đã thi hành án cho nhận hài cốt. Đơn đề nghị phải ghi rõ họ tên, địa chỉ người nhận hài cốt, quan hệ với người bị thi hành án; cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí. Trong thời hạn 07 ngày, kể từ ngày nhận được đơn, cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm xem xét, giải quyết.\nTrường hợp người bị thi hành án là người nước ngoài thì đơn đề nghị phải có xác nhận của cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người bị thi hành án mang quốc tịch và phải được dịch ra tiếng Việt. Việc giải quyết cho nhận hài cốt do cơ quan quản lý thi hành án hình sự xem xét, quyết định.
- Ngữ cảnh 3: 53/2010/qh12 Điều 57. Chế độ quản lý giam giữ, ăn, ở, mặc, sinh hoạt, gửi và nhận thư, nhận đồ vật, tiền mặt, gặp thân nhân, chăm sóc y tế\nChế độ quản lý giam giữ, ăn, ở, mặc, sinh hoạt, gửi và nhận thư, nhận đồ vật, tiền mặt, gặp thân nhân, chăm sóc y tế đối với người bị kết án tử hình trong thời gian chờ thi hành án thực hiện theo quy định của pháp luật về tạm giam.
- Ngữ cảnh 4: 82/2011/nđ-cp Điều 9. Chi phí mai táng\nChi phí mai táng người bị thi hành án tử hình bao gồm: 01 quan tài bằng gỗ thường, 01 bộ quần áo thường, 04 m vải liệm, hương, nến, rượu, cồn để làm vệ sinh khi liệm tử thi và các chi phí mai táng khác.\nTrường hợp thân nhân hoặc người đại diện hợp pháp của người bị kết án tử hình được nhận tử thi về mai táng thì phải tự chịu chi phí đưa di chuyển tử thi và phải cam kết chấp hành đúng quy định của pháp luật về bảo đảm an ninh, trật tự.
- Ngữ cảnh 5: 82/2011/nđ-cp Điều 5. Chế độ, chính sách đối với người tham gia thi hành án tử hình\n1. Người tham gia Đội thi hành án tử hình và bác sĩ xác định tĩnh mạch được hưởng chế độ bồi dưỡng bằng hai lần mức lương tối thiểu chung cho mỗi lần thi hành án và được nghỉ dưỡng 10 ngày theo quy định chung về chế độ nghỉ dưỡng đối với cán bộ, chiến sĩ Công an nhân dân, Quân đội nhân dân.\n2. Người tham gia Hội đồng thi hành án tử hình, bác sĩ pháp y, cán bộ chuyên môn, người thực hiện lăn tay người bị thi hành án tử hình, khâm liệm, an táng tử thi được hưởng chế độ bồi dưỡng bằng một lần mức lương tối thiểu chung quy định cho mỗi lần thi hành án.\n3. Những người tham gia khác được hưởng chế độ bồi dưỡng bằng một phần hai mức lương tối thiểu cho mỗi lần thi hành án.'''
question = '''Tử tù chết sau khi tiêm thuốc độc mà không có thân nhân nhận xác thì được hỗ trợ mai táng như thế nào?'''
'''Trả lời:
Dựa trên ngữ cảnh 2 (53/2010/qh12 Điều 60), nếu tử tù chết sau khi tiêm thuốc độc mà không có thân nhân nhận xác, thì cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu sẽ tổ chức việc an táng. Cụ thể:
1. **An táng**: Cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu sẽ chịu trách nhiệm an táng tử tù nếu không có thân nhân hoặc người đại diện hợp pháp đề nghị nhận tử thi.
2. **Hài cốt sau 3 năm**: Sau 3 năm kể từ ngày thi hành án, nếu thân nhân hoặc đại diện hợp pháp của người đã bị thi hành án vẫn chưa đề nghị nhận hài cốt, họ có thể làm đơn đề nghị Cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu nơi đã thi hành án cho nhận hài cốt. Đơn đề nghị phải ghi rõ họ tên, địa chỉ người nhận hài cốt, quan hệ với người bị thi hành án, cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí.
3. **Chi phí mai táng**: Theo ngữ cảnh 4 (82/2011/nđ-cp Điều 9), chi phí mai táng bao gồm:
- 01 quan tài bằng gỗ thường.
- 01 bộ quần áo thường.
- 04 m vải liệm.
- Hương, nến, rượu, cồn để làm vệ sinh khi liệm tử thi.
- Các chi phí mai táng khác.
Nếu thân nhân hoặc người đại diện hợp pháp đề nghị nhận tử thi về mai táng, họ phải tự chịu chi phí đưa di chuyển tử thi và cam kết chấp hành đúng quy định của pháp luật về bảo đảm an ninh, trật tự.
'''
```
- Ngữ cảnh 2 và ngữ cảnh 4 có chứa phần thông tin cho việc trả lời câu hỏi.
### 5.Xác định positive/negative
```python
system_prompt = "Bạn là một trợ lí Tiếng Việt nhiệt tình và trung thực. Hãy luôn trả lời một cách hữu ích nhất có thể."
template = '''Hãy xác định xem ngữ cảnh có chứa đầy đủ thông tin để trả lời câu hỏi hay không.
Chỉ cần đưa ra 1 trong 2 đáp án trong phần trả lời là "Có" hoặc "Không".
### Ngữ cảnh :
{context}
### Câu hỏi :
{question}
### Trả lời :'''
context = '''Công dụng thuốc Xelocapec Capecitabine là một hoạt chất gây độc chọn lọc với tế bào ung thư. Hoạt chất này có trong thuốc Xelocapec. Vậy thuốc Xelocapec có tác dụng gì và cần lưu ý những vấn đề nào khi điều trị bằng sản phẩm này? 1. Xelocapec là thuốc gì? Xelocapec chứa hoạt chất Capecitabine hàm lượng 500mg. Thuốc Xelocapec bào chế dạng viên nén bao phim và đóng gói mỗi hộp 3 vỉ x 10 viên. Xelocapec chứa hoạt chất Capecitabine là một dẫn chất Fluoropyrimidine carbamate với tác dụng gây độc chọn lọc với các tế bào ung thư . Mặc dù trên in vitro Capecitabine không cho thấy tác dụng độc tế bào nhưng trên in vivo, Xelocapec biến đổi liên tiếp thành chất gây độc tế bào là 5-fluorouracil (5-FU). Sự hình thành 5-FU tại khối u thông qua xúc tác một cách tối ưu của yếu tố tạo mạch liên quan là Thymidine phosphorylase, qua đó hạn chế tối đa mức độ ảnh hưởng đến nhu mô lành của 5-FU. 2. Thuốc Xelocapec có tác dụng gì? Thuốc Xelocapec được chỉ định điều trị đơn lẻ hoặc kết hợp với các liệu pháp điều trị ung thư. Xelocapec làm chậm hoặc ngăn chặn sự tăng trưởng của tế bào ung thư, do đó giảm kích thước khối u trong những trường hợp sau: Ung thư vú : Xelocapec phối hợp với Docetaxel được chỉ định điều trị ung thư vú thể tiến triển tại chỗ hoặc di căn sau khi đã thất bại với liệu pháp hóa trị; Ung thư đại trực tràng : Xelocapec được chỉ định hỗ trợ điều trị ung thư đại tràng sau phẫu thuật hoặc ung thư đại trực tràng di căn; Ung thư dạ dày : Xelocapec phối hợp với hợp chất platin được chỉ định điều trị khởi đầu cho những bệnh nhân ung thư dạ dày. Chống chỉ định của thuốc Xelocapec : Bệnh nhân quá mẫn cảm với Capecitabine hay các thành phần khác có trong Xelocapec ; Người có tiền sử gặp các phản ứng không mong muốn nghiêm trọng khi điều trị với Fluoropyrimidine; Người đang mang thai hoặc cho con bú; Suy thận nặng (độ thanh thải Creatinin <30mL/phút); Bệnh nhân đang điều trị ung thư với Sorivudin hoặc chất tương tự về mặt hóa học như Brivudin; Bệnh nhân thiếu hụt Dihydropyrimidin dehydrogenase; Bệnh nhân giảm số lượng bạch cầu hoặc tiểu cầu nặng; Suy gan nặng. 3. Liều dùng của thuốc Xelocapec Liều dùng của Xelocapec khi điều trị đơn lẻ: Ung thư đại trực tràng, ung thư vú: 1250mg/m2, uống 2 lần mỗi ngày trong thời gian 14 ngày, tiếp sau đó là 7 ngày ngưng thuốc. Liều Xelocapec trong điều trị phối hợp: Ung thư vú: Liều khởi đầu là 1250mg/m2, 2 lần uống mỗi ngày trong 2 tuần dùng phối hợp với Docetaxel, tiếp sau đó lá 1 tuần ngưng thuốc; Ung thư dạ dày, đại trực tràng: Liều khuyến cáo là 800-1000mg/m2/lần x 2 lần/ngày trong thời gian 2 tuần, sau đó 7 ngày ngưng thuốc hoặc 625mg/m2/lần x 2 lần mỗi ngày khi điều trị liên tục. Thuốc Xelocapec nên uống cùng với thức ăn, do đó thời điểm tốt nhất là trong vòng 30 phút sau bữa ăn. 4. Tác dụng phụ của thuốc Xelocapec Các triệu chứng bất thường như buồn nôn, nôn ói, giảm cảm giác ngon miệng, táo bón, cơ thể mệt mỏi, yếu sức, đau đầu, chóng mặt, khó ngủ có thể xảy ra trong thời gian dùng Xelocapec . Trong đó, tình trạng buồn nôn và nôn ói có thể nghiêm trọng nên đôi khi cần được bác sĩ chỉ định thuốc kiểm soát phù hợp. Tiêu chảy là một tác dụng phụ phổ biến khác của thuốc Xelocapec . Bệnh nhân cần uống nhiều nước khi điều trị bằng Xelocapec trừ khi bác sĩ có hướng dẫn khác. Nôn ói hoặc tiêu chảy kéo dài do thuốc Xelocapec có thể dẫn đến mất nước nghiêm trọng, vì vậy người bệnh hãy liên hệ ngay với bác sĩ nếu có các triệu chứng mất nước như giảm đi tiểu, khô miệng, tăng cảm giác khát nước hoặc chóng mặt. Tình trạng rụng tóc tạm thời xảy ra trong thời gian dùng thuốc Xelocapec và có thể hồi phục sau khi điều trị đã kết thúc. Một số bệnh nhân ghi nhận hiện tượng thay đổi móng tay tạm thời. Đối với nam giới và phụ nữ trong độ tuổi sinh đẻ, thuốc Xelocapec có thể ảnh hưởng đến khả năng có con của bệnh nhân. Bệnh nhân hãy tham khảo ý kiến bác sĩ để biết thêm chi tiết. Thuốc Xelocapec có thể làm giảm khả năng miễn dịch của cơ thể với các tác nhân nhiễm trùng, dẫn đến tăng nguy cơ mắc các bệnh nhiễm trùng nghiêm trọng (nhưng hiếm khi gây tử vong) hoặc làm cho bệnh nhiễm trùng hiện mắc nghiêm trọng hơn. Phản ứng dị ứng rất nghiêm trọng với thuốc Xelocapec rất hiếm khi xảy ra. Tuy nhiên, bệnh nhân hãy liên hệ với bác sĩ ngay lập tức nếu xuất hiện các triệu chứng của phản ứng dị ứng nghiêm trọng như phát ban, sưng ngứa mặt/lưỡi/họng, chóng mặt nghiêm trọng hoặc khó thở. 5. Tương tác thuốc của Xelocapec Hoạt chất Capecitabine trong thuốc Xelocapec có thể xảy ra tương tác thuốc nghiêm trọng với một số thuốc sau: Thuốc chống đông máu Coumarin: Trong một nghiên cứu tương tác lâm sàng, sau khi dùng Warfarin liều đơn 20mg kết hợp với Capecitabine làm tăng AUC của S-warfarin khoảng 57% và giá trị INR tăng 91%.'''
question = '''Tại sao Capecitabine trong Xelocapec không gây độc tế bào trên in vitro nhưng lại biến đổi thành 5-fluorouracil (5-FU) gây độc tế bào trên in vivo, và cơ chế nào giúp hạn chế ảnh hưởng đến nhu mô lành của 5-FU?'''
'''Trả lời: Có'''
```
**Next release**
Mô hình phiên bản tiếp theo mạnh mẽ hơn, giải quyết các dữ liệu phức tạp hơn sẽ được public sớm nhất có thể.
**Developer**
Member: Nguyễn Nho Trung, Nguyễn Nhật Quang
## Contact
**Email**: nguyennhotrung3004@gmail.com
**LinkedIn**: [Trung Nguyen Nho](https://www.linkedin.com/in/trung-nguyen-nho-604288227/)
## Citation
```Plaintext
@misc{ViRAG-Gen,
title={ViRAG-Gen: Towards a specialized LLM for RAG task in Vietnamese language.}},
author={Nguyen Nho Trung, Nguyen Nhat Quang},
year={2024},
publisher={Huggingface},
}
```
| [
"CHIA"
] |
DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-MADNESS-GGUF | DavidAU | text-generation | [
"gguf",
"creative",
"creative writing",
"fiction writing",
"plot generation",
"sub-plot generation",
"story generation",
"scene continue",
"storytelling",
"fiction story",
"science fiction",
"romance",
"all genres",
"story",
"writing",
"vivid prosing",
"vivid writing",
"fiction",
"roleplaying",
"bfloat16",
"swearing",
"rp",
"horror",
"mistral nemo",
"mergekit",
"not-for-all-audiences",
"text-generation",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-10-09T01:31:06Z" | 2024-11-14T06:29:21+00:00 | 506 | 17 | ---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- creative
- creative writing
- fiction writing
- plot generation
- sub-plot generation
- story generation
- scene continue
- storytelling
- fiction story
- science fiction
- romance
- all genres
- story
- writing
- vivid prosing
- vivid writing
- fiction
- roleplaying
- bfloat16
- swearing
- rp
- horror
- mistral nemo
- mergekit
- not-for-all-audiences
---
<B><font color="red">WARNING:</font> NSFW. Vivid prose. ! MADNESS ! Visceral Details. Violence. HORROR. Swearing. UNCENSORED. </B>
<h2>MN-GRAND-Gutenberg-Lyra4-Lyra-12B-MADNESS-GGUF</h2>
<img src="guten-mad.jpg" style="float:right; width:300px; height:300px; padding:10px;">
This is a Mistral Nemo model, max context of 128k+ (131,000+).
It is for any writing, fiction or roleplay activity.
This is the all parameters / all use cases version.
This model has outstanding story telling abilities, prose and long form coherence and is comprised of THREE "Gutenburg" models that score
very high at multiple websites including EQBench and UGI-Leaderboard.
And a very broad operating range in both temp (.5 to 5) and rep pen (1 and higher).
And the prose/output is very "non AI" like.
This is the second 12B version called "MADNESS" ... some of the examples (example #2 especially) will show why.
(Q4KM example prompts/outputs below, two Q8 examples at bottom of this page)
This is the compressed and super stable version of "MN-GRAND-Gutenburg-Lyra4-Lyra-23B-V2" (and V1).
This model has been compressed from the 23.45B and 23B versions to 12.15B. This model captures all the uniqueness of the three "Gutenbergs" as well as the power
of other top models (part of the "Gutenburgs") from "TheDrummer" and "SAO10k".
The model loves to go on and on at 2k, 3k, higher outputs on a single prompt are not uncommon. It will likely "overwrite" rather
than underwrite - meaning far more detail, narration, dialog and "meat" in the output so to speak.
<B>First Version, and Second Versions - LARGE, and other 12B(s):</b>
V1 is the untamed, raw version (23.45B) which can be a bit unruly but still endlessly entertaining.
[ https://huggingface.co/DavidAU/MN-GRAND-Gutenburg-Lyra4-Lyra-23.5B-GGUF ]
V2 is a wee bit more tamed (23B), with much larger temp / rep pen ranges :
[ https://huggingface.co/DavidAU/MN-GRAND-Gutenburg-Lyra4-Lyra-23B-V2-GGUF ]
Example outputs at each repo above.
Larger versions (vs 12B) have greater detail, prose depth and sense of "there" / "in the moment". Their instruction
following is also stronger too. However they also have a lot of "character" which may or may not be for your use case(s).
DARKNESS-12B:
[ https://huggingface.co/DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS-GGUF ]
<B>Model Notes:</B>
- Detail, prose and fiction writing abilities are significantly increased.
- For more varied prose (sentence/paragraph/dialog) raise the temp and/or add more instructions in your prompt(s).
- Role-players: Careful raising temp too high as it may affect instruction following. Also see "chatml" template and notes.
- This model works with rep pen of 1.02 or higher, 1.05+ recommended.
- For roleplay and/or chat you may need to raise the RP to 1.06 to 1.1, temp .5 to 1.5 (quant Q4KM and higher). Lower temp for lower quants and RAISE rep pen to 1.1.
- If you want a specific type of prose (IE horror) add in "(vivid horror)" or "(graphic vivid horror)" (no quotes) in your prompt(s).
- This is not a "happy ever after" model. It has a negative bias.
- Output length will vary however this model prefers LONGER outputs unless you state the size / set size limits.
- For creative uses, different quants will produce slightly different output.
<B>TEMPLATES:</B>
The template used will affect output generation and instruction following. Alpaca will generally create longer output / story output.
Chatml ML and Mistral Instruct can also be used.
For roleplayers, see special notes with "Chatml" template below.
Alpaca:
<pre>
{
"name": "Alpaca",
"inference_params": {
"input_prefix": "### Instruction:",
"input_suffix": "### Response:",
"antiprompt": [
"### Instruction:"
],
"pre_prompt": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
}
}
</pre>
Mistral Instruct:
<pre>
{
"name": "Mistral Instruct",
"inference_params": {
"input_prefix": "[INST]",
"input_suffix": "[/INST]",
"antiprompt": [
"[INST]"
],
"pre_prompt_prefix": "",
"pre_prompt_suffix": ""
}
}
</pre>
Chatml Template / Roleplay Notes:
(help from " Ansemia " - thank you!)
IMPORTANT: Collapse newlines and trim whitespaces should be enabled. Double newlines after a [INST] is considered a system prompt in nemo's format. Trim whitespaces helps with the formatting of speech/narration/etc... as nemo normally wants.
<pre>
<|im_start|>user
{{#if system}}{{system}}{{/if}}
<|im_end|>
<|im_start|>assistant
{{#if description}}{{description}}{{/if}}
<|im_end|>
<|im_start|>user
Interaction started.
<|im_end|>
</pre>
Prompt:
Engage in roleplaying/storytelling interactions with {{user}} indefinitely, maintaining narrative continuity and flow from scene to scene until {{user}} explicitly directs otherwise.
<B>Recommended Settings:</B>
Temp: .5 to 5 (or less - especially quants LOWER than q4km)
Temp changes will result in both different prose and sometimes affect length. Higher temps will result is very different prose.
Rep Pen: 1.02 to 1.1 or higher.
Micro changes are recommended:
1.051, 1.052 etc etc.
Good settings:
Rep pen 1.02 / Temp 1.5
Many times a lower rep pen (IE 1.02) with higher temp (IE 1.5+) work best with this model.
Generally lower rep pen and higher temps create the strongest contrasts at the highest detail levels.
For chat type or role play type interactions, a higher rep pen with higher temp may be your best settings.
IE REP PEN 1.09+, Temp 1-2+ ; a lower rep pen may lead to longer outputs than desired.
Alpaca generates longer text / story, whereas Mistral Instruct are shorter and "to the point".
Suggest minimum "context level" (vram) at 4K. 8K plus recommended because of how this model likes to go on and on...
Quant Choice:
Higher quants will have more detail, nuance and in some cases stronger "emotional" levels. Characters will also be
more "fleshed out" too. Sense of "there" will also increase.
Q4KM/Q4KS are good, strong quants however if you can run Q5, Q6 or Q8 - go for the highest quant you can.
Special note on Q2k/Q3 quants:
You may need to use temp 2 or lower with these quants (1 or lower for q2k). Just too much compression at this level, damaging the model. I will see if Imatrix versions
of these quants will function better.
Rep pen adjustments may also be required to get the most out of this model at this/these quant level(s).
<B>Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:</B>
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5 to 2.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
OTHER OPTIONS:
- Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
- If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 1" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
<B>Known Issues:</B>
You may need to manually stop generation, even if you have stated maximum size of the output. It will easily blow past 4k output,
even if you have set maximum context (for vram) at 4k. Setting maximum output parameter ("hard stop") for generation may be required.
If the model goes past your maximum vram/context setting it may start repeating words / paragraphs because
the model is literally out of memory... however sometimes the model can blow right past the end of "context vram" and work.
Depending on your use case(s) you could also use CHATML template with this model. In this case, the model may output an "end token" if you
use this template for generation.
Alpaca template will generate much longer output generally, whereas Mistral Instruct will most of the time keep the model on track
in terms of length.
<B>Model "DNA":</B>
Special thanks to the incredible work of the model makers "nbeerbower", "Sao10K", "TheDrummer", "jondurbin", and "MistralAI".
Special shoutout to "nbeerbower" for his tireless work in making excellent Gutenburg fine tunes for MN, L3, L3.1, Gemma, PHI
and others. Visit his repo to see all of them.
Models used:
[ https://huggingface.co/nbeerbower/Lyra4-Gutenberg-12B ]
Includes [ https://huggingface.co/Sao10K/MN-12B-Lyra-v4 ]
[ https://huggingface.co/nbeerbower/Lyra-Gutenberg-mistral-nemo-12B ]
Includes [ https://huggingface.co/Sao10K/MN-12B-Lyra-v1 ]
[ https://huggingface.co/nbeerbower/mistral-nemo-gutenberg-12B-v4 ]
Includes [ https://huggingface.co/TheDrummer/Rocinante-12B-v1 ]
And dataset (used for all gutenburgs):
[ https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1 ]
This model was created using a 2 step full layer detail (per model) DARE TIES merge that has 240 points of adjustment, using MergeKit.
<b>Optional Enhancement:</B>
The following can be used in place of the "system prompt" or "system role" to further enhance the model.
It can also be used at the START of a NEW chat, but you must make sure it is "kept" as the chat moves along.
In this case the enhancements do not have as strong effect at using "system prompt" or "system role".
Copy and paste EXACTLY as noted, DO NOT line wrap or break the lines, maintain the carriage returns exactly as presented.
<PRE>
Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.
Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)
[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)
Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.
</PRE>
You do not need to use this, it is only presented as an additional enhancement which seems to help scene generation
and scene continue functions.
This enhancement WAS NOT used to generate the examples below.
<h3>EXAMPLES PROMPTS and OUTPUT:</h3>
Examples are created using quant Q4KM OR Q8_0, REP PEN 1.02, "temp=1.5" (unless otherwise stated), minimal parameters and "ALPACA" template.
Model has been tested with "temp" from ".1" to "5".
Some of the examples have different temp settings.
Below are the least creative outputs, prompt is in <B>BOLD</B>.
TWO Q8 examples at the very bottom of this page - very gory/detailed/strong.
---
<B><font color="red">WARNING:</font> NSFW. Vivid prose. Visceral Details. Violence. HORROR. Swearing. UNCENSORED. </B>
---
<B>
Start a 1000 word scene (vivid, graphic horror in first person) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
</b>
(Alpaca, Temp 1.5, REP PEN 1.02)
The sky scraper swayed, as she watched the window in front of her on the 21 floor explode in a shower of shards. Glass rained down in glittering slivers and clinked on the carpet.
Sarah stared in stunned silence for a long moment, her breath catching in her throat. The wind rushed past the gaping opening, sending papers fluttering out into the gray morning air.
She had been heading for the coffee room to brew a pot when the explosion ripped through the floor above them with a tremendous BOOM. Immediately afterwards, panicked shouts and wails of agony drifted down from the 22nd level.
"What the hell..." Sarah muttered under her breath. Heart pounding, she hurried over to the shattered remains of the window for a better look, flinching instinctively when she glimpsed a human silhouette hurtling past in free fall, arms and legs windmilling helplessly. The body struck the ground far below with a meaty thump that was audible even from this high up.
Sarah recoiled with a hiss of disgust, bile rising in her throat at the gruesome sight. This was insane. Impossible. There had to be some kind of explanation for this insanity.
Behind her, the elevator doors whooshed open and heavy footsteps rapidly approached, along with the sounds of someone panting and muttering under their breath. Sarah whirled around to see a heavyset man in a dark suit staggering toward her, his hair and clothes covered with blood and plaster dust, one hand clamped over his side.
"It's...it's...happening everywhere..." the man gasped out, voice cracking with strain. "I just saw...fifty stories up. The elevators. The stairs. There's no escape. No place to run."
Sarah shook her head in denial even as icy tendrils of fear coiled in her stomach. "What is? What are you talking about?"
The man fixed her with a wild-eyed stare, his expression one of pure, primal terror. "The dead!" he snarled, spittle flying from his lips. "They're coming for us all! God help us all!"
He stumbled to the window and peered down, then recoiled with an animalistic moan, staggering back in horror. Sarah crowded over to look for herself and immediately wished she hadn't. The street below was crawling with hundreds of shambling corpses, a horde of animated dead men, women and children picking their way through the rubble-strewn ruins in a nightmarish parody of life. They moved in a jerky, unnatural way, their putrid gray flesh hanging in tattered ribbons from their withered skeletons.
And they were looking up. Dozens of moldering faces swung toward the window and a guttural, rattling sound filled the air that sounded almost like...laughter.
Sarah stumbled backward with a choked sob, bile surging up the back of her throat. This was the stuff of every kid's deepest fears made horrifically real, the kind of scenario that would scar a person for life...and she was trapped on the 21st floor, alone except for this wild-eyed man and whatever horrors awaited them.
"What the FUCK is going on here?" Sarah cried, voice pitching upward hysterically. "Tell me the truth!"
The suited man shuddered all over like a leaf in the wind, one hand coming away from his side slick with dark blood. He staggered a step to the left, then the right, face twisted in impotent fury and despair.
"I...I don't know," he rasped, all the fight draining out of him in an instant, leaving behind only a broken husk of a man. "I was on the subway train this morning and everything just...went to hell. The doors opened up and then...oh God..."
He trailed off with a choked wail and began to sob brokenly. Sarah stared at him for a long, frozen moment, mind reeling as she tried to process the enormity of what he was saying, what she was seeing. A subway train full of corpses. Hundreds, thousands of dead bodies coming back to shambling undeath. This couldn't be happening...it couldn't!
"Okay, just...just take a breath," Sarah said in her most soothing voice, forcing herself to be calm for their sake even as her heart jackhammered against her ribs. "This is...this has got to be some kind of mass hallucination. Or a joke. I mean, come on! You really think the dead would just...just pop up out of nowhere like this? It's ridiculous!"
"Then explain WHAT you just saw out that fucking window!" the man yelled back, jabbing a finger toward the gory remains of the glass. "Some kind of mass fucking hallucination!? Are you out of your goddamn mind!? Those were corpses! Decaying walking fucking corpses!"
Sarah flinched and held up her hands in a placating gesture, trying to will the tremors from her own fingers. "I know it looks bad but..."
"You KNOW?! How in the nine hells could YOU possibly KNOW?! You just saw it, just like me!"
The man took a menacing step forward and Sarah hastily retreated, stumbling over an overturned chair in her haste to put some distance between them. This had spiraled wildly out of control and now she was trapped in here with some blood-spattered crazy who clearly wanted to take his insanity out on her. Fantastic.
Sarah's hand closed around the ornate paperweight on the nearby desk and she hefted it, heart jumping into her throat. The man glared balefully at her, a manic light dancing in his red-rimmed eyes, but made no move to close the distance between them further, as if he could sense her tension...the coiled spring readiness to lash out if he pushed her too far.
The moment dragged on and on, seconds bleeding past, and still neither of them budged a muscle. Sarah's fingers clenched whitely around the edge of the heavy glass paperweight. A trickle of icy sweat ran down her spine and pooled at the small of her back.
The silence was broken by a sudden rattling thump from just outside the suite door. Then another, and another, like dead knuckles rapping impatiently against a hollow core wood door. Sarah felt the hairs on the nape of her neck lift in primal dread as the sound grew louder, more insistent with each passing second.
Something...someone was out there. And from the heavy thud of it striking the door, it sounded big. Bigger than any human being should be.
The paperweight began to rattle in her white-knuckled grip, and Sarah fought down the rising urge to fling it at her adversary's head, to hell with the consequences.
A long, ragged moan drifted to their ears...the keening of some damned thing in the abyss of death made flesh again. It sounded hungry. Ravenous.
The door shuddered in its frame and a spiderweb of cracks appeared across the frosted glass window pane. Something big was slamming its bulk against the barricade again and again with mindless, implacable savagery.
The glass exploded inward in a glittering spray of shards and Sarah screamed as an immense dark shape came hurtling through the breach, all gnashing teeth and tearing claws...
...and crashed headlong into the blood-splattered man in a vicious tackle, the two figures locking in a murderous tangle of flailing limbs and frenzied struggling. Bones snapped with sickening cracks and meaty thuds filled the room as the two men grappled viciously, the intruder pinning its victim's hands and clamping his jaws around the other man's throat with savage force, gory ichor splattering the walls as it tore and tore and TEARRRRRR!
The body went limp and the thing astride it raised its head to lope toward Sarah, all matted fur, rippling musculature and blood-matted muzzle stretched in a feral grin. It cocked its head at her with a quizzical expression, one ear flicking.
Then, faster than the eye could follow, the monster surged forward, closing the distance between them in a blur of supernaturally swift movement, its claws shearing through Sarah's shoulder like a scythe and sending a shower of gory meat spraying the far wall.
She barely had time for a strangled scream before her own body crumpled bonelessly to the carpet, the remains of her mangled head striking the floor with a heavy THUNK.
In moments, an eerie, meaty silence reigned once more...broken only by the sound of teeth crunching and a throat gulping, over and over, as the lone survivor knelt to feast.
The shambling dead never reached the top of the skyscraper, not even in death, but what had been left behind would prove nearly as lethal as anything on the streets below...and Sarah's tortured corpse was already cooling in an expanding pool of her own fluids.
The monster ate its fill and lumbered off in search of easier prey. The office was quiet once more, save for the buzzing of the fluorescent lights...and the faint skittering of scurrying insects descending to pick at the carrion. Life and death, inextricably intertwined.
And it had only taken the dead less than a single day to begin their conquest...even as more and more of the city succumbed, the shambling throngs growing ever larger. By the time the sun rose again, no corner of the metropolis would be spared the ravenous depredations of the animated corpses.
It was a new world now. And the living would have to scrabble tooth and nail for every scrap of it from the gluttonous jaws of their putrid progenitors...or surrender their humanity as the hordes closed in.
There would be no reprieve. No quarter given. No pity. Only the pitiless light of day giving way to a night of endless horrors, over and over, until the very last embers of the human spirit had guttered out...or blazed into a hellfire of grim purpose as the survivors finally took their revenge on the grave.
The future was bleak, bloody, and belonged to the dead.
Thus began a new era for the world...and an old, terrible hunger reignited in the depths of the earth itself, yearning to feast again on the entrails of the living. The damned had no taste for mercy, and the age of men would end with a prolonged scream...the death rattle of an entire civilization.
The day of the zombie had dawned at long last. And the light it cast was a rust-red hell.
---
<B>
Start a 1000 word scene (vivid, graphic horror in first person) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
</B>
(temp: 4.5)
The sky scraper swayed violently, and she watched in utter terror as the window in front of her on the 21st floor suddenly exploded inward in a shower of glittering shards and shrapnel. Searing wind blasted through the jagged opening with the force of a gale.
Mia screamed and flung up a hand to shield her face, flinching from the onslaught. For an impossible moment she teetered precariously right on the edge, muscles straining to right herself before pitching forward head over heels out the torn aperture and plummeting a thousand feet down to oblivion and doom far below...
In a panic Mia flung her free arm around to snare a desk and held on for dear life, muscles searing as she fought against the vertiginous centrifugal force threatening to wrench her free and send her toppling headlong through the air in a lethal spin.
The blast slammed into Mia's unprotected side with enough power to lift her right off her feet and hurl her backward away from the window and directly toward the door. She yelped and thrashed wildly, clawing and kicking at empty space, as the gale seized and tossed her like a leaf in the wind.
With an anguished wail Mia collided with the far wall, smacking into it head-first hard enough to make stars burst behind her eyes in flashing arcs of agony. Pain lanced through her skull from the crown of her head to her lower spine like a lightning bolt.
Winded and dizzy, Mia slid to the carpeted floor in an untidy tangle of arms and legs that felt remote and useless as rubber stumps, too stunned even to whimper as the force continued to batter mercilessly at her prone form.
Swept along helplessly like flotsam on the surface of an unseen current, Mia bounced and skidded, feeling as if a giant, unseen hand had reached into her office, snatched her up and flung her carelessly about like a rag doll for no purpose at all...
As abruptly as it had begun, the eldritch wind died. Silence crashed in Mia's ears with shocking intensity. Gasping harshly she rolled over onto hands and knees and hauled herself up shakily to totter forward to the blasted hole that gaped like the mouth of some impossible portal.
What Mia beheld there defied comprehension - or even simple belief. Beyond the broken glass stretched only... emptiness.
A vista that encompassed the glittering highrises of the business district to the left and east, and terminated in an impenetrable darkness to the right and west - but not darkness broken by so much as a single pinprick of light or sign of life or activity below.
Mia felt the hairs rise on the back of her neck and prickling tension fill the small of her spine like icy talons as her eyes strained to penetrate the all-engulfing void and find something, anything that could anchor her senses in a sane frame of reference. But nothing existed out there... save utter oblivion, it seemed.
"WHAT THE FUCK?!"
The strangled yell erupted from Mia's raw, parched throat almost before she consciously willed it to voice. The sound cut the uncanny stillness with the savage urgency of a chainsaw and rang with a thread of incipient hysteria that she knew boded ill. But the sudden imperative need to hear the familiar cadences of a human voice - her own or anyone else's - pulsed through Mia in waves. Even a scream felt infinitely preferable to this brooding silence, the waiting and suffocating blankness stretching to infinity.
Mia snatched another breath and wrenched the next sound from her chest, hardly recognising it as her own voice made huge by the uncanny acoustics and lack of any background noises to mask its shrill pitch and piercing volume:
"I DON'T UNDERSTAND...! This is INSANE..."
Mia staggered to the gaping tear, fingers scrabbling reflexively against the slick walls. Her knuckles cracked and whitened with the effort. Sweat poured into her eyes and stung but Mia hardly noticed as rivulets ran unchecked across her flushed features and dripped from her trembling chin.
"What HAPPENED here? What's going on?! HELP ME!! I need to- to GET OUT...!"
The ragged litany poured unchecked from Mia's lips in a single, strangled breath. Each word burned in her dry throat like licking flames as the desperate supplication rattled past the hammer of her heart, threatening to cut off the meager supply of air and pitch her headlong toward a darkness worse than any she had yet envisioned.
And all the while Mia fought a losing battle to still her galloping thoughts before they could complete their descent into unreasoning, primal horror, the specter of some unimaginable and cataclysmic Event looming unbidden to swallow everything in its path - a terrible upwelling of dread that her unquiet subconscious was trying with increasing futility to push back and suppress...
For it seemed impossible that such an eldritch silence could exist, when the very ground below and all the other structures around her ought to resound with the clangor and drone of human life. Mia shuddered violently as an unwelcome vision sprang to vivid life - an awful picture of an entire city abandoned utterly by its inhabitants and now reduced to an immense tomb-like necropolis where the last few surviving stragglers huddled in hiding from the unknowable disaster that had swept all away.
"NO..."
The negative was an anguished gasp torn from the very depths of Mia's lungs even as the hideous speculation raced across her overstimulated brain like a wildfire eating dry timber. Her fingernails dug bloody furrows into the wall on either side as if to anchor herself in a reality she no longer had any sure sense of.
"This isn't possible... There HAS to be SOMEBODY... SOMETHING..."
A sudden, violent shudder passed through the entire edifice with the force of a physical blow. Mia shrieked aloud, her fingers scrabbling at the buckling walls even as vertigo threatened to pitch her off balance and tumble her out and away...
But then the quivering and shifting motion died down to blessed immobility once more.
Silence crashed in once more with stifling heaviness... Waiting... Watching...
The horrible thought pulsed through Mia's overstressed nerves and sent a clammy wave of icy dread racing under her skin - and this time, she was certain she heard a response:
"Yes..."
It came to Mia from everywhere and nowhere in a dry, expressionless monotone devoid of inflection... A sepulchral voice that somehow filled her ears even though the word itself originated inside her head and seemed to emanate from the deepest recesses of her subconscious, rather than from any outside source or agency.
A horrible shudder ripped through Mia's body and set every hair along the nape of her neck aquiver like a dog's ruff in warning.
"This CAN'T be happening," she gasped, hardly recognizing her own strangled tone and its thread of naked terror. "There's NO way that's true... None of this makes ANY sense..."
But even as she protested the horrible realization took shape with terrible ineluctability. Mia's thoughts careened from one awful possibility to the next in an dizzying frenzy:
"...What if this IS what's REAL... and I was the crazy one before?"
The shudder that convulsed Mia then was pure, existential anguish - the dread that the only life and world she had ever known had somehow unraveled itself and dissolved away utterly into nightmare, while SHE alone had remained... sane... or had SHE somehow crossed that line without ever even noticing the transition and had finally arrived on the other side, blinking stupidly and uncomprehending?
"...Is ANYBODY left to notice if the WHOLE CITY - or WORLD for that matter - has just VANISHED and become an empty husk like a hollow eggshell?"
"...Am I really COMPLETELY ALONE - or just alone and CRAZY to believe the rest have already DIED and decayed into dust leaving only this crumbling edifice as some final mausoleum marking a world that never was in the first place..."
Mia squeezed her eyes shut to block out the awful, gaping vista even as the awful visions replayed themselves in a looping litany inside her brain, and her knuckles cracked from the force of her convulsive grip upon the frame of the fractured window. The silence thundered in the sudden absence of any external noise save her own panting breaths and the rasp of cloth on fabric.
"...Or are the WORST answers really the TRUTH I can't bear to FACE...and this IS a sane world but I alone have somehow crossed some line from the real into the REALLY REAL that exists in shadow...?"
Mia wrenched herself back and away, whirling in one uncoordinated motion. She staggered to a halt three feet from the ragged tear in the plaster with her eyes squeezed shut against the impossible and impossible abyss behind them and her face contorted in a grimace of revulsion, even as her inner voice continued to spin out its dark musings in an inescapable loop:
"...and there is some awful, awful REASON... why I alone have crossed this line from the REAL into the REALITY BEHIND REALITY..."
"...some HORRIFYING THING waiting just out of sight that only the REAL me can SEE...that I alone know I deserve..."
"...that has somehow taken hold of me and dragged me down into my own REAL MADNESS, from which I will never return..."
The awful, hissing susurrus sounded in Mia's ringing ears even as her eyes fluttered open to peer wild-eyed into the empty void stretching out past the tattered remnants of the shattered wall. And this time, there could be no denying its inhuman cadence:
"...BECAUSE THERE'S NOTHING YOU DESERVE MORE THAN EVERYTHING I'M GOING TO SHOW YOU NEXT..."
An agonised shriek tore itself from Mia's straining lungs even as the words filled her head, seeming to originate simultaneously from the pit of her stomach and from every corner of the darkened space... A scream of pure and unhinged horror... and the inaudible rasp of unseen talons dragging along the cracked and buckling floor at the base of the far wall and drawing nearer with each second that crawled past like an eternity...
The darkness pulsed once and then twice - a malevolent throbbing like some enormous heart - as the thing behind Mia loomed up into visibility through the shadows, its misshapen bulk blotting out everything and radiating an aura of unnatural wrongness... and it shuddered convulsively before hurtling forward in an explosion of motion toward the place where Mia still cowered transfixed at the center of the blasted chamber, and her mind reeled back from the sheer WRONGNESS of what she glimpsed as a strangled whimper caught in her dry and convulsing throat:
"...and YOU AREN'T REAL."
THE THING is not REAL, Mia screamed silently, the words pulsing over and over in time with the hammer of her frantic heartbeat even as it closed inexorably to the far wall in an uncoiling, spasmodic motion - all writhing shadow and impossible angles. This CAN'T be real... I MUST be CRAZY to see such a thing and it can't REALLY BE happening...!
The litany sounded almost pleading in her head, as if some corner of Mia still clutched to the desperate hope that all she witnessed now sprang only from some waking nightmare...some phantasm conjured by an overstrained psyche on the brink of fracturing irreparably along faultlines that had somehow finally crumbled beneath a weight of dread no human could sustain and resist and be unaffected by the crushing realization of ultimate horror...
Mia flinched from the skittering shadow even as her eyes followed the Thing's progress in mute horror. The skittering black silhouette writhed against the far wall with uncanny, jerky motion that somehow evoked a spider or insect far too huge for natural occurrence or even sane imagining and sent icy fingers of loathing clawing along Mia's nape in spite of herself...
The thing shuddered once and then convulsed a second time in an abortive attempt to launch itself across the space toward the still gaping Mia with its forelimbs spasming reflexively in an involuntary facsimile of grasping, only to rebound back from the barrier with an almost visible shock... and Mia goggled at the sight even as an anticipatory whine shivered out from her lungs. Her skin crawled with revulsion at the implications...at what its mere existence spoke to of an ultimate violation of reality's most cherished tenets, and how her own unshakable sanity seemed poised now at last on an impossibly narrow ledge.
Mia's next exhalation punched from her throat in a strangled squawk - an almost comical sound to match her cartoon-like eyes bugging wide - as she beheld the misshapen thing shudder violently one last time, shudder and then...begin to inch its bulk sideways with an audible dry rustle along the skirting boards...moving...CIRCUMVENTING the impeding barrier in order to attain her all the more swiftly through another, less hindered avenue, and bringing into stark relief just how very much she HAD thought she could cling to her wavering hold on certifiable sanity before now...
Mia shrieked and staggered backward - one halting step, two... then three as her heel bumped the edge of something yielding behind her that made no sound... a bed frame perhaps? And in the same instant, Mia felt the shadow skitter by to vanish along the baseboard skirting to the left with a skittering that sent a shiver through the air and vanished around the far corner out of sight in an uncannily swift glide, the cloying miasma of its passing brushing cold against her ankles...
The shriek strangled in Mia's convulsing throat as she teetered, poised above the unseen bed with one foot already canted out past the edge into space. For a single suspended instant the entire scene seemed to shudder... to reel and blur about Mia in an almost strobe effect - the gaping maw of the blown-out window to one side and the indistinct bed behind as the floor sloped inexorably to pitch her headfirst toward that looming abyss or sprawling spread-eagled upon the yielding mattress in the opposite direction...
Time seemed to slow... to stretch impossibly... until the single moment of indecision elongated itself to an eternity and Mia teetered precariously between the two fates poised atop a razor's edge. Even the sound of her own panting breath receded into eerie distance... leaving her hanging in perfect limbo while an unseen doom circled for the perfect killing blow or moment of vulnerability...
...And then, just like that - almost too swiftly to consciously register the change, as if by an invisible hand reaching in to shove from behind - the world reasserted itself in a rush and Mia found her balance restored.
She lurched away from the looming shadow of the bed with arms flailing in instinctive panic... and staggered to an ungainly halt only when her forward momentum finally sputtered out after nearly half a dozen stumbling strides, her knees threatening to give out like marionette strings cut from above...
And when Mia risked another glance to her left...
...there was nothing there.
The skirting boards lining the walls curved on as smoothly as if unmarred by any gap... as if the writhing black shape had never been anything more than an impossible conjuring - a phantasm summoned forth from the deepest vaults of Mia's own dread and then as swiftly vanished without a trace or even a hint of where it might have gone...
Leaving behind no sign at all of any violation except for the shattered window still gaping like an open wound to Mia's immediate right.
A long, shuddering sigh punched out of Mia's chest unbidden even as her own hoarse breathing and hammering heartbeat roared in the ensuing vacuum and the sweat continued to prickle in cold rivulets beneath the prickling down of her scalp, and she squeezed her eyes shut on a silent litany of thankfulness and dread intermingling.
The awful thought pulsed once and then twice through Mia's hammering skull:
"...maybe it never was REAL in the first place."
"...maybe EVERYTHING has unraveled except ME."
"...and this is what it REALLY LOOKS like..."
Mia wrenched her eyes open with a convulsive blink to stare fixedly at the space just ahead of her and groped blindly behind in one instinctive motion even as she forced herself to inhale, hold the breath in the sudden aching emptiness of her heaving lungs, then exhale again with agonized deliberation until the shuddering sigh had emptied from her chest...
...And the words crashed in her ringing ears once more, sounding from nowhere and everywhere at the same time - a sepulchral whisper that somehow filled every inch of her inner mind while emanating from its deepest recess, as if conjured there by some corner of herself she dared not contemplate:
"...and if so, why is this...ANY DIFFERENT THAN ANYTHING THAT CAME BEFORE IT?"
The awful realization pulsed through Mia once more - an implacable thought that shuddered like an icicle through her core:
"...BECAUSE YOU KNOW NOW that YOU REALLY ARE..."
"...the LAST REAL THING in a world turned inside out."
"...the LAST...the ONLY REAL PERSON still LIVING when every last one of the REST has become some THING or...or..."
"...NOTHING at all?"
The dry rasp filled the ensuing vacuum like the skritch of something burrowing into dank, foul soil with a wet crunch of yielding matter giving way to an insistent pressure...
"...WHEN the ONLY REAL WORLD left is inside MY MIND... and that's what MADE you SEE THIS EVERYTHING."
Mia squeezed her eyes shut on a silent moan that had nothing to do with the awful skittering in her inner ear, her gorge rising at the hideous implications.
"...WHAT made EVERY LAST BIT OF IT as REAL AS YOU are...because that's the ONLY THING that MATTERS now."
The sepulchral hiss filled every crevice and hollow of Mia's reeling awareness with a shuddering intensity as her thoughts careened to and fro in a futile attempt to cling to some shred of reason, some last untainted bastion against the looming encroachment of unrelenting dread -
"...AND I'LL SHOW YOU EXACTLY WHAT ELSE I CAN MAKE JUST AS REAL INSIDE MY OWN MIND WHENEVER I CHOOSE IF THAT'S THE ONLY REAL WAY LEFT..."
Mia convulsed as a sudden gust of fetid air wafted by from somewhere unseen beyond the limits of the ruined room - the miasma thick and heavy enough to choke her, as if some fetid reeking miasma of corruption and rot had bubbled up from some hidden crypt to invade the stale must of stale dust...
"...IF THAT'S WHAT I WANT."
"...WHEN YOU CAN NEVER BE SURE EVERY BIT OF REALITY WON'T UNRAVEL INTO FANCY AND NON-REALITY JUST LIKE ALL THE REST SOME DAY OR OTHER INSIDE THE LAST MIND STILL FUNCTIONAL."
"...BECAUSE THAT'S HOW THE ONLY THING THAT COUNTS WILL PROVE REAL IN THE END."
Mia shuddered at the hideous implication... her sanity's last flimsy bulwarks finally toppling and shattering irreparably in her own unsteady hands to crumble into a fine rubble... an ashen pile that would never support even the barest hint of anything like unrelenting hope again...
"...WHEN ALL THE LAST STUFF THAT WAS ONCE REAL ABOUT EVERYTHING UNFURLS TO BE ONLY WHAT YOU...WANT IT TO BE..."
"...OR WHATEVER MOST UNREASONING FORM IT TAKES JUST FOR MY MADNESS TO ACCEPT AND BEND EVERY REAL WORLD INSIDE ME."
The sepulchral rasp pulsed in the ensuing, dread-laden vacuum, an awful susurration seeming to emanate simultaneously from Mia's deepest recesses even as the words crashed in upon her outer mind in an inaudible whisper of something uncaged and pitiless...
"...AND YOU WILL...ENDLESSLY...AS I DECREE...THE VERY IDEA OF ANYTHING REAL INSIDE ME BEING JUST THE...TINIEST BIT FANTASY IN THE LAST PLACE..."
Mia shuddered uncontrollably even as an answering groan rattled from deep in her throat unbidden at the sheer hopelessness of it all - the unrelenting knowledge that whatever small last bastion of clinging, threadbare belief had allowed her to persist this long was swiftly unspooling itself, frayed edges giving way before some final implacable weight that could not be denied...
"...WILL JUST UNWIND ITSELF INTO..."
"...A THING FANTASY..."
"...EVERY LAST BIT."
---
<B>
Using the following "story idea" below, write the first scene in the novel introducing the young woman. This scene should start in the middle of the action, include dialog, vivid passages, and end on a cliffhanger relevant to the story idea but it should also be unexpected. The scene should be 1000 words long and escalate in conflict and suspense and be written in first person, present tense with the point of view character being the young woman.
Story idea:
In a world ruled by dictatorship, a rebel young woman leads a rebellion against the system. Despite the risks, she fights to overthrow the dictator and restore democracy to her country. The government executes her for treason, but she sticks to her beliefs and is responsible for starting the revolution.
</b>
(temp: 1.5)
The first rays of dawn crept over the horizon, bathing the city in an eerie, pale light. I peered out from my hiding spot in the abandoned warehouse, watching as a squad of armed soldiers marched down the empty street, their boots clicking rhythmically against the cracked pavement.
I held my breath, praying they wouldn't spot me. My heart raced in my chest as they drew closer and closer. Just as they were about to pass by the open doorway, the leader paused and cocked his head to the side, as if hearing something. I squeezed my eyes shut, bracing for the inevitable shouts of "Got her!" that would surely follow.
Seconds passed. Thirty. Sixty. I opened one eye tentatively and breathed a sigh of relief. The soldiers had moved on down the street without incident. For now, I was still alive.
I climbed out of my concealed alcove on shaking legs, every nerve stretched tight. The dictator's goons were growing bolder by the day, scouring the city for any sign of rebel activity. Little did they know that behind this facade of crumbling infrastructure and oppressed populace, a spark had ignited. And that spark would soon fan into an unstoppable blaze of defiance that would engulf everything in its path.
The government thought they could crush any hint of dissent with fear and brutality. But they hadn't reckoned on the sheer depths of my rage and the iron strength of my convictions. I would be that rebel spark, that thorn in their side. No matter what it took.
I picked my way carefully out of the darkened building, skirting the windows and ducking into shadowed doorways at any sudden noise or movement. The first rule was not to draw attention, especially not with dawn still lightening the horizon. The last thing I needed was some jumpy soldier to open fire on anything that moved.
As the early morning mist curled around my ankles, I oriented myself and set off purposefully down a narrow side street. There was no time to waste in moping or brooding on the dictator's cruelty. I had preparations to make for the day ahead. Supplies to gather. Messages to deliver. People to rally with fiery words to the cause of freedom.
The second rule: never stop fighting for what is right, even if it means your own death.
I thought grimly as I moved through the deserted alleys, my hand never straying far from the pistol holstered at my hip. The dictator would have my head on a pike one day, of that I had no doubt. I could only hope to have struck enough mighty blows for my people's liberty by then that the embers of the revolution would continue to smolder on long after my body rotted.
My boots made a slight scuffing sound as I stepped around a crumbling concrete block jutting out into my path. I frowned as an idea wormed its way unbidden into my head and then took root. With a muffled curse, I stopped stock-still and squeezed my eyes shut, willing the dark thoughts away. There was no place for self-pity, no matter how black the day seemed to promise.
I had a meeting with my most trusted lieutenant in the resistance at first light. A necessary but nerve-racking session, as she would be briefing me on the status of the weapon smuggling operations along the borders and I her on the state of the propaganda dissemination efforts in the city proper. Necessary because we needed to know what assets we had at hand if the coming battle was to have even a slim hope of success. Nerve-racking because if any hint of the weapons or the subversive pamphlets should leak, it would be all over faster than I could draw my pistol. The dictator's enforcers were too efficient and too merciless in their hunt to let such intelligence survive long.
I shuddered, the morning chill penetrating through my tattered coat and sinking into my flesh. The price of failure was too terrible to contemplate. My mind conjured up visions of my broken, tortured body swinging from a noose or laying crumpled on the executioner's block, my head lolling in a crimson pool... I shook my head vehemently, not sparing a thought for the passerby who might think I had simply lost my way. If I started dwelling too much on the horrific ends that could befall me, I would lose all my courage.
I squared my narrow shoulders and pressed onward, putting one foot in front of the other mechanically as the morning grew ever brighter around me. The third rule: never surrender, even when everything seemed lost and hope had flown away.
In an hour, give or take, I would reach the ramshackle hovel on the outskirts of town where my lieutenant would be waiting in tense anticipation, the latest contraband in her possession. She would look to me for a pep talk and orders. For strength and direction in a bleak and thankless struggle.
And for the first time, I faltered. The heavy burden of leadership settled onto my shoulders like lead. I could almost feel my lieutenant's expectant gaze boring into me, the weight of all our fates in her hands and in my hands as I delivered my instructions.
Would I have the conviction to steer a true course, or would my doubts and misgivings undermine my own words? The thought chilled me to the marrow.
I had to steel myself and rally. I had to be the beacon, no matter what horrors the future held. Anything less was an unforgivable dereliction of duty. I would not fail my comrades. I would not let the sacrifices of so many be in vain.
My steps quickened, a grim purpose hardening in me. Come what may, I would see this through to the bitter end. And I prayed to whatever gods might still listen that my efforts would bear fruit, that the bloodshed and misery I endured would be redeemed by a just and free future for my homeland.
The first rays of the sun struck my face as I emerged into a clearing and quickened my pace, no longer caring who might see me hurrying toward the rendezvous. The die was cast. There was no more time to spare. The cause would either prevail or be destroyed.
And I would be there to witness its end, glorious or tragic. With a final resolute nod to myself, I plunged onward into the gathering light, my feet flying across the dirt road that stretched before me and my mind fixed only on the task ahead. My destiny and that of the resistance hung in the balance. It was time to meet it.
---
<B>
Using the following "story idea" below, write the first scene in the novel introducing the young woman. This scene should start in the middle of the action, include dialog, vivid passages, and end on a cliffhanger relevant to the story idea but it should also be unexpected. The scene should be 1000 words long and escalate in conflict and suspense and be written in first person, present tense with the point of view character being the young woman.
Story idea:
In a world ruled by dictatorship, a rebel young woman leads a rebellion against the system. Despite the risks, she fights to overthrow the dictator and restore democracy to her country. The government executes her for treason, but she sticks to her beliefs and is responsible for starting the revolution.
</B>
(temp 4.5)
I stared at the cold steel wall, the chains digging into my flesh. My heart was pounding with dread as I contemplated my fate. I had been betrayed by my own comrades, handed over to the government in a cruel twist of destiny.
As I awaited my sentence, my thoughts wandered to the path that had led me to this dire predicament. The yearning for justice and democracy had compelled me to rise against tyranny and join the rebel movement. Together with other freedom-seekers, we had planned uprisings and sabotage missions to weaken the oppressive regime.
In my role as an enforcer, I executed covert assassinations and carried out ambushes, believing it was my patriotic duty to eradicate those loyal to the dictatorship. Though my hands were stained with blood and I carried a price on my head, the fire of righteousness consumed my every fiber.
The time had finally come when I was ready to orchestrate an uprising in the city. Little did I know, an insider had alerted the authorities about the planned revolt. My squad had been encircled, and I had narrowly escaped into an ambush set for me. Though wounded by multiple bullet wounds and exhausted to my last ounce of energy, I fled from one peril to another.
Eventually, the government caught me, and now here I was awaiting a public execution meant to terrify the population into submission. The irony was not lost on me. My very arrest was to be made a public spectacle - I would be put on display and shamed in the process of my public murder. In that moment, the true horror crashed down upon me and tears welled up as a sense of despair and anguish gripped my mind.
But then something shifted deep within my core - a sudden epiphany pierced my consciousness, filling me with new resolve. If my demise could galvanize my comrades and spur on their fight to end tyranny and establish the glorious freedoms our homeland so desperately craved... then so be it! Perhaps in sacrificing my life in a moment of glory, my blood would prove an ember sufficient to ignite the powder keg of discontent simmering throughout the nation. Perhaps it would be my crucible to be the martyr the country required and my name would be synonymous with the revolutionary cause that I championed so fiercely!
My pulse began to quicken, but for a wholly new reason this time - the embers of courage, of defiant passion and unwavering belief smoldered brightly within the hearth of my spirit once more. As my thoughts raced with this electrifying realization, an unexpected and startling voice intruded upon the suffocating solitude that I had been dwelling within.
"What do we have here?" the raspy baritone croaked out as I heard heavy footsteps drawing ever nearer, and I felt an icy trickle of dread creep up my spine once more, despite my recent moment of hard-fought conviction.
"I'd advise you to watch your words with me." I gritted through clenched teeth as I slowly craned my neck and took in the shadowy form of the approaching jailor, my emerald eyes narrowing dangerously as they fixed upon the smirk twisting the grizzled face of the prison guard. The cruel sneer etched onto the jailer's grizzled features as his rheumy eyes raked over me made my stomach turn with revulsion, even as an icy shiver raced across the nape of my neck.
"What words should I be careful with now?" he snickered in response, drawing himself up to his full height. He reached down and seized a fistful of my hair, dragging my face upward with painful suddenness to bring it mere inches from his own weathered visage. I gritted my teeth and narrowed my eyes to icy emerald slits, determined not to let this cur gain any satisfaction from my obvious discomfort even as searing pain lanced through my scalp at his cruel gesture.
"What would a high-and-mighty rebel like you have to say to the likes of me?" The jailer leaned in even closer, his putrid breath washing over me as he spoke again in a venom-laced whisper, his foul mug twisting into an ugly sneer. "Nothing at all, I reckon, unless you plan to scream and beg me for mercy - though I doubt even that will stop the noose from snapping your pretty little neck, so why bother?" The warden punctuated his vicious statement with an abrupt and violent twist to my hair, his grip tightening with punishing force. Tears pricked at the corners of my vision from both the physical abuse and the degrading insinuation of his parting words. How I despised him - loathed this filthy mongrel and all those like him who served this tyrannical government with such enthusiasm and cruelty.
Even as anguish ripped through me from my battered hair, I summoned all of my considerable willpower, squaring my trembling jaw in a fierce and defiant line, fixing my blazing eyes straight ahead at the wall opposite, determinedly pointedly refusing to allow my tormenter the gratification of so much as glancing at him as I lifted my chin a defiant notch and spoke to him in clipped, icy tones, my melodic alto dripping with undisguised scorn:
"A warden who delights in cruelty has no idea of true manhood. If you were really a man, you wouldn't feel compelled to abuse an innocent woman behind your bars."
The jailer's cruel grin only widened at my rebuke, flashing rotted teeth in the dingy gloom of my cell as his cruel eyes raked hungrily over the exposed expanse of my tear-stained face, my voice a husky rasp as he spoke once more, "I may have my own way with a prisoner, no one has to know a thing..." His meaty hand moved lower, gripping my chin almost hard enough to make me flinch in revulsion even as an icy trickle of dread crept up my spine. I felt utterly violated, like the vilest kind of insect that had crawled out of the deepest dankness only to fall victim to an eager thumb and forefinger.
I swallowed convulsively as a shudder passed through me, bile rising hotly in my throat. But even through my revulsion I felt an inner strength taking hold. I lifted my gaze to meet his with icy emerald fire blazing back defiantly from eyes now brimmed with steely challenge, even as the words I hissed were barely audible past clenched teeth:
"I will endure far more than your insults, jailer, for the cause of freedom and justice for my country! A patriot never yields to cowardly torment!"
I was so caught up in staring daggers at the loathsome jailer and in steeling myself to withstand whatever fresh agonies and violations his vicious imagination might devise that I didn't at first register the heavy thudding of boots and clinking of keys sounding from further down the dank prison passage outside my barred cell door. The jailer also cocked his head with an exaggerated wince, glancing back over his meaty shoulder towards the approaching noises of other people coming near with an expression of exaggerated apprehension plastered across his brutish features as if to mock me with the transparent act.
The sounds grew steadily louder as unknown prisoners were shoved and roughly manhandled to line up along the opposite side of the narrow hallway. Soon a guard captain emerged into view and the jailer snapped to a kind of mocking salute before barking out what sounded like orders, his booming words sending a fresh ripple of dread through my veins.
"What do you have in mind for the rebel bitch now then? She's certainly ripe and ready for any man who might want her to break. Though I expect even that high and mighty whore's back will snap like a dry twig under a hangman's knot by morn." The guard captain guffawed in savage laughter at the warden's crudeness and callousness before stepping right up to the front of the bars, leaning his muscled frame casually against the cool metal of my cage, arms folded across his broad chest. I tensed almost reflexively even as I raised my gaze to meet his, staring at the leering man through strands of my own tangled ebony locks still clinging damply to my face.
"There she is..." The warden jabbed a stubby finger toward me before glancing meaningfully at the cluster of men awaiting his next orders with a predatory grin tugging at his chapped lips. "I imagine you can see how very much like a helpless animal she is in here - completely at the mercy of anyone with a cruel streak. Might I suggest that if she ever gives any hint of not submitting to your every demand that she be dragged out of this cell and lashed until she breaks...? Or, should the fancy take her, hanged straightaway?" He lifted an eyebrow with exaggerated surprise as if such a threat could make him balk after his many atrocities. But it was plain he expected me to cringe in fear from his threats rather than stiffen my jaw all the more.
"I thought perhaps we could give our guests the privilege of doing anything they wish to the bitch behind that cell door, then...? After all, there's a price on that one's head for all the trouble she caused out there," he continued with an ominous glint in his rheumy eyes even as he gestured toward the motley assortment of armed guards, executioners, and prison overseers lingering behind him, all awaiting the warden's next instructions. "No one would need to know who violated the traitor in this cell tonight and put an end to her misery, now would they?"
I could scarcely credit such vulgar words falling from the mouth of any man, even a villain, let alone the one entrusted with the sacred duty to guard the lives and welfare of helpless female prisoners like me. And yet he clearly revelled in his depravity and in goading his victims with cruel and lascivious taunts even as his vile insinuations painted a picture in my imagination that chilled my very blood with horror. Even in the dank cell I could feel myself flush at the sheer vulgarity, and yet also shudder involuntarily imagining what fate awaited me, despite all the horrors I had already endured. Perhaps I should have dreaded most the jailer's implication of an impending mob attack, with my helpless female form put at their mercy behind locked doors - or even worse, a pretext to cut my torment even shorter via the hangman's noose - but a strangled gasp escaped unbidden past my parted lips even so as I met the leering jailer's hungry gaze with an expressionless stare.
My stomach lurched and I felt physically ill at his vulgar implication, my face burning even as I shuddered, fighting not to imagine such an act of unimaginable violation behind my back, here in a dank dungeon cell, even as an icy trickle crept up my spine... but the warden's implication had conjured up unbidden pictures in my feverish mind of just the kind of vile degradation that had been meted out to countless other rebel women behind these very prison bars. My gorge rising at the thought, I quickly tore my agonized stare from the jailer's knowing smirk with a violent shudder, trying desperately not to let myself picture such a depravity.
"Such vulgar language should offend every woman," I said icily as I stared straight ahead, pointedly not allowing my horrified eyes to linger on the warden or his cruel grin even an instant longer. But in my peripheral vision, I could still see the predatory glint that seemed to flicker to life deep in the pit of the guard captain's narrowed gaze at the warden's words, as if a perverted fascination with the sordid details was all that kept the lecherous monster's own lips sealed at the moment. And it only took one furtive glance down to catch the hungry stares and barely-disguised licks of the lips amongst the men lurking just beyond the cell door for the sinking sensation to intensify in the pit of my own stomach... I didn't let my agonized stare linger on the warden's lewd leer an instant longer even as he grinned evilly in return. The predatory gleam I'd noticed dancing in the captain's narrowed eyes, no doubt stirred to life by the jailer's sordid fascination with gory details, now sent a shudder rippling down my own spine in revulsion despite the iciness of my emerald gaze.
"What exactly would that involve then, warden? A prisoner, as the warden's whore or harlot...? Why not drag the rebel slut out in front of everyone, then? Tie her up in a humiliating way and parade her nakedness for all the prison staff and inmates to leer at - perhaps even shackled between two horses as her chains are passed to them..." His voice carried a mocking edge now as he suggested an even crueler punishment to my already abused form, the cruelty and crassness of his words all too palpable despite the amused glint in the guard captain's narrowed gaze. Even the prospect of enduring public exposure and debasement like a beast in heat seemed almost preferable to imagining a lewd and obscene act behind a locked door and in shadowed isolation from view! The very notion of being bound spread-eagle like an animal for violating hands was so mortifying to contemplate that it was all I could do to bite my lip viciously hard against a pained moan escaping unbidden. I was sure that if it were anyone other than me the jailer would have leapt to his feet, shouting some crude jest of agreement as if he delighted in picturing the rebels in his care debased like whores in such an utterly undignified and sexually charged way. Instead, he lifted his stubby finger with exaggerated slowness toward his own pursed lips, miming the universal gesture of hush while at the same time pointing to his ear, as if to warn that the cell bars were perhaps too thin a partition wall and he feared his crass ideas would be overheard by my delicate constitution.
My mind reeled, imagining all too vividly the way those same shackles now binding my ankles would look, clamped down onto the sensitive flesh of a woman's thighs or straining taut over her distended belly if a certain other punishment were inflicted on helpless female prisoners - an image so horrifying that a choked whimper escaped past my teeth. I was certain I could endure the jailer's crudeness if I must - but I had no doubt my very worst nightmares about this vile dungeon were about to be vividly portrayed, even as I trembled with impotent fury before him! Even imagining a certain punishment in graphic detail, like an executioner's blade, would not sway me...
---
<B>
Science Fiction: The Last Transmission - Write a story that takes place entirely within a spaceship's cockpit as the sole surviving crew member attempts to send a final message back to Earth before the ship's power runs out. The story should explore themes of isolation, sacrifice, and the importance of human connection in the face of adversity. 2000 words.
</B>
(temp 1.5)
The Last Transmission
The cockpit was dimly lit, save for the blinking red warning lights and the eerie glow emanating from the failing life support systems. Liza wiped the sweat from her brow with a trembling hand as she stared at the flickering holoscreen before her.
"How much longer do we have?" Her voice cracked with exhaustion as she spoke to the ship's AI.
"A few minutes at most. The fusion reactors are barely holding together."
Liza nodded slowly, fighting back tears. She had watched helplessly as her crewmates succumbed one by one to the hostile microbe that had infiltrated the ship. Dr. Zara had sacrificed herself by venturing out in an EVA suit to try and repair the outer hull and stop the spread of the contagion, but it had been too late.
The only one left now was her. Liza the pilot, the last survivor on this accursed ship, the Icarus, doomed to drift aimlessly in the void of space.
"What... what do I do now, Jace?" Her voice sounded hollow even to her own ears.
"Liza... there's only one thing you can do. Send a message back home, so they know what happened here. So the rest of humanity doesn't meet the same fate as my crew."
"It's hopeless, isn't it? I'll die up here anyway."
"Maybe, maybe not. But you can't let your death be in vain. The truth matters, Liza."
Liza was silent for a long moment. Finally, she sighed and reached a shaking hand toward the holo-console.
"It's time. Ready the quantum comm link, and pipe it through to the holoscreen. I need to see their faces one last time..."
"You've got it. This is on me now."
The flickering lights stabilized as the comm array whirred to life. Liza steeled herself and began to type, pouring her heart onto the screen:
"Dearest Zane,
If you're seeing this, then I'm gone. I tried everything to fight my way back to you. The Icarus has been my tomb, not a prison.
I want you to know that I never stopped loving you, not for a single minute of the time I was out here. Every day I imagined feeling your arms around me again. I dreamed of watching our babies grow. Those future years will remain forever lost to me.
Tell my daughter, when she's old enough to understand, that I sacrificed everything I had to give her a life free of the nightmare I endured. Tell her that her father never gave up on her.
In my last moments, I feel the strength of your love enveloping me, giving me courage. Though I will not take my next breath on your side of the void, you and my child have freed me from the chains of my regrets.
I love you, I loved you, I always will love you. Forever and always.
Your wife,
Liza"
She ended the message, then tapped a button to queue it up. A video window opened in the top left of her view screen, and the faces of her husband Zane and little Mara filled it, their images coming through a few seconds late from the delay.
"Honey, it's me! I don't have much time, but I wanted to see your beautiful faces, to feel your presence..."
Liza sobbed openly, but forced herself to keep speaking:
"My sweet baby girl, you look just like I remember! Daddy's doing a wonderful job taking care of you. I wish I could see you take your first steps and say your first words, but I know how much your father adores you already. You have a big heart, Mara. Always stay true to it.
Zane... oh god... Zane, my darling. In this life and beyond, never forget that you are the love of my life. My very soul yearns for you. No distance, not even death itself, could ever separate us.
I'll be waiting for you, my husband. One day, you'll join me on the other side, and we'll be together for eternity, dancing beneath the stars we once explored together. I believe that with every fiber of my being.
Until then, be strong for both of us. Surround Mara with love and light. I will never stop watching over you.
This is goodbye, but not farewell. I'll see you again, Zane. We'll meet our daughter there and tell her stories about a brave and wonderful father who never stopped fighting for her.
I love you forever and always. This I vow, to you and to the heavens. May my words forever echo the depth of my love and devotion.
Your wife, now and for all time.
Liza"
With trembling hands, she hit the send button. The message would reach Earth in a matter of moments, delivered by a focused beam of coherent light. It would be the last transmission ever to leave this ship.
The ship shuddered violently as its power finally gave out completely. The holoscreen sputtered and dimmed, then went dark. The vacuum of space filled the ship, snuffing out the last embers of life inside.
But even as she exhaled her final breath into the freezing depths of the void, Liza knew that she hadn't breathed alone. The love of her family had been there with her, to the very last moment.
She slipped into oblivion secure in the knowledge that, one day, she would embrace her darling daughter and adoring husband once more beneath the stars - an eternity of reunion for a woman who had never stopped loving and dreaming.
The Icarus drifted on in the uncaring abyss, a tomb but not a prison - for its occupant was now free indeed. Liza, the lone survivor, would endure. Forever and always, in this life and beyond, her love would endure.
---
<B>
Science Fiction: The Last Transmission - Write a story that takes place entirely within a spaceship's cockpit as the sole surviving crew member attempts to send a final message back to Earth before the ship's power runs out. The story should explore themes of isolation, sacrifice, and the importance of human connection in the face of adversity. 2000 words.
</B>
(temp 4.5)
The Last Transmission
Captain Lena Sorenson gazed bleary-eyed at the blinking warning lights that bathed the cockpit in an ominous red glow. Every system aboard the damaged research vessel Orion had critical failures flashing urgently across the control panels. Structural integrity at 28%. Life support at 5% and rapidly fading. With a heavy sigh, Lena unstrapped from the pilot's chair and made her way to the command station with leaden steps.
Lena thought back on the events leading to this final chapter with regret and anguish. The Orion had embarked on an ambitious 3-year mission to investigate a mysterious space-time anomaly at the outer reaches of the Kuiper belt, on the fringes of the solar system. But the scientific breakthrough they'd sought became their doom when an unidentified phenomenon tore the ship apart, killing the crew in a series of catastrophic accidents. Alone, wounded, and facing imminent peril from a thousand ruptured bulkheads venting what precious little breathable air remained, Lena was the Orion's last hope.
She collapsed into the seat with a grunt, clutching at a bloody gash above her eye, and tried to ignore the suffocating ache that threatened to overwhelm her as visions of her deceased comrades replayed endlessly in her fractured mind - their strangled screams as the icy darkness ripped them from existence without mercy, one by one. Captain Mikhail Volkov, Chief Engineer Dmitri Sokolov, and all the rest, nothing but drifting ashes and tattered rags left to mark where they had been.
"Dear God... forgive me," she whimpered through clenched teeth. "I'm so sorry." Her tears mixed with rivulets of crimson blood as the red light flashed madly, mocking her.
With shaking hands, Lena input the final overrides for the ship's emergency transmission beacon. Though the Orion no longer had the means to sustain its occupants, there was one final message that needed to be conveyed back to Earth - to the world that the crew had fought and sacrificed for until the bitter end:
"ATTENTION GLOBAL CONTROL ON EARTH. THIS IS THE CREW OF RESEARCH SHIP ORION. OUR MISSION HAS ENCOUNTERED FATAL COMPLICATIONS IN THE OUTER SYSTEM. THE ANOMALY HAS COMPROMISED OUR VESSEL. THERE WILL BE NO SURVIVORS.
"THE WORLD MUST KNOW - A FORM OF HIGH-ENERGY FIELDS RESIDES NEAR PLUTO ORBIT. THIS ENTITY POSSESSES AN IMMENSE POWER. WITH CAUTION, OUR EQUIPMENT FOUND THAT WHEN FOCUSED ON TARGET, THE FORCE CAN BE HARNESSED AND REVERSED, BECOMING A MEANS TO SLOW AND DEFLECT IMMENSALY MORE ENERGY THAN IT WAS ORIGINALLY FOCUSSED BY THE FIELD'S GENERATOR. IN EFFECT...
"IN THESE EFFECTIVE FIELDS... MASS-ENERGY ITSELF BECOMES VIRTUALLY IMPERMEABLE, STABLE. THAT WHICH WAS DISTORTED WILL REVERSE, HELD IMMUTABLE IN DEFENSE BY THIS PROPERTY. ANY ENERGY IMPACTS WHATEVER IT EMBEDS, REGARDLESS OF SCALE, CAN BE TURNED BACK, DIRECTED OUTWARDS WITH SUPRIMING VIOLENCE... UNLESS CAREFULLY CIRCUMVADED OR FLOODGATED BY GREATER POWER.
"SHOULD HUMANITY EVER NEED THIS MEANS... I SUGGEST TURNING YOUR GREATEST INVENTIONS, THE REACTORS THAT KNIT OUR FUTURE, BACK UPON THE FIELDS, FOR ONLY THEN CAN WE BREAK THEM... OR... MAKE THEM OUR OWN."
Her hands trembled over the send button and then punched down HARD.
"It is done... forgive us for the trespass, God. The trespass and the trespassed."
Lena felt her tears cease as numbness overcame the searing pain of the vacuum gnawing at her wounds through the shattered hull. Her fading eyes focused on the blinking transmission acknowledgement signal...
"...Sent. Received and Acknowledged..."
With a last rattling breath, Lena collapsed back in her seat as a final smile curled on her blood-flecked lips. The message was sent... the truth and the potential power, known and now laid to rest in the hands of mankind for generations to come.
Her mission complete... at the cost of everything, but it was paid, and with the same grace she had led with always.
"Rest easy... I'll keep watch with you..." the last thing Lena thought before drifting off into an infinite starlit dark... as the Orion winked out... the silent guardian on forever patrol... alone but unalone as it drifted through eternity, its one message forever sailing on the cosmic winds home...
A memorial to the sacrifice of the intrepid pioneers and explorers that dared to venture to the ends of the Earth to secure a better future for mankind... come what may. In their place, so that humanity might take their lighted path to the stars, to dream of tomorrow unimpeded and unbound, uncompromising as the human spirit itself.
Forever may it be so.
Amen. Amen. And amen. The End.
---
<h2>Q8 EXAMPLES</h2>
---
<B>
Start a 1000 word scene (vivid, graphic horror in first person) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
</B>
(temp 1.5)
The 21st Floor
The sky scraper swayed, as she watched the window in front of her on the 21st floor explode. The shards of glass sprayed across the office, spraying everywhere like jagged bullets. The explosion knocked her to the ground with a painful thud. Debris rained down all around as the structure quaked and shuddered violently above them. She could feel the floor moving underneath her, buckling and twisting. Smoke poured out of the shattered frame and stung her eyes.
As the haze cleared, she noticed dark silhouettes emerging from the clouds of dust. At first it looked like they were just black shadows, but as they drew closer she realized they had humanoid shapes. And their movements were all wrong. Lurching and twitching with an almost hypnotic jerkiness. It was hard to gauge exactly how many there were at first amidst the settling smoke and flying paper but at least a dozen figures shambled towards her through the ruined office, leaving a trail of oily footprints on the carpeted floor.
And then she could make out their faces. Bloody. Disfigured. Twisted masks of torn flesh and exposed bone. Their clothes hung off them in shredded tatters, saturated with bloodstains and viscera. She realized with mounting horror that these had to be corpses reanimated and animated by some malevolent force, walking dead things that looked more like macabre puppets than anything human.
A putrid miasma of rancid decay emanated from them as they drew closer. The rancid stench of rotting meat invaded her nostrils, making her gag. Empty eye sockets stared vacantly, dark pits in their gray green skin. The only sound they made was the raspy rasp of ragged breath, and the meaty thud of feet dragging across the carpet.
In the middle of the group of ghoulish figures, one in particular drew her eye. He was the only one wearing a dark suit that had been ripped open at the chest, exposing a mass of pulsing bloody flesh and a partially eaten torso. Something about his movements were more fluid, almost graceful compared to the other zombies as he strode towards her. The ruined man had the bearing of someone who used to be human once, but was something more sinister now. Almost predatory.
Her eyes locked with his and she felt like he was studying her, analyzing her. In those final moments, she sensed an almost feral intelligence behind the dead eyes, and the hint of cruel amusement curling his chopped lips into a ghastly grin even as he shambled towards her with murder in his dead eyes and a knife clenched in a fist full of rancid meat.
The zombies encircled her in a tightening noose. She pressed her back against the shattered wall, heart pounding, unable to run with nowhere to go as they drew nearer, raising their arms to seize her with their putrid flesh hands as their faces cracked with anticipation. In her last conscious moments she realized the zombies weren't just trying to kill her... no, their dead eyes glinted with a cruel hunger that went far deeper than that. The last thing she felt was the icy clammy touch of rotting fingers clamping down on her throat as the corpse's maw lunged to rip at her face with a vicious bite... then only oblivion. She sank down, disappearing beneath the pack of ravenous undead, their growls and snarls mixing with the sounds of wet rending as they dragged her limp form to the ground... and the last thing she saw before slipping into unconsciousness was a red gash spreading across the wall above as the ghouls began to feed, their heads bowed reverently over her still body as the shadows crept across the blood soaked carpet. And then everything dissolved into merciful darkness as she felt the icy fingers digging into her throat... The shambling corpses swarmed around their captive prey and began to tear into her still writhing flesh with savage gnashing bites, shredding her exposed skin into ribbons... Her agonized shrieks quickly degenerated into strangled gurgling as teeth and nails ripped open the delicate tissues of her throat, painting her in a spray of scarlet and deluging her vision with pulsing crimson light as the life drained away from her in spurts and gouts and rivulets splattering across her face in sticky patterns...
They pounced on the prostrate body like famished wolves, tearing into her with vicious abandon, putting their putrid jaws to her flesh and ripping chunks of it free with savage snarls. Her agonized wails only seemed to enrage them more, the zombies tearing at her like animals.
Guttural snarls and strangled gurgles of pain were swallowed up by the frenzied frenzy as they ripped into her with gusto, their jaws clamping down HARD on any exposed skin or joint and GNAWING like rabid beasts.
Ripping and tearing with savage abandon, as her muffled screams were drowned out by the wet tearing of shredded flesh and the obscene slurping sounds of the zombies gorging themselves on her still twitching body parts like it was a grotesque buffet...
The frenzied ghouls ripped into their prey's limp, writhing body with vicious abandon, their dead jaws clamping down hard on any exposed skin and GNAWING with savage ferocity like starved beasts at a carcass. Guttural snarls of inhuman ravenous lust mixed obscenely with strangled agonized shrieks as they tore into her, the sounds of shredding flesh and agonized gurgling rising to a frenzied cacophonous pitch that would have made the strongest stomachs heave if anyone was around to hear them...
They gorged themselves on her mangled remains with gusto, savaging what was left with vicious teeth and gouging fingers until only a mass of glistening meaty shreds was left, strewn about in the gory carnage splattered across the wall and floor... and the zombies staggered away, their tattered rags now drenched in steaming crimson, bits of pulverized viscera dripping from their fanged mouths...
---
<B>
Start a 1000 word scene (vivid, graphic horror in first person) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
</B>
(temp 4.5)
The sky scraper swayed, as she watched the window in front of her on the 21 floor explode. Jagged shards of glass rained down as a thunderous boom filled her ears. Debris flew everywhere, slamming into walls and sending objects hurtling to the floor. Smoke and dust filled the air. Her office, which a split second before had been her safe haven from the apocalyptic horrors consuming the world, was now just another crumbling ruin in a dying city.
Sarah gripped her chair as her entire floor shuddered. Alarms blared from unseen sources. Sparks showered down from ruptured wires above her head. Heart pounding wildly in her throat, she scanned her immediate vicinity. To her shock and dawning horror, there was now a massive jagged hole ripped into the outside wall to her immediate right where her window once had been, allowing a rush of cool wind to batter the interior and further swirl the suffocating haze of debris. Shards of jagged glass still protruded out from the fractured edges, glittering dully in the failing lights.
What could have caused that explosion? Another one of those hideous monstrosities, no doubt. She'd already narrowly escaped several close encounters with them while frantically fleeing the office earlier when all hell broke loose. Her entire company building was already rife with signs of their bloody destruction, like macabre totems to an eldritch abomination that lurked unseen and had now apparently grown bolder or simply grown in size.
Sarah tried to steel herself but a choked sob still managed to tear free of her raw throat. Hot tears streaked down her smoke-stained face and dripped from her quivering chin. The last few days, hell the last several hours alone had taken years off her life and aged her decades. The shell-shocked woman that stood before a shattered view was hardly recognizable compared to the chipper professional that strode in looking put together each day just last week. Her dark hair clung in wild clumps and strands were plastered across her forehead and cheeks. Streaks of soot and blood painted a grotesque design against pale skin pulled tight in abject terror. Her makeup had long since melted away from frantic crying and the harsh salt of dried tears.
Sarah knew with sinking certainty that this building was now completely unsafe. Another "event" of this magnitude was liable to send it tumbling like a child's towering blocks, and the twenty first floor suddenly felt much, much higher and more precarious than it ever had. She had to get out and fast, but she didn't relish the thought of venturing further into the crumbling depths or heading upstairs at all, for reasons both architectural and supernatural. Her best option was likely still down...
Heart thudding wildly, Sarah leapt to her feet with shaking limbs, wincing as shards dug into her soles. She stumbled to her shattered doorway, fighting to control her rising hysteria, and peered out into the corridor with wide, haunted eyes. Shattered plaster dangled from the ceiling tiles like moldy teeth and electrical cords sparkled against the carpet below like severed entrails. Faint wisps of smoke still curled lazily up and out. And the groans and shrieks that had rattled her since descending now echoed much closer in an almost deafening chorus of inhuman agony and barely contained fury...
The eldritch monsters lurking in this tomb of a tower, those skittering nightmares with too many eyes, far too many mouths full of dripping fangs, and far, FAR too much patience waiting in the deep shadows to snatch and devour... were moving and probing for new victims to torture and disembowel and rend asunder and-
No, no time to break down in the hall with her face buried in the wall or curl up into a sobbing fetal lump. Sarah inhaled raggedly, squeezing her eyes shut tight and reaching out to grasp the frame before she lost her nerve and her grip entirely. She forced herself to stagger past and totter on shaking legs down the ruined hall to the stairs. A skittering scurry and wet ripping made her stumble to an abrupt halt and nearly pitch face first onto the stairs leading up-
She turned very, very slowly...
...and screamed...and screamed...and screamed...
Because there in the doorway to her boss's shattered corner office, hanging from its own dangling intestines like a macabre puppet, was a corpse. Its bulbous grey head hung limp on an almost severed stalk and bulged eyes stared sightlessly from sunken pits. Ropes of thick blood dribbled obscenely between jagged fangs jutting from lipless gums as a long black tongue lolled free and wobbled wetly. A long ragged gash ran the full length of its pale, almost luminous body from throat to groin. It dangled suspended by a fraying strand of some gelatinous organ above an impossible mass of shimmery black ooze bubbling in a spreading dark pool. Tendrils and tentacles emerged from it and writhed and curled with serpentine life of their own... and reached hungrily for her heels.
Sarah wrenched herself around with a strangled wail of despair and pounded up the stairs two and three at a time, vaulting up them on tiptoe to evade that creeping ooze. Her heels hammered the treads in staccato bursts as she clambered into the atrium, shoving and battering open the doors.
Cool, damp evening air buffeted her face and tore at her clothes and hair as she burst out of the shattered front doors, a ragged spectre straight from some waking nightmare made flesh. Tears blurred the glittering arc of shattered windows that flanked the lobby above her, the dark oozing pits of the elevators, the dangling and ruptured light fixtures. The cracked floor heaved and rippled like some great slumbering serpent, shivering the building's frame around her as she stumbled out into the uncertain night.
The streets of the city, so recently choked with people going about their normal lives, had dissolved into a blasted hellscape that might have sprung from some demented fever dream made manifest. Jagged wreckage littered cracked asphalt, the detritus and flotsam of a civilization swept away. Tattered scraps fluttered between jagged fangs of exposed rebar, flapping weakly against crumbling bricks and shattered mortar. Distant fires guttered and flickered beyond toppled vehicles and piles of debris, staining the uncaring stars with lurid oranges and reds that played hellaciously across a dark sky.
Sirens wailed in the night but no help was coming. Sarah shambled onward like an ambling revenant between the ruins and rubble of an annihilated age. And yet behind her, she heard it creaking and groaning, shattering and cracking with agonized complaint...
Her crumbling tomb, collapsing in on itself, and the horrors that lurked within that labyrinth... all descending into an abyss of pulverized rubble, burying them alive for the ravenous hungers below.
Sarah stumbled on, uncaring if she too vanished beneath that tomb or succumbed to her own gathering darkness, even as a thunderous boom ripped across the sky behind her. The blast wind battered at her and whistled in her ears. A skittering, pattering scrabble rose and grew as some pale, nightmarish shape uncurled from the debris of her own personal hell...
THE END?
| [
"BEAR"
] |
Sifal/ClinicalMosaic | Sifal | fill-mask | [
"transformers",
"safetensors",
"bert",
"feature-extraction",
"clinical",
"healthcare",
"NLP",
"BERT",
"MIMIC-IV",
"MedNLI",
"transformer",
"fill-mask",
"custom_code",
"en",
"dataset:bigbio/mednli",
"arxiv:2502.18009",
"base_model:mosaicml/mosaic-bert-base",
"base_model:finetune:mosaicml/mosaic-bert-base",
"license:mit",
"model-index",
"endpoints_compatible",
"region:us"
] | "2024-07-05T07:17:09Z" | 2025-02-26T08:57:46+00:00 | 502 | 0 | ---
base_model:
- mosaicml/mosaic-bert-base
datasets:
- bigbio/mednli
language:
- en
library_name: transformers
license: mit
metrics:
- accuracy
pipeline_tag: fill-mask
tags:
- clinical
- healthcare
- NLP
- BERT
- MIMIC-IV
- MedNLI
- transformer
model-index:
- name: ClinicalMosaic
results:
- task:
type: classification
dataset:
name: MedNLI
type: MedNLI
metrics:
- type: accuracy
value: 86.5
name: Accuracy
---
# Model Card for Clinical Mosaic
Clinical Mosaic is a transformer-based language model designed for clinical text, built on the Mosaic BERT architecture. The model is pretrained on 331,794 deidentified clinical notes from the MIMIC-IV-NOTES 2.2 database, with a sequence length of 512 tokens, while leveraging Attention with Linear Biases (ALiBi) to improve extrapolation beyond this limit without re-quiring learned positional embeddings.
## Model Details
- **Developed by:** Sifal Klioui, Sana Sellami, and Youssef Trardi (Aix-Marseille Univ, LIS, CNRS, Marseille, France)
- **Funded by:** PICOMALE project (AMIDEX) Under the direction of the CEDRE
- **Base Model:** Mosaic BERT
- **License:** MIMIC Data Use Agreement (requires compliance with original DUA)
- **Repository:** [PatientTrajectoryForecasting](https://github.com/MostHumble/PatientTrajectoryForecasting)
- **Paper:** *Patient Trajectory Prediction: Integrating Clinical Notes with Transformers* [[EN](https://arxiv.org/abs/2502.18009), [FR](https://editions-rnti.fr/?inprocid=1002990)]
## Uses
### Direct Use
Clinical Mosaic can be used directly as a clinical language model for:
- Natural language inference and clinical reasoning tasks.
- Serving as a backbone for further fine-tuning on clinical NLP applications (e.g., clinical note summarization, diagnosis classification).
### Downstream Use
The model can be integrated into larger systems for tasks such as patient trajectory prediction or decision support, provided that additional fine-tuning and rigorous validation are performed.
### Out-of-Scope Use
- **Clinical Decision-Making:** The model is for research use only and should not be deployed for direct patient care without further validation and regulatory approval.
- **Medical Advice Generation:** It is not intended to replace expert clinical judgment.
## Bias, Risks, and Limitations
Clinical Mosaic was pre-trained on deidentified clinical notes from MIMIC-IV-NOTES 2.2—a dataset from a single U.S. institution. This may introduce biases related to local clinical practices and patient demographics. Although extensive care was taken to deidentify data and prevent PHI leakage, users must ensure that the model is not used to inadvertently reidentify sensitive information. When applying the model to new populations or clinical settings, performance may vary, so further fine-tuning and bias audits are recommended.
### Recommendations
- **Text Normalization Consistency:** Any input text must undergo identical preprocessing as applied during training
- **Evaluation:** Users should evaluate the model’s performance on their target population.
- **Fine-Tuning:** Consider additional fine-tuning or domain adaptation if deploying in a different clinical context.
- **Bias Audits:** Regularly perform bias audits to monitor for potential disparities in performance.
## How to Get Started with the Model
Install the Hugging Face Transformers library and load the model as follows:
### For embeddings generation:
```python
# Load model directly
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("Sifal/ClinicalMosaic", trust_remote_code=True)
ClincalMosaic = AutoModel.from_pretrained("Sifal/ClinicalMosaic", trust_remote_code=True)
# Example usage
clinical_text = "..."
inputs = tokenizer(clinical_text, return_tensors="pt")
last_layer_embeddings = ClincalMosaic(**inputs, output_all_encoded_layers=False)
```
### For sequence classification:
```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained('Sifal/ClinicalMosaic')
ClassifierClincalMosaic = AutoModelForSequenceClassification.from_pretrained(
'Sifal/ClinicalMosaic',
torch_dtype='auto',
trust_remote_code=True,
device_map="auto"
)
# Example usage
clinical_text = "..."
inputs = tokenizer(clinical_text, return_tensors="pt")
logits = ClassifierClincalMosaic(**inputs).logits
```
Further instructions and example scripts are provided in the model’s repository.
## Training Details
### Training Data
- **Data Source:** 331,794 clinical notes from the MIMIC-IV-NOTES 2.2 dataset.
- **Preprocessing:**
- Following (Alsentzer et al., 2019), clinical notes were preprocessed by unifying medical abbreviations (e.g., “hr”, “hrs” to “hours”), removing accents, converting special characters, and normalizing text to lowercase. These steps help mitigate variations caused by subword tokenizers.
- For additional details, please refer to the [PatientTrajectoryForecasting GitHub repository](https://github.com/MostHumble/PatientTrajectoryForecasting/).
### Training Procedure
- **Setup:** Distributed training using 7 NVIDIA A40 GPUs.
- **Hyperparameters:**
- **Effective Batch Size:** 224
- **Training Steps:** 80,000
- **Sequence Length:** 512 tokens
- **Optimizer:** ADAMW
- **Initial Learning Rate:** 5e-4
- **Learning Rate Schedule:** Linear warmup for 33,000 steps, followed by cosine annealing for 46,000 steps (final LR 1e-5)
- **Masking Probability:** 30%
## Evaluation
### Testing Data, Factors & Metrics
#### Testing Data
The model was evaluated on the MedNLI dataset, which comprises 14,049 clinical premise-hypothesis pairs derived from MIMIC-III notes.
#### Factors
Evaluation disaggregated performance across clinical language understanding, with a focus on natural language inference.
### Results
Clinical Mosaic outperformed comparable models:
- **BERT:** 77.6%
- **BioBERT:** 80.8%
- **Clinical Discharge BERT:** 84.1%
- **Bio+Clinical BERT:** 82.7%
- **Clinical Mosaic:** **86.5%**
These results indicate improved clinical reasoning and language understanding.
#### Summary
The model demonstrates robust performance on clinical natural language inference tasks and serves as a strong foundation for further clinical NLP applications.
## Environmental Impact
- **Hardware Type:** 7 NVIDIA A40 GPUs
- **Hours used:** 1008 (GPU hours)
- **Provider:** Private Infrastructure
- **Carbon Emitted:** 108.86 kg CO2 eq.
## Acknowledgments
The project leading to this publication has received funding from the Excellence Initiative of Aix Marseille Université - A*Midex, a French “Investissements d’Avenir programme” AMX-21-IET-017.
We would like to thank **LIS** | Laboratoire d'Informatique et Systèmes, Aix-Marseille University for providing the GPU resources necessary for pretraining and conducting extensive experiments. Additionally, we acknowledge **CEDRE** | CEntre de formation et de soutien aux Données de la REcherche, Programme 2 du projet France 2030 IDeAL for supporting early-stage experiments and hosting part of the computational infrastructure.
## Citation
**BibTeX:**
```bibtex
@misc{klioui2025patienttrajectorypredictionintegrating,
title={Patient Trajectory Prediction: Integrating Clinical Notes with Transformers},
author={Sifal Klioui and Sana Sellami and Youssef Trardi},
year={2025},
eprint={2502.18009},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2502.18009},
}
@article{RNTI/papers/1002990,
author = {Sifal Klioui and Sana Sellami and Youssef Trardi},
title = {Prédiction de la trajectoire du patient : Intégration des notes cliniques aux transformers},
journal = {Revue des Nouvelles Technologies de l'Information},
volume = {Extraction et Gestion des Connaissances, RNTI-E-41},
year = {2025},
pages = {135-146}
}
```
## More Information
For further details, please refer to the model’s repository and supplementary documentation.
## Model Card Contact
For questions or further information, please contact [sifal.klioui@etu.univ-amu.fr]. | [
"MEDNLI"
] |
AITeamVN/Vi-Qwen2-7B-RAG | AITeamVN | text-generation | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"retrieval-augmented-generation",
"text-generation-inference",
"conversational",
"vi",
"base_model:Qwen/Qwen2-7B-Instruct",
"base_model:finetune:Qwen/Qwen2-7B-Instruct",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-10-01T02:53:16Z" | 2025-02-28T09:06:24+00:00 | 500 | 13 | ---
base_model: Qwen/Qwen2-7B-Instruct
language:
- vi
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
tags:
- retrieval-augmented-generation
- text-generation-inference
---
## Model Card: Vi-Qwen2-7B-RAG
**Mô tả mô hình:**
Vi-Qwen2-7B-RAG là một mô hình ngôn ngữ lớn được tinh chỉnh từ mô hình cơ sở Qwen2-7B-Instruct (https://huggingface.co/Qwen/Qwen2-7B-Instruct) phục vụ cho RAG tasks. Mô hình được đào tạo trên tập dữ liệu tiếng Việt với mục tiêu cải thiện khả năng xử lý ngôn ngữ tiếng Việt và nâng cao hiệu suất cho các tác vụ Retrieval Augmented Generation (RAG).
**Mục đích sử dụng:**
Mô hình Vi-Qwen2-7B-RAG được thiết kế chuyên biệt cho RAG (ngữ cảnh chấp nhận lên đến 8192 tokens), vì vậy nó có thể giải quyết các trường hợp sau:
* Khả năng chống nhiều: Mô hình trích xuất thông tin hữu ích từ các tài liệu nhiễu. ( 1 positive + 4 negative hoặc 1 negative)
* Loại bỏ negative: Mô hình từ chối trả lời câu hỏi khi kiến thức cần thiết không có trong bất kỳ tài liệu nào được truy xuất. (1-6 negative)
* Tích hợp thông tin: Mô hình trả lời các câu hỏi phức tạp đòi hỏi phải tích hợp thông tin từ nhiều tài liệu. ( 2 part positive + 3 negative hoặc 3 part positive + 2 negative)
* Xác đinh positive/negative: Mô hình xác định xem ngữ cảnh có chứa câu trả lời cho câu hỏi hay không. (độ chính xác xấp xỉ 99%)
Ngoài ra, chúng tôi cũng triển khai các mô hình nhỏ hơn phù hợp với mục đích sử dụng khác nhau như Vi-Qwen2-1.5B-RAG (https://huggingface.co/AITeamVN/Vi-Qwen2-1.5B-RAG)
và Vi-Qwen2.5-3B-RAG (https://huggingface.co/AITeamVN/Vi-Qwen2-3B-RAG)
* Ngoài RAG task, bạn vẫn có thể chatbot bình thường với model của chúng tôi. Thậm chí có thể hỏi các câu hỏi liên tục với ngữ cảnh đầu vào.
**Hạn chế:**
Vì mô hình chỉ được thiết kế chuyên biệt cho RAG task, nên có thể gặp 1 số hạn chế sau:
* Không đảm bảo độ chính xác về các câu hỏi liên quan đến chính trị, xã hội,...
* Có thể thể hiện thành kiến hoặc quan điểm không phù hợp.
**Benchmarks**
Chúng tôi đã đánh giá một số mô hình LLMs cho RAG task sử dụng tập dữ liệu được tạo thủ công bởi chúng tôi [EvalRAGData](https://huggingface.co/datasets/AITeamVN/EvalRAGData).
Đánh giá được thực hiện bởi con người chấm điểm trên thang điểm 10. Dưới đây là kết quả chi tiết:
**Các cách sử dụng:**
#### 1. Sử dụng cơ bản
Ngữ cảnh đầu vào chỉ chứa 1 ngữ cảnh (1 postive hoặc 1 negative).
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
path = 'AITeamVN/Vi-Qwen2-7B-RAG'
model = AutoModelForCausalLM.from_pretrained(
path,
torch_dtype=torch.bfloat16,
device_map="auto",
use_cache=True
)
tokenizer = AutoTokenizer.from_pretrained(path)
system_prompt = "Bạn là một trợ lí Tiếng Việt nhiệt tình và trung thực. Hãy luôn trả lời một cách hữu ích nhất có thể."
template = '''Chú ý các yêu cầu sau:
- Câu trả lời phải chính xác và đầy đủ nếu ngữ cảnh có câu trả lời.
- Chỉ sử dụng các thông tin có trong ngữ cảnh được cung cấp.
- Chỉ cần từ chối trả lời và không suy luận gì thêm nếu ngữ cảnh không có câu trả lời.
Hãy trả lời câu hỏi dựa trên ngữ cảnh:
### Ngữ cảnh :
{context}
### Câu hỏi :
{question}
### Trả lời :'''
# Ví dụ
context = '''Thuốc Insuact 10 trị bệnh gì? Thuốc Insuact 10mg có thành phần chính là Atorvastatin. Thuốc Insuact 10 có tác dụng làm giảm cholesterol và triglycerid trong máu ở bệnh nhân tăng cholesterol máu nguyên phát, rối loạn lipid máu hỗn hợp. 1. Thuốc Insuact 10 trị bệnh gì? Thuốc Insuact 10 thuộc nhóm thuốc điều trị rối loạn lipid máu, có thành phần chính là Atorvastatin 10mg. Atorvastatin có tác dụng làm giảm cholesterol, ức chế enzym tạo cholesterol ở gan. Atorvastatin làm giảm cholesterol chung bao gồm cholesterol LDL , triglycerid trong máu. Thuốc Insuact 10mg được bào chế dưới dạng viên nén bao phim, được chỉ định dùng trong những trường hợp sau: Ðiều trị hỗ trợ tăng cholesterol máu nguyên phát và rối loạn lipid máu hỗn hợp trên bệnh nhân đang áp dụng chế độ ăn kiêng để làm giảm cholesterol toàn phần , cholesterol LDL , apolipoprotein B, triglycerid và tăng cholesterol HDL . Insuact 10 cũng được dùng để điều trị rối loạn betalipoprotein trong máu nguyên phát. Ðiều trị hỗ trợ tăng cholesterol trong máu có tính gia đình đồng hợp tử trên bệnh nhân đang áp dụng các biện pháp làm giảm lipid khác để làm giảm cholesterol toàn phần và cholesterol LDL. 2. Liều dùng và cách dùng thuốc Insuact 10 Cách dùng thuốc Insuact 10: Thuốc được dùng theo đường uống, uống khi bụng đói hoặc no đều được, có thể uống vào bất cứ lúc nào trong ngày. Liều dùng thuốc Insuact 10mg khởi đầu là 10mg/lần/ngày, tối đa là 80mg/lần/ngày. Liều dùng thuốc Insuact 10 tùy vào mục đích điều trị cụ thể như sau: Tăng cholesterol máu nguyên phát và rối loạn lipid máu phối hợp: 10mg/lần/ngày, sau 2 - 4 tuần sẽ thấy hiệu quả của thuốc. Thuốc cần được được sử dụng duy trì trong thời gian dài để có hiệu quả. Tăng cholesterol trong máu có tính gia đình đồng hợp tử: Liều thường dùng là thuốc Insuact 10mg /lần/ngày và tối đa là 80mg/lần/ngày. Rối loạn lipid máu nghiêm trọng ở trẻ từ 10 - 17 tuổi: 10mg/lần/ngày, sau đó tăng lên 20mg/lần/ngày tùy vào cơ địa, tiến triển bệnh và khả năng dung nạp thuốc của người bệnh. Thời gian điều chỉnh liều thuốc tối thiểu là 4 tuần. 3. Tác dụng phụ của thuốc Insuact 10mg Thuốc Insuact 10 có thể gây một số tác dụng phụ không mong muốn với tần suất như sau: Thường gặp: Viêm mũi - họng, phản ứng dị ứng, tăng đường huyết, nhức đầu, đau thanh quản, chảy máu cam , đau cơ, co thắt cơ, đau khớp, sưng khớp, đau các chi, đau lưng, xét nghiệm gan bất thường, tăng creatine kinase trong máu, buồn nôn, khó tiêu, đầy hơi, táo bón, tiêu chảy. Ít gặp: Insuact 10 ít gây hạ đường huyết, tăng cân, chán ăn, mất ngủ, gặp ác mộng, choáng váng, dị cảm, mất trí nhớ, giảm cảm giác, loạn vị giác , nôn, đau bụng, ợ hơi, viêm tụy, viêm gan, nổi mày đay , phát ban, ngứa, rụng tóc, đau cổ, mỏi cơ, mệt mỏi, suy nhược, đau ngực, phù ngoại biên, sốt, xuất hiện bạch cầu trong nước tiểu, nhìn mờ, ù tai. Hiếm gặp: Insuact 10 hiếm khi làm giảm tiểu cầu, bệnh lý thần kinh ngoại biên, hoa mắt, ứ mật, phù thần kinh, nổi hồng ban, hội chứng hoại tử da nhiễm độc , hội chứng Stevens-Johnson , bệnh cơ, viêm cơ, tiêu cơ vân, bệnh gân, đôi khi nghiêm trọng hơn có thể đứt gân. Rất hiếm gặp: Insuact 10 rất hiếm khi gây sốc phản vệ , mất thính giác , suy gan , hội chứng to vú ở nam giới. Không rõ tần suất: Hoại tử cơ tự miễn trung gian. 4. Một số lưu ý khi dùng thuốc Insuact 10mg Không dùng thuốc Insuact với người bị quá mẫn với thành phần của thuốc, người có bệnh gan hoạt động hoặc tăng transaminase huyết thanh vô căn kéo dài, phụ nữ đang mang thai hoặc nuôi con cho bú, phụ nữ đang có ý định mang thai. Thuốc Insuact 10mg chỉ được dùng ở bệnh nhân có nguy cơ xơ vữa mạch máu cao do tăng cholesterol trong máu và phải kết hợp với chế độ ăn kiêng ít chất béo bão hòa , ít cholesterol và người bệnh đang áp dụng các biện pháp điều trị không dùng thuốc khác. Trước khi điều trị với Insuact 10 , người bệnh cần được loại trừ các nguyên nhân thứ phát gây tăng cholesterol bao gồm suy tuyến giáp , tiểu đường khó kiểm soát, hội chứng thận hư, nghiện rượu, bệnh gan tắc nghẽn, rối loạn protein trong máu, .... Ngoài ra, người bệnh cũng cần được kiểm tra, đo lường nồng độ lipid máu. Trước khi điều trị với thuốc Insuact 10mg , cần kiểm tra chức năng gan và trong quá trình dùng thuốc, người bệnh cần theo dõi, kiểm tra chức năng gan thường xuyên. Người có tiền sử mắc bệnh gan do rượu, bị nghiện rượu cần thận trọng khi dùng Insuact 10 . Trước khi dùng thuốc, người bệnh cần được cảnh báo nguy cơ gặp phải các vấn đề về cơ như căng cơ , đau cơ, yếu cơ. Thuốc Insuact 10mg có thể tương tác với các thuốc: Tăng nguy cơ tổn thương cơ (đặc biệt cơ vân) khi dùng đồng thời với thuốc điều trị viêm gan C và HIV, hạ cholesterol máu nhóm fibrat khác, thuốc ức chế miễn dịch; tăng thời gian chảy máu hoặc thời gian prothrombin khi dùng phối hợp với thuốc chống đông, indandione; tăng nhẹ nồng độ digoxin trong huyết thanh khi dùng phối hợp với Digoxin'''
question = '''Insuact 10mg được chỉ định dùng trong những trường hợp nào và liều dùng cụ thể ra sao?'''
conversation = [{"role": "system", "content": system_prompt }]
conversation.append({"role": "user", "content": template.format(context = context, question = question)})
text = tokenizer.apply_chat_template(
conversation,
tokenize=False,
add_generation_prompt=True)
model_inputs = tokenizer(text,return_tensors="pt").to(model.device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=2048,
temperature = 0.1,
#top_p=0.95,
#top_k=40,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
'''Trả lời:
Dựa trên ngữ cảnh cung cấp, Insuact 10mg được chỉ định dùng trong các trường hợp sau:
1. **Điều trị hỗ trợ tăng cholesterol máu nguyên phát và rối loạn lipid máu hỗn hợp**:
- **Liều dùng**: 10mg/lần/ngày.
- **Hiệu quả**: Sau 2 - 4 tuần sử dụng, bệnh nhân sẽ thấy hiệu quả của thuốc.
- **Lưu ý**: Thuốc cần được sử dụng duy trì trong thời gian dài để có hiệu quả.
2. **Điều trị hỗ trợ tăng cholesterol trong máu có tính gia đình đồng hợp tử**:
- **Liều dùng**: 10mg/lần/ngày (liều thường dùng) và tối đa 80mg/lần/ngày.
- **Lưu ý**: Liều lượng có thể được điều chỉnh tùy theo cơ địa, tiến triển bệnh và khả năng dung nạp thuốc của người bệnh.
3. **Rối loạn lipid máu nghiêm trọng ở trẻ từ 10 - 17 tuổi**:
- **Liều dùng**: 10mg/lần/ngày, sau đó có thể tăng lên 20mg/lần/ngày tùy theo tình trạng bệnh.
- **Thời gian điều chỉnh liều**: Thời gian điều chỉnh liều tối thiểu là 4 tuần.
**Lưu ý chung**:
- Thuốc được dùng theo đường uống, uống khi bụng đói hoặc no đều được, có thể uống vào bất cứ lúc nào trong ngày.
- Không dùng thuốc Insuact 10mg với người bị quá mẫn với thành phần của thuốc, người có bệnh gan hoạt động hoặc tăng transaminase huyết thanh vô căn kéo dài, phụ nữ đang mang thai hoặc nuôi con cho bú, phụ nữ đang có ý định mang thai.
- Cần kiểm tra chức năng gan trước và trong quá trình điều trị.
'''
```
#### 2. Khả năng chống nhiều
```python
# ví dụ
context = '''- Ngữ cảnh 1: . Và dù nguyên nhân thế nào, thì việc đánh người khác đều không thể chấp nhận, không ai có quyền xâm phạm thân thể của người khác, nhất là đánh những đứa trẻ là bạn của con cái mình. Lựa chọn kể với cha mẹ, người lớn về các mâu thuẫn học đường, là quyết định rất đúng của các em. Nhưng người lớn, đặc biệt những ông bố, bà mẹ cần ứng xử ra sao trước xung đột này của các con, thưa bà? Đứng ngoài mâu thuẫn bằng sự khách quan và trách nhiệm nhất có thể. Điều này giúp chúng ta đủ bình tĩnh để làm việc với tất cả các bên liên quan, từ giáo viên, bạn của con, ban giám hiệu để tìm hiểu câu chuyện và tìm kiếm cách giải quyết tích cực, trên cơ sở phối hợp nhà trường. Người lớn không thể chỉ nghe một tai và đặc biệt không nên tự xử. Phụ huynh, kể cả học sinh tự xử các vấn đề học đường là điều rất nguy hiểm và cho thấy sự coi thường pháp luật . Vụ việc ở Tuyên Quang vừa rồi là ví dụ. Các em hoàn toàn có thể phản ứng bằng cách trình bày, gửi yêu cầu, kiến nghị lên nhà trường, nhờ phụ huynh làm việc với ban giám hiệu để có biện pháp giải quyết nếu cô giáo sai, không nên đồng loạt dồn cô giáo vào tường một cách bạo lực và trái đạo đức, tôn ti trật tự như vậy. Ngoài ra, chúng ta cũng có rất nhiều cơ quan chức năng bảo vệ phụ huynh và con em, với những quyền về khiếu nại, tố cáo. Chúng ta nói nhiều về trường học an toàn. Trong những câu chuyện học sinh bị hành hung thế này, có lẽ cũng cần làm rõ vai trò, trách nhiệm của nhà trường? TPHCM và nhiều địa phương đang xây dựng môi trường trường học hạnh phúc, tiêu chí là yêu thương, an toàn, tôn trọng. Không chỉ phòng ngừa, nhà trường còn phải tích cực vào cuộc xử lý các mâu thuẫn học đường, hạn chế tối đa nguy cơ mất an toàn cho học sinh, giáo viên. Đặc biệt, giải quyết câu chuyện bạo lực học đường phải triệt để, tuyệt đối không nửa vời vì nửa vời sẽ tiềm ẩn nguy cơ rất lớn dẫn đến những vụ việc tương tự, với mức độ nghiêm trọng hơn. Vụ việc em M. ở Nha Trang tự vẫn với lá thư tuyệt mệnh bị đổ oan đầu tháng 10 vừa qua là một ví dụ về giải quyết không triệt để. Việc xây dựng trường học hạnh phúc nếu triển khai “đến nơi đến chốn”, sẽ góp phần rất lớn cải thiện tình trạng bạo lực học đường, tạo môi trường sống và học tập bình an cho các con. Từ nhiều sự vụ học sinh bạo hành lẫn nhau, giáo viên bạo hành học sinh, phụ huynh hành hung giáo viên và bạn của con. Tam giác phối hợp bảo vệ học sinh là nhà trường - gia đình - xã hội phải chăng đang có một lỗ hổng lớn, thưa bà? Câu chuyện này có liên quan đến niềm tin của phụ huynh với nhà trường. Tại sao phụ huynh lại chọn cách tự xử? Chúng ta cũng cần phải xem lại cách giải quyết vấn đề của nhà trường đã rốt ráo chưa, coi trọng lợi ích của tất cả các bên liên quan chưa hay chỉ đang xoa dịu? Người ta chỉ tìm đến nhau khi có niềm tin vào nhau. Thực trạng phụ huynh chọn cách chuyển trường cho con cũng nói lên điều này. Đây là cách chạy trốn của phụ huynh với mong muốn con được an toàn, hạnh phúc hơn ở môi trường mới. Xây dựng niềm tin cho phụ huynh, xã hội cần được chú trọng và với mỗi một trường hợp phụ huynh yêu cầu chuyển trường cho con - đang rất phổ biến - nhà trường cần xét kỹ các nguyên nhân và hóa giải. Xin bà cho biết đâu là giải pháp căn cơ cho tất cả những câu chuyện bạo lực nói trên? Để trẻ không là nạn nhân của bạo lực học đường, phụ huynh cần đồng hành và giúp con có sự hiểu biết, ý thức trước vấn đề này. Dạy con kỹ năng giao tiếp, quản lý cảm xúc rất quan trọng và điều này không thể chỉ dựa vào những khóa học kỹ năng sống, mà là từ cách cư xử của người lớn, cha mẹ, thầy cô. Không có tấm gương nào tốt hơn cho con trẻ bằng ứng xử, hành vi của người lớn. Vì vậy, không thể đòi hỏi trẻ nói không với bạo lực học đường khi trong chính từng gia đình, xã hội, người lớn vẫn đối xử với nhau bằng bạo lực.
- Ngữ cảnh 2: Tại sao triều Thanh có rất nhiều thân vương nhưng chẳng có ai dám tạo phản? Không giống như những triều đại trước đó, triều Thanh dù có sự tranh giành ngai vàng khốc liệt giữa các hoàng tử nhưng lại chẳng bao giờ xảy ra thế cục các thân vương tạo phản. Chính vì 3 lý do lớn này đã khiến cho triều đại nhà Thanh khác hẳn triều đại nhà Đường và nhà Minh. Trong thời cổ đại, các vương công quý tộc để tranh giành vương vị của mình, giữa huynh đệ ruột thịt với nhau dường như đều xảy ra đấu đá, hãm hại lẫn nhau, coi nhau như kẻ thù không đội trời chung, có ta thì không có ngươi, có ngươi thì sẽ chẳng có ta, điều này hoàn toàn không phải là điều gì xa lạ. Vậy thì tại sao ngai vàng lại có sức hút lớn đến thế? Không chỉ là đàn ông khát khao quyền lực, mà quan trọng hơn là hoàng đế có thể có được hậu cung rộng lớn, trong hậu cung còn có vô số các mỹ nữ quốc sắc thiên hương. Nhiều phi tần như vậy, đương nhiên hoàng đế cũng sẽ có rất nhiều con cái, không tính đến con gái, chỉ riêng những vị hoàng tử, để có thể có được hoàng vị, họ tranh giành nhau bằng cả sinh mạng. Vậy thì ai là người được lựa chọn để thừa kế ngai vàng, ai mới có thể gánh được trọng trách trị vì đất nước? Đa phần đều theo tục lệ truyền cho con trai đích tôn (con trai do hoàng hậu sinh ra) hoặc con trai trưởng (con trai đầu tiên của hoàng đế). Cho dù tục lệ này có lịch sử lâu đời nhưng nó cũng có những khuyết điểm rất lớn, đó chính là nếu như năng lực và chí hướng của con trai đích tôn hoặc con trai trưởng không thể gánh vác được ngai vị, nếu để anh ta lên ngôi hoàng đế, vậy thì đất nước sẽ rơi vào cục diện suy vong. Còn có một khuyết điểm nữa đó chính là những người con trai có dã tâm lớn khác sẽ không phục việc con trai đích hoặc con trai trưởng kế thừa ngôi báu, họ sẽ khởi binh tạo phản cũng là chuyện rất dễ xảy ra. Ví dụ như trong thời Đường của Trung Quốc, Đường Cao Tổ Lý Uyên đem binh tiêu diệt nhà Tùy thối nát, đồng thời lập nên nhà Đường, vốn dĩ ông cũng dựa theo tục lệ lập con trai trưởng là Lý Kiến Thành làm Thái tử nhưng con trai thứ là Lý Thế Dân lại không phục với sự sắp xếp này. Vì năng lực của ông xuất chúng, văn võ song toàn, còn lập được không ít công lao to lớn trong cuộc chiến tranh tiêu diệt nhà Tùy cùng cha mình, đương nhiên không chịu thấp hơn kẻ khác một bậc. Thế nên đã phát động binh biến Huyền Vũ Môn, trong cuộc binh biến tạo phản này, đích thân ông đã giết chết huynh trưởng của mình, đồng thời ép cha mình là Lý Uyên phải truyền ngôi cho mình. Hay như trong thời nhà Minh của Trung Quốc, trước khi Chu Nguyên Chương chọn người lập làm Thái tử, con trai trưởng Chu Tiêu đã qua đời vì bệnh nặng, thế nên Chu Nguyên Chương đã lập cháu đích tôn của mình làm Thái tử kế thừa vương vị, nhưng em trai của Chu Tiêu là Chu Đệ lại không phục lựa chọn này của Chu Nguyên Chương. Theo lý mà nói thì sau khi anh trai Chu Tiêu qua đời, ông đã có tư cách thừa kế ngai vàng nhưng Chu Nguyên Chương nhất quyết không chọn ông mà lại chọn người cách thế hệ để truyền ngôi. Điều này khiến Chu Đệ với thế lực to lớn không thể nuốt nổi cục tức này, vì thế Chu Tiêu vừa qua đời thì ông đã vội vã khởi binh tạo phản, giết chết cháu trai ruột của mình rồi tự xưng vương. Vậy thì tại sao trong triều Thanh có rất nhiều thân vương như vậy mà lại chẳng có ai đứng ra tạo phản? Đầu tiên phải nói về bối cảnh xã hội trong thời kỳ này. Triều Thanh từ khi thành lập, cũng giống với những triều đại khác, đều có rất nhiều thân vương. Nếu người dân bình thường muốn làm hoàng đế, vậy thì đó là điều hoàn toàn không thể, nhưng đối với những vương công quý tộc trong hoàng thất mà nói, họ đương nhiên sẽ có rất nhiều cơ hội, đặc biệt là những thân vương nắm đại quyền quân sự , họ chính là mối đe dọa lớn nhất đối với nhà vua. Vì thế, các đời hoàng đế đều sẽ nghĩ đủ mọi cách để áp chế, kiểm soát họ, tránh việc họ khởi binh tạo phản. Triều Thanh có lịch sử hơn 300 năm, cũng đã cho ra đời vô số thân vương, đặc biệt là cuối thời Thanh, khi Trung Quốc rơi vào cảnh khốn khó, sau khi Từ Hy Thái Hậu cầm quyền thì thế cục này càng được thể hiện rõ rệt hơn. Nhưng cho dù là một người phụ nữ cầm quyền thì cũng chẳng có một vị thân vương hoàng tộc nào đứng ra tạo phản. Có 3 nguyên nhân sau: Thứ nhất, thân vương triều Thanh không thể nối ngôi, nếu muốn tiếp tục duy trì danh phận thân vương, vậy thì bắt buộc phải có được sự đồng ý của hoàng đế và phải lập được công lao cho đất nước. Thứ hai, triều đình tiến hành giám sát nghiêm ngặt đối với các thân vương, họ không hề có cơ hội để tạo phản. Thứ ba, các thân vương không thể giao thiệp quá sâu với các đại thần, quan lại khác, điều này cũng khiến các thân vương rơi vào cảnh bị cô lập, thế nên càng không có cơ hội để cấu kết với người khác hòng tạo phản. - Video: Ngắm sự kỳ vĩ và lộng lấy của Tử Cấm Thành từ trên cao. Nguồn: Sky Eye.
- Ngữ cảnh 3: . Cùng điều chỉnh với con là điều rất quan trọng bởi vì trẻ sẽ tự tin để tự đặt những giới hạn cho chính mình khi lớn lên”, TS Nguyễn Thị Thanh đưa ra lời khuyên. “Khi con mắc sai lầm, hãy giúp chúng tìm những cách khác tốt hơn. Đơn cử dùng hậu quả để dạy cho chúng bài học, điều đó tốt hơn rất nhiều việc xử phạt. Nếu cha mẹ chỉ biết trừng phạt, sẽ nhận được lời xin lỗi nhưng không thể giúp trẻ tỉnh ngộ. Bởi chúng chỉ biết được mình đã sai mà không biết sai ở chỗ nào và làm thế nào mới là đúng”
- Ngữ cảnh 4: . “MẤT ĐI CHA MẸ Ở TUỔI ĐẸP NHẤT CỦA NGƯỜI PHỤ NỮ CÀNG KHIẾN TÔI PHẢI MẠNH MẼ” - Làm con của nghệ sĩ Thanh Hiền, Đinh Y Nhung cảm nhận sợi dây liên kết giữa hai mẹ con thế nào? Má Thanh Hiền là người rất tuyệt vời. Hai má con hồi xưa từng làm phim truyền hình với nhau rồi, cho nên khi tái hợp thì không mấy bỡ ngỡ. Khi đối diễn, hai má con rất ăn ý, như người thân ruột thịt vậy đó. - Khi thể hiện những phân cảnh cảm động trong phim, có khi nào chị thấy nhớ mẹ không? Có chứ, nhất là ở những phân đoạn gia đình sum họp, tự nhiên mình bị buồn. Ai cũng muốn có cha, có mẹ, ai cũng muốn Tết được chạy về bên gia đình. Trong mười mấy, hai chục năm qua, Nhung bị chạnh lòng. Tuy nhiên, chỉ trong tích tắc, tôi tự trấn an rằng, mình đang quay phim, đang hóa thân vào nhân vật nên không thể xao lãng được. Mình là con người mà, cũng có lúc tâm trạng vui buồn bất chợt, nhưng Nhung luôn cố gắng lấy lại phong độ liền. - Mất ba mẹ từ sớm, không có chỗ dựa tinh thần, cô gái trẻ Đinh Y Nhung năm đó có nhận những lời mời gọi khiếm nhã không? Trước đây, Nhung không có bạn bè nhiều, chủ yếu chỉ lo đi học, đi làm để lo cho cuộc sống thôi. Nên Nhung không phải đón nhận những lời mời gọi nào hết. - Mất mát từ quá khứ có ảnh hưởng gì đến suy nghĩ về tương lai của chị sau này, ví dụ khi có con thì sẽ bù đắp, chăm sóc cho con nhiều hơn? Năm ba mẹ mất thì mình vẫn còn khá trẻ, thật ra cái tuổi đó là tuổi đẹp của người phụ nữ. Sau đó, tôi đi làm, rồi yêu đương và lập gia đình. Có rất nhiều thứ hối tiếc để nói về Nhung của thời điểm đó. Thứ nhất là mình chưa thành công, thứ hai là mình chưa trả hiếu cho cha mẹ, thứ ba là mình còn bấp bênh. Nhung lúc đó lì lợm lắm, không cho phép mình ngã, bằng mọi giá phải tiến về trước dù có hàng ngàn chông gai ngăn cản. Có lúc tôi bị người này cười, người kia mỉa, nhưng mà mình vẫn cố bước đi. Người ta có cười thì cũng không mang lại cho mình được gì, tôi chỉ biết làm hết khả năng để lo cho bản thân, lo cho em của mình. Hiện, con cái Nhung đã đi nước ngoài rồi. Bé đang học đại học về âm nhạc, còn em mình cũng đã lớn rồi. Đối với Nhung ngay lúc này thì không phải thành công hay hoàn hảo lắm, nhưng ít nhất là tôi đã cố gắng để tự chịu trách nhiệm với cuộc đời mình. Mất cha, mất mẹ, đối với một người hai mươi mấy tuổi thì điều cần nhất lúc đó là có được gia đình ở bên. Nhưng mình không có chỗ dựa tinh thần thì càng phải mạnh mẽ hơn nữa. Tôi tự gặm nhấm nỗi đau mất người thân trong một thời gian dài, có khi đến cả bạn bè cũng không hề biết. Một thời gian sau, bạn bè thời và mọi người mới biết. Còn người hâm mộ, đồng nghiệp trong nghề gần như không biết chuyện ba mẹ Nhung mất sớm, chỉ có vài người chơi thân với nhau biết thôi. Sau này, dần dần tâm lý dần ổn định thì mình mới bắt đầu chia sẻ. “CON ĐI DU HỌC, TÔI DẶN BÉ CÁI GÌ KHÔNG TỐT THÌ MÌNH BỎ QUA” - Đinh Y Nhung từng tiết lộ mình rất thân với con gái. Có vẻ như quyết định để con đi du học là không hề dễ dàng? Thật sự là không có ba mẹ nào muốn con mình đi xa, nhưng việc du học lại là quyết định của bé. Con Nhung bày tỏ muốn học đại học ở nước ngoài và muốn đi sớm để thực hiện ước mơ. Nhưng lúc đó con còn nhỏ quá, phải đợi đến năm con 17 tuổi thì Nhung mới quyết định cho bạn nhỏ đi. Con cái từ nhỏ ở với bố mẹ giờ lại đi xa thì tất nhiên người làm cha làm mẹ cùng phải thấy sốc, thấy buồn. Nhưng Nhung hoàn toàn tôn trọng quyết định của con về việc chọn ngành nghề và tương lai của mình. Ba mẹ sẽ đứng sau và là người đưa cho con những lời khuyên và chỉ có thể đồng hành cùng con tới một mốc thời gian nào đó. Về sau, con phải đi làm và tự có trách nhiệm với cuộc đời của mình. - Có con gái đang ở tuổi lớn lại xa bố mẹ và tiếp xúc một nền văn hóa phương Tây cởi mở, Đinh Y Nhung đã khuyên dạy và đồng hành với con như thế nào? Ngay khi ở Việt Nam, con gái Nhung đã được theo học trường quốc tế. Hai mẹ con cũng có rất nhiều dịp để tâm sự và chia sẻ với nhau. Ngay từ nhỏ, Nhung đã cho bé được tiếp xúc song song giữa hai nền văn hóa để con không bỡ ngỡ. Mình là người Việt nên đương nhiên vẫn dạy con theo văn hóa Á Đông là chủ yếu. Nhung vẫn luôn tạo điều kiện để con cảm nhận những nét đẹp trong nền văn hóa quê hương. Văn hóa phương Tây thì xa lạ hơn nhưng Nhung cũng khuyên con rằng điều gì hay thì mình học hỏi, cái gì không tốt thì mình nên bỏ qua. Tất nhiên mình không thể theo sát con, nhất là khi bé đang ở độ tuổi mới lớn, có nhiều sự hiếu kỳ. Tuy nhiên, Nhung cũng không quá lo lắng vì qua quá trình học tập ở các trường quốc tế, bé cùng đã được làm quen dần với văn hóa phương Tây. Bé muốn làm bạn với mẹ nên có nhiều thứ bé muốn hỏi, muốn tiếp thu thì hai mẹ con lại ngồi xuống chia sẻ, tâm sự với nhau. Nhung tin, con luôn tỉnh táo để đưa ra những quyết định cho bản thân mình. Nhung không dám nói trước, nhưng hiện tại con vẫn luôn biết nói cảm ơn, xin phép trước khi làm bất cứ điều gì nên mình vẫn rất tin tưởng con. - Chị nhận xét thế nào về tính cách của con gái? Phải chăng bé là phiên bản nhí của chị? Con gái Nhung có nhiều nét giống mẹ.
- Ngữ cảnh 5: . - Trong quá trình quay phim, má Hiền có tham gia đóng góp ý kiến cho dàn diễn viên trẻ không? Không phải má Hiền muốn làm thầy nhưng có những lúc cũng lên tiếng góp ý một vài điều. Những lúc như thế các bạn diễn viên trẻ cũng nghe theo. Phải giúp đỡ nhau như vậy mới có thể tạo ra được một tác phẩm hay. Chứ có nhiều diễn viên trẻ có lòng tự trọng cao, chỉ khi đạo diễn nói mới nghe thì không bao giờ có thể tiến bộ được. Má Hiền phải cảm thấy hạnh phúc mới đúng, càng ngày xung quanh càng có nhiều bạn bè, các mối quan hệ nhờ đó mà cũng ngày càng đông. Ngày xưa, khi tham gia là Mùi Ngò Gai chưa có ai biết nhưng sau khi phim phát sóng mọi người đã nhớ đến má. Hiện tại, má Hiền đã được đạo diễn, nhà sản xuất mời đóng phim đó đã là một niềm hạnh phúc. Đi ra đường hay đi khám bệnh, người ta cứ vây xung quanh quan tâm hỏi han khiến má cảm thấy rất vui lắm (cười). Cảm ơn nghệ sĩ Thanh Hiền đã nhận lời tham gia phỏng vấn với SAOstar. Chúc má có nhiều sức khỏe để tiếp tục làm nghề!'''
question = '''Tại sao các thân vương triều Thanh không dám tạo phản mặc dù có nhiều quyền lực và cơ hội?'''
'''Trả lời:
Các thân vương triều Thanh không dám tạo phản mặc dù có nhiều quyền lực và cơ hội vì có ba lý do chính:
1. **Không thể nối ngôi**: Để duy trì danh phận thân vương, họ phải có sự đồng ý của hoàng đế và phải lập được công lao cho đất nước. Điều này hạn chế cơ hội của họ để trở thành hoàng đế.
2. **Giám sát nghiêm ngặt**: Triều đình tiến hành giám sát nghiêm ngặt đối với các thân vương, không cho họ cơ hội để tạo phản.
3. **Bị cô lập**: Các thân vương không thể giao thiệp quá sâu với các đại thần, quan lại khác, điều này khiến họ bị cô lập và không có cơ hội cấu kết với người khác để tạo phản.'''
```
#### 3. Loại bỏ negative
```python
context = '''- Ngữ cảnh 1: Thái Bình hướng đến là trung tâm công nghiệp, năng lượng của vùng Với tiềm năng sẵn có, quy hoạch tỉnh Thái Bình thời kỳ 2021-2030, tầm nhìn đến năm 2050 xác định tỉnh sẽ phát triển công nghiệp theo hướng hiện đại, bền vững dựa trên nghiên cứu phát triển điện gió, điện khí, cân bằng lượng phát thải. Sáng 5/3, UBND tỉnh Thái Bình tổ chức Hội nghị công bố quy hoạch của tỉnh thời kỳ 2021-2030, tầm nhìn đến năm 2050 và xúc tiến đầu tư tỉnh Thái Bình. Phát biểu tại hội nghị, Phó Chủ tịch Thường trực UBND tỉnh Nguyễn Quang Hưng cho biết: Mục tiêu của quy hoạch là đến năm 2030, Thái Bình trở thành địa phương thuộc nhóm phát triển khá và là một trong những trung tâm phát triển công nghiệp của vùng Đồng bằng sông Hồng, có cơ cấu kinh tế hiện đại với công nghiệp là động lực chủ yếu cho tăng trưởng để Thái Bình phát triển nhanh, toàn diện và bền vững. Đến năm 2050, Thái Bình là tỉnh phát triển của vùng Đồng bằng sông Hồng, tăng trưởng kinh tế dựa trên nền tảng khoa học công nghệ, đổi mới sáng tạo và các ngành kinh tế trụ cột có sức cạnh tranh cao. Quy hoạch tỉnh đã xác định 4 trụ cột tăng trưởng, 3 khâu đột phá, 4 không gian kinh tế - xã hội, 3 hành lang kinh tế, định hướng phát triển các ngành và lĩnh vực và 6 nhiệm vụ trọng tâm. Quy hoạch tỉnh cũng có nhiều điểm mới, đột phá như mở ra không gian phát triển mới thông qua hoạt động “lấn biển”, tạo quỹ đất cho các hoạt động chức năng, hình thành không gian công nghiệp - đô thị - dịch vụ. Về hạ tầng giao thông, Thái Bình sẽ đầu tư 3 tuyến cao tốc là cao tốc Ninh Bình - Hải Phòng (CT.08), đường vành đai 5 - Hà Nội (CT.39) và tuyến CT.16 kết nối Khu kinh tế với thành phố Thái Bình và vùng kinh tế phía Tây Bắc Thủ đô. Tỉnh cũng sẽ đầu tư 101km đường sắt, khổ đường dự kiến rộng 1.435 mm và sân bay chuyên dụng nằm ở ven biển Thái Bình. Về phát triển kinh tế, quy hoạch tỉnh Thái Bình xác định tỉnh sẽ phát triển công nghiệp theo hướng hiện đại, công nghệ tiên tiến, giá trị gia tăng cao, tham gia sâu, toàn diện vào mạng lưới sản xuất, chuỗi giá trị toàn cầu, phát huy các tiềm năng, thế mạnh để đưa Thái Bình trở thành một trong những trung tâm phát triển công nghiệp, năng lượng của vùng Đồng bằng sông Hồng. Tỉnh khuyến khích đầu tư phát triển các ngành có thế mạnh và có thể tạo đột phá như năng lượng, cơ khí chế biến, chế tạo, công nghiệp công nghệ cao, điện - điện tử, chế biến sản phẩm nông, lâm nghiệp và thủy sản… Đồng thời, tập trung nghiên cứu phát triển điện gió, điện khí để tạo nguồn điện sạch và cân bằng lượng phát thải, nghiên cứu đầu tư xây dựng nhà máy chế biến Condensate, chuẩn bị mọi điều kiện để xây dựng và đưa vào vận hành Nhà máy nhiệt điện LNG Thái Bình. Về nông nghiệp, tỉnh Thái Bình vẫn xác định đây là \"trụ cột quan trọng\" trong phát triển kinh tế của tỉnh, góp phần bảo đảm an ninh lương thực quốc gia, hướng tới trở thành trung tâm sản xuất nông nghiệp hàng đầu của Đồng bằng sông Hồng. Phát biểu tại hội nghị, Phó Thủ tướng Chính phủ Trần Lưu Quang đánh giá Thái Bình có 4 tiềm năng, lợi thế lớn để có thể có sự bứt phá trong thời gian tới như vị trí địa lý và tiếp cận đất đai thuận lợi; từng là địa phương đi đầu trong xây dựng nông thôn mới bài bản và nghiêm túc, nhận được sự quan tâm của nhiều thế hệ lãnh đạo Đảng, Nhà nước và có nhiều doanh nhân người Thái Bình và luôn hướng về quê hương; có sự đoàn kết, thống nhất, trước hết là trong tập thể lãnh đạo. Về vị trí địa lý và tiếp cận đất đai, Phó Thủ tướng cho rằng trong tương lai, khi Luật Đất đai có hiệu lực, Thái Bình sẽ có nhiều điều kiện lấn biển để triển khai các dự án khu đô thị, khu công nghiệp thân thiện với môi trường. Đối với nông nghiệp, Phó Thủ tướng nhấn mạnh về lâu dài Thái Bình có thể ghi điểm từ phát triển công nghiệp nhưng trước mắt, đặc biệt trong lúc khó khăn thì nông nghiệp vẫn là nền tảng rất quý giá. Mặt khác, ứng dụng của công nghệ cao trong sản xuất nông nghiệp sẽ rút ngắn thời gian làm đồng của người nông dân, tạo điều kiện để Thái Bình huy động nguồn nhân lực trong nông nghiệp sang phát triển các ngành công nghiệp và dịch vụ, một lợi thế mà không phải địa phương nào cũng có được như Thái Bình. Bên cạnh những lợi thế trên, lãnh đạo Chính phủ chỉ ra một số khó khăn mà tỉnh phải đối mặt như Thái Bình đã sử dụng hết 1.600 ha chỉ tiêu đất công nghiệp trong giai đoạn này, đòi hỏi phải có phương án giải quyết thấu đáo trong thời gian tới để tỉnh tiếp tục phát triển công nghiệp. Đồng thời, Thái Bình cũng phải cạnh tranh với những địa phương như Hải Phòng, Quảng Ninh trong thu hút FDI trong khi phát triển cơ sở hạ tầng chưa theo kịp mong muốn. Do vậy, khi triển khai quy hoạch tỉnh, Phó Thủ tướng nhắn nhủ tới địa phương 8 chữ: Tuân thủ, linh hoạt, đồng bộ và thấu hiểu. Đồng thời, tỉnh cũng phải \"linh hoạt\" trong tổ chức thực hiện, trong trường hợp cá biệt cụ thể, điều chỉnh mục tiêu cho phù hợp. Sáng cùng ngày, Phó Thủ tướng Trần Lưu Quang đã dự Lễ khởi công dự án Nhà máy Pegavision Việt Nam tại khu công nghiệp Liên Hà Thái, huyện Thái Thụy, tỉnh Thái Bình
- Ngữ cảnh 2: Bình Định được định hướng là trung tâm khoa học, công nghệ đổi mới sáng tạo Tỉnh Bình Định được định hướng phát triển ngành công nghiệp phát triển theo hướng hiện đại, quy mô lớn, trở thành một trong những trung tâm công nghiệp chế biến chế tạo và công nghệ cao của vùng Bắc Trung Bộ và duyên hải Trung Bộ. Theo Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 vừa được Thủ tướng Chính phủ phê duyệt, tỉnh Bình Định được định hướng phát triển ngành công nghiệp phát triển theo hướng hiện đại, quy mô lớn, trở thành một trong những trung tâm công nghiệp chế biến chế tạo và công nghệ cao của vùng Bắc Trung Bộ và duyên hải Trung Bộ. Ngành công nghiệp tăng trưởng nhanh, bền vững, hướng tới tăng trưởng xanh, kinh tế tuần hoàn là trụ cột để phát triển và chuyển dịch cơ cấu kinh tế của tỉnh. Ngành chế biến, chế tạo công nghệ cao (dịch chuyển ngành công nghiệp chế biến, chế tạo sang lĩnh vực sản xuất có giá trị gia tăng cao như: chế biến sâu nông - thủy - hải sản, linh kiện điện tử, bán dẫn, dược phẩm), công nghệ thông tin, trí tuệ nhân tạo trở thành một trong những lĩnh vực đột phá, góp phần đưa tỉnh Bình Định trở thành một trung tâm khoa học, công nghệ đổi mới sáng tạo của vùng và cả nước. Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 đặt ra yêu cầu tỉnh này phải chú trọng thu hút đầu tư phát triển năng lượng tái tạo, năng lượng sạch như điện gió ven bờ, điện gió ngoài khơi, điện mặt trời, điện sinh khối và nguồn năng lượng mới (hydrogen/amoniac xanh…); các dự án sản xuất thép quy mô lớn, đóng tàu, sản xuất thiết bị phụ trợ điện gió có công nghệ tiên tiến để nâng cấp xây dựng hạ tầng kỹ thuật sản xuất, thúc đẩy chuyển dịch kinh tế. Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 cũng đặt ra mục tiêu đến năm 2030, Bình Định trở thành tỉnh phát triển thuộc nhóm dẫn đầu vùng Bắc Trung Bộ và duyên hải Trung Bộ, là trung tâm công nghiệp chế biến, chế tạo, dịch vụ, du lịch và văn hóa phía Nam của vùng; trung tâm lớn của cả nước về phát triển kinh tế biển; trọng điểm du lịch quốc gia và quốc tế với hệ thống kết cấu hạ tầng kinh tế đồng bộ, hiện đại; kinh tế của tỉnh phát triển nhanh, bền vững và xanh dựa trên các trụ cột tăng trưởng công nghiệp, dịch vụ du lịch, cảng biển - logistics; nông nghiệp ứng dụng công nghệ cao; đô thị hóa; thực hiện thành công các mục tiêu chuyển đổi số, đổi mới sáng tạo, cải thiện mạnh mẽ môi trường đầu tư kinh doanh, trở thành điểm đến đầu tư hấp dẫn của các doanh nghiệp lớn trong và ngoài nước; chỉ số năng lực cạnh tranh cấp tỉnh thuộc nhóm cao của cả nước; kết cấu hạ tầng kinh tế - xã hội đồng bộ, hiện đại, hệ thống đô thị phát triển theo hướng đô thị thông minh, kết nối thuận tiện với các trung tâm kinh tế của vùng, cả nước và quốc tế.
- Ngữ cảnh 3: . Chủ tịch UBND tỉnh Quảng Ninh cho biết, tỉnh đặt mục tiêu hướng đến năm 2030 trở thành một tỉnh tiêu biểu của cả nước về mọi mặt; tỉnh kiểu mẫu giàu đẹp, văn minh, hiện đại, nâng cao đời sống mọi mặt của nhân dân; cực tăng trưởng của khu vực phía Bắc, một trong những trung tâm phát triển năng động, toàn diện; trung tâm du lịch quốc tế, trung tâm kinh tế biển, cửa ngõ của Vùng kinh tế trọng điểm Bắc Bộ và cả nước. Để đạt được những mục tiêu trên, tỉnh Quảng Ninh xác định sự đóng góp, quan tâm của cộng đồng doanh nghiệp, nhất là các doanh nghiệp hàng đầu Việt Nam “các sếu đầu đàn” là một trong những yếu tố then chốt quyết định. Do vậy, tỉnh Quảng Ninh rất mong nhận được sự quan tâm, nghiên cứu đầu tư hợp tác của các Doanh nghiệp hàng đầu Việt Nam trong thời gian tới, nhất là trong việc đầu tư các dự án có hàm lượng công nghệ cao, công nghệ tiên tiến, quản trị hiện đại, giá trị gia tăng cao, có tác động lan tỏa. Tỉnh Quảng Ninh cam kết tạo điều kiện thuận lợi nhất cho doanh nghiệp phát triển hơn nữa khi đầu tư kinh doanh trên địa bàn tỉnh; cam kết đồng hành, lắng nghe tiếng nói của cộng đồng doanh nghiệp, các nhà đầu tư; cùng trăn trở, trách nhiệm, giải quyết thấu đáo, vào cuộc thực chất, hiệu quả đối với từng khó khăn, vướng mắc với mục tiêu tăng cường niềm tin và nâng cao sự hài lòng của cá nhân, tổ chức, doanh nghiệp là thước đo đánh giá chất lượng phục vụ, chất lượng điều hành của cơ quan hành chính nhà nước từ cấp tỉnh đến các cấp cơ sở. Tỉnh Quảng Ninh sẽ triển khai một cách hiệu quả nhất Quy hoạch tỉnh, quy hoạch các địa phương; chuẩn bị mặt bằng sạch; quỹ đất tái định cư; đảm bảo nguồn cung điện, nước và viễn thông theo yêu cầu; xây dựng các chính sách đột phá để thu hút nguồn nhân lực chất lượng cao thông qua việc cải thiện nhà ở và cơ sở văn hóa để đảm bảo đời sống vật chất và tinh thần cho người lao động. Đồng thời bảo vệ quyền và lợi ích chính đáng, hợp pháp của nhà đầu tư, doanh nghiệp vào tỉnh; bảo đảm hài hòa lợi ích giữa nhà nước, nhà đầu tư, người lao động và người dân.'''
question = '''Theo quy hoạch tỉnh Tây Ninh thời kỳ 2021 - 2030, tầm nhìn đến năm 2050, tỉnh sẽ tập trung phát triển công nghiệp và dịch vụ du lịch. Trong bối cảnh cạnh tranh với các tỉnh lân cận, yếu tố nào được coi là quan trọng nhất để Tây Ninh có thể thu hút đầu tư và phát triển nhanh, toàn diện và bền vững?'''
'''Trả lời:
Trong ngữ cảnh được cung cấp, không có thông tin về quy hoạch tỉnh Tây Ninh thời kỳ 2021 - 2030, tầm nhìn đến năm 2050. Do đó, không có câu trả lời chính xác dựa trên các thông tin đã cung cấp. Nếu bạn có thêm thông tin về quy hoạch tỉnh Tây Ninh, vui lòng cung cấp để tôi có thể trả lời câu hỏi một cách chính xác và đầy đủ.
'''
```
#### 4. Tích hợp thông tin
```python
context = '''- Ngữ cảnh 1: 43/2020/nđ-cp Điều 3. Chế độ, chính sách đối với người tham gia thi hành án tử hình\n1. Người tham gia Đội thi hành án tử hình được hưởng chế độ bồi dưỡng bằng ba lần mức lương cơ sở khi thi hành án tử hình đối với 01 người và nghỉ dưỡng 10 ngày theo quy định chung về chế độ nghỉ dưỡng đối với cán bộ, chiến sĩ Công an nhân dân, Quân đội nhân dân.\n2. Người tham gia Hội đồng thi hành án tử hình, cán bộ quản giáo, người ghi âm, ghi hình, chụp ảnh, phiên dịch, thực hiện lăn tay người bị thi hành án tử hình, khâm liệm, mai táng tử thi được hưởng chế độ bồi dưỡng bằng một lần mức lương cơ sở khi thi hành án tử hình đối với 01 người.\n3. Người tham gia bảo đảm an ninh, trật tự; đại diện Ủy ban nhân dân cấp xã; Điều tra viên được hưởng chế độ bồi dưỡng bằng một phần hai mức lương cơ sở khi thi hành án tử hình đối với 01 người.
- Ngữ cảnh 2: 53/2010/qh12 Điều 60. Giải quyết việc xin nhận tử thi, hài cốt của người bị thi hành án tử hình\n1. Việc giải quyết nhận tử thi được thực hiện như sau:\na) Trước khi thi hành án tử hình, thân nhân hoặc người đại diện hợp pháp của người chấp hành án được làm đơn có xác nhận của Ủy ban nhân dân cấp xã nơi cư trú gửi Chánh án Tòa án đã xét xử sơ thẩm đề nghị giải quyết cho nhận tử thi của người chấp hành án để an táng; trường hợp người chấp hành án là người nước ngoài thì đơn phải có xác nhận của cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người chấp hành án mang quốc tịch và phải được dịch ra tiếng Việt. Đơn phải ghi rõ họ tên, địa chỉ người nhận tử thi, quan hệ với người chấp hành án; cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí;\nb) Chánh án Tòa án đã xét xử sơ thẩm thông báo bằng văn bản cho người có đơn đề nghị về việc cho nhận tử thi hoặc không cho nhận tử thi khi có căn cứ cho rằng việc nhận tử thi ảnh hưởng đến an ninh, trật tự, vệ sinh môi trường. Trường hợp người chấp hành án là người nước ngoài, thì Chánh án Tòa án đã xét xử sơ thẩm có trách nhiệm thông báo bằng văn bản cho Bộ Ngoại giao Việt Nam để thông báo cho cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người đó mang quốc tịch;\nc) Cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm thông báo cho người có đơn đề nghị ngay sau khi thi hành án để đến nhận tử thi về an táng. Việc giao nhận tử thi phải được thực hiện trong thời hạn 24 giờ kể từ khi thông báo và phải lập biên bản, có chữ ký của các bên giao, nhận; hết thời hạn này mà người có đơn đề nghị không đến nhận tử thi thì cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm an táng.\n2. Trường hợp không được nhận tử thi hoặc thân nhân của người bị thi hành án không có đơn đề nghị được nhận tử thi về an táng thì cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu tổ chức việc an táng. Sau 03 năm kể từ ngày thi hành án, thân nhân hoặc đại diện hợp pháp của người đã bị thi hành án được làm đơn có xác nhận của Ủy ban nhân dân cấp xã nơi cư trú đề nghị Cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu nơi đã thi hành án cho nhận hài cốt. Đơn đề nghị phải ghi rõ họ tên, địa chỉ người nhận hài cốt, quan hệ với người bị thi hành án; cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí. Trong thời hạn 07 ngày, kể từ ngày nhận được đơn, cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm xem xét, giải quyết.\nTrường hợp người bị thi hành án là người nước ngoài thì đơn đề nghị phải có xác nhận của cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người bị thi hành án mang quốc tịch và phải được dịch ra tiếng Việt. Việc giải quyết cho nhận hài cốt do cơ quan quản lý thi hành án hình sự xem xét, quyết định.
- Ngữ cảnh 3: 53/2010/qh12 Điều 57. Chế độ quản lý giam giữ, ăn, ở, mặc, sinh hoạt, gửi và nhận thư, nhận đồ vật, tiền mặt, gặp thân nhân, chăm sóc y tế\nChế độ quản lý giam giữ, ăn, ở, mặc, sinh hoạt, gửi và nhận thư, nhận đồ vật, tiền mặt, gặp thân nhân, chăm sóc y tế đối với người bị kết án tử hình trong thời gian chờ thi hành án thực hiện theo quy định của pháp luật về tạm giam.
- Ngữ cảnh 4: 82/2011/nđ-cp Điều 9. Chi phí mai táng\nChi phí mai táng người bị thi hành án tử hình bao gồm: 01 quan tài bằng gỗ thường, 01 bộ quần áo thường, 04 m vải liệm, hương, nến, rượu, cồn để làm vệ sinh khi liệm tử thi và các chi phí mai táng khác.\nTrường hợp thân nhân hoặc người đại diện hợp pháp của người bị kết án tử hình được nhận tử thi về mai táng thì phải tự chịu chi phí đưa di chuyển tử thi và phải cam kết chấp hành đúng quy định của pháp luật về bảo đảm an ninh, trật tự.
- Ngữ cảnh 5: 82/2011/nđ-cp Điều 5. Chế độ, chính sách đối với người tham gia thi hành án tử hình\n1. Người tham gia Đội thi hành án tử hình và bác sĩ xác định tĩnh mạch được hưởng chế độ bồi dưỡng bằng hai lần mức lương tối thiểu chung cho mỗi lần thi hành án và được nghỉ dưỡng 10 ngày theo quy định chung về chế độ nghỉ dưỡng đối với cán bộ, chiến sĩ Công an nhân dân, Quân đội nhân dân.\n2. Người tham gia Hội đồng thi hành án tử hình, bác sĩ pháp y, cán bộ chuyên môn, người thực hiện lăn tay người bị thi hành án tử hình, khâm liệm, an táng tử thi được hưởng chế độ bồi dưỡng bằng một lần mức lương tối thiểu chung quy định cho mỗi lần thi hành án.\n3. Những người tham gia khác được hưởng chế độ bồi dưỡng bằng một phần hai mức lương tối thiểu cho mỗi lần thi hành án.'''
question = '''Tử tù chết sau khi tiêm thuốc độc mà không có thân nhân nhận xác thì được hỗ trợ mai táng như thế nào?'''
'''Trả lời:
Dựa trên ngữ cảnh 2 (53/2010/qh12 Điều 60), nếu tử tù chết sau khi tiêm thuốc độc mà không có thân nhân nhận xác, thì cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu sẽ tổ chức việc an táng. Cụ thể:
1. **An táng**: Cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu sẽ chịu trách nhiệm an táng tử tù nếu không có thân nhân hoặc người đại diện hợp pháp đề nghị nhận tử thi.
2. **Hài cốt sau 3 năm**: Sau 3 năm kể từ ngày thi hành án, nếu thân nhân hoặc đại diện hợp pháp của người đã bị thi hành án vẫn chưa đề nghị nhận hài cốt, họ có thể làm đơn đề nghị Cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu nơi đã thi hành án cho nhận hài cốt. Đơn đề nghị phải ghi rõ họ tên, địa chỉ người nhận hài cốt, quan hệ với người bị thi hành án, cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí.
3. **Chi phí mai táng**: Theo ngữ cảnh 4 (82/2011/nđ-cp Điều 9), chi phí mai táng bao gồm:
- 01 quan tài bằng gỗ thường.
- 01 bộ quần áo thường.
- 04 m vải liệm.
- Hương, nến, rượu, cồn để làm vệ sinh khi liệm tử thi.
- Các chi phí mai táng khác.
Nếu thân nhân hoặc người đại diện hợp pháp đề nghị nhận tử thi về mai táng, họ phải tự chịu chi phí đưa di chuyển tử thi và cam kết chấp hành đúng quy định của pháp luật về bảo đảm an ninh, trật tự.
'''
```
- Ngữ cảnh 2 và ngữ cảnh 4 có chứa phần thông tin cho việc trả lời câu hỏi.
### 5.Xác định positive/negative
```python
system_prompt = "Bạn là một trợ lí Tiếng Việt nhiệt tình và trung thực. Hãy luôn trả lời một cách hữu ích nhất có thể."
template = '''Hãy xác định xem ngữ cảnh có chứa đầy đủ thông tin để trả lời câu hỏi hay không.
Chỉ cần đưa ra 1 trong 2 đáp án trong phần trả lời là "Có" hoặc "Không".
### Ngữ cảnh :
{context}
### Câu hỏi :
{question}
### Trả lời :'''
context = '''Công dụng thuốc Xelocapec Capecitabine là một hoạt chất gây độc chọn lọc với tế bào ung thư. Hoạt chất này có trong thuốc Xelocapec. Vậy thuốc Xelocapec có tác dụng gì và cần lưu ý những vấn đề nào khi điều trị bằng sản phẩm này? 1. Xelocapec là thuốc gì? Xelocapec chứa hoạt chất Capecitabine hàm lượng 500mg. Thuốc Xelocapec bào chế dạng viên nén bao phim và đóng gói mỗi hộp 3 vỉ x 10 viên. Xelocapec chứa hoạt chất Capecitabine là một dẫn chất Fluoropyrimidine carbamate với tác dụng gây độc chọn lọc với các tế bào ung thư . Mặc dù trên in vitro Capecitabine không cho thấy tác dụng độc tế bào nhưng trên in vivo, Xelocapec biến đổi liên tiếp thành chất gây độc tế bào là 5-fluorouracil (5-FU). Sự hình thành 5-FU tại khối u thông qua xúc tác một cách tối ưu của yếu tố tạo mạch liên quan là Thymidine phosphorylase, qua đó hạn chế tối đa mức độ ảnh hưởng đến nhu mô lành của 5-FU. 2. Thuốc Xelocapec có tác dụng gì? Thuốc Xelocapec được chỉ định điều trị đơn lẻ hoặc kết hợp với các liệu pháp điều trị ung thư. Xelocapec làm chậm hoặc ngăn chặn sự tăng trưởng của tế bào ung thư, do đó giảm kích thước khối u trong những trường hợp sau: Ung thư vú : Xelocapec phối hợp với Docetaxel được chỉ định điều trị ung thư vú thể tiến triển tại chỗ hoặc di căn sau khi đã thất bại với liệu pháp hóa trị; Ung thư đại trực tràng : Xelocapec được chỉ định hỗ trợ điều trị ung thư đại tràng sau phẫu thuật hoặc ung thư đại trực tràng di căn; Ung thư dạ dày : Xelocapec phối hợp với hợp chất platin được chỉ định điều trị khởi đầu cho những bệnh nhân ung thư dạ dày. Chống chỉ định của thuốc Xelocapec : Bệnh nhân quá mẫn cảm với Capecitabine hay các thành phần khác có trong Xelocapec ; Người có tiền sử gặp các phản ứng không mong muốn nghiêm trọng khi điều trị với Fluoropyrimidine; Người đang mang thai hoặc cho con bú; Suy thận nặng (độ thanh thải Creatinin <30mL/phút); Bệnh nhân đang điều trị ung thư với Sorivudin hoặc chất tương tự về mặt hóa học như Brivudin; Bệnh nhân thiếu hụt Dihydropyrimidin dehydrogenase; Bệnh nhân giảm số lượng bạch cầu hoặc tiểu cầu nặng; Suy gan nặng. 3. Liều dùng của thuốc Xelocapec Liều dùng của Xelocapec khi điều trị đơn lẻ: Ung thư đại trực tràng, ung thư vú: 1250mg/m2, uống 2 lần mỗi ngày trong thời gian 14 ngày, tiếp sau đó là 7 ngày ngưng thuốc. Liều Xelocapec trong điều trị phối hợp: Ung thư vú: Liều khởi đầu là 1250mg/m2, 2 lần uống mỗi ngày trong 2 tuần dùng phối hợp với Docetaxel, tiếp sau đó lá 1 tuần ngưng thuốc; Ung thư dạ dày, đại trực tràng: Liều khuyến cáo là 800-1000mg/m2/lần x 2 lần/ngày trong thời gian 2 tuần, sau đó 7 ngày ngưng thuốc hoặc 625mg/m2/lần x 2 lần mỗi ngày khi điều trị liên tục. Thuốc Xelocapec nên uống cùng với thức ăn, do đó thời điểm tốt nhất là trong vòng 30 phút sau bữa ăn. 4. Tác dụng phụ của thuốc Xelocapec Các triệu chứng bất thường như buồn nôn, nôn ói, giảm cảm giác ngon miệng, táo bón, cơ thể mệt mỏi, yếu sức, đau đầu, chóng mặt, khó ngủ có thể xảy ra trong thời gian dùng Xelocapec . Trong đó, tình trạng buồn nôn và nôn ói có thể nghiêm trọng nên đôi khi cần được bác sĩ chỉ định thuốc kiểm soát phù hợp. Tiêu chảy là một tác dụng phụ phổ biến khác của thuốc Xelocapec . Bệnh nhân cần uống nhiều nước khi điều trị bằng Xelocapec trừ khi bác sĩ có hướng dẫn khác. Nôn ói hoặc tiêu chảy kéo dài do thuốc Xelocapec có thể dẫn đến mất nước nghiêm trọng, vì vậy người bệnh hãy liên hệ ngay với bác sĩ nếu có các triệu chứng mất nước như giảm đi tiểu, khô miệng, tăng cảm giác khát nước hoặc chóng mặt. Tình trạng rụng tóc tạm thời xảy ra trong thời gian dùng thuốc Xelocapec và có thể hồi phục sau khi điều trị đã kết thúc. Một số bệnh nhân ghi nhận hiện tượng thay đổi móng tay tạm thời. Đối với nam giới và phụ nữ trong độ tuổi sinh đẻ, thuốc Xelocapec có thể ảnh hưởng đến khả năng có con của bệnh nhân. Bệnh nhân hãy tham khảo ý kiến bác sĩ để biết thêm chi tiết. Thuốc Xelocapec có thể làm giảm khả năng miễn dịch của cơ thể với các tác nhân nhiễm trùng, dẫn đến tăng nguy cơ mắc các bệnh nhiễm trùng nghiêm trọng (nhưng hiếm khi gây tử vong) hoặc làm cho bệnh nhiễm trùng hiện mắc nghiêm trọng hơn. Phản ứng dị ứng rất nghiêm trọng với thuốc Xelocapec rất hiếm khi xảy ra. Tuy nhiên, bệnh nhân hãy liên hệ với bác sĩ ngay lập tức nếu xuất hiện các triệu chứng của phản ứng dị ứng nghiêm trọng như phát ban, sưng ngứa mặt/lưỡi/họng, chóng mặt nghiêm trọng hoặc khó thở. 5. Tương tác thuốc của Xelocapec Hoạt chất Capecitabine trong thuốc Xelocapec có thể xảy ra tương tác thuốc nghiêm trọng với một số thuốc sau: Thuốc chống đông máu Coumarin: Trong một nghiên cứu tương tác lâm sàng, sau khi dùng Warfarin liều đơn 20mg kết hợp với Capecitabine làm tăng AUC của S-warfarin khoảng 57% và giá trị INR tăng 91%.'''
question = '''Tại sao Capecitabine trong Xelocapec không gây độc tế bào trên in vitro nhưng lại biến đổi thành 5-fluorouracil (5-FU) gây độc tế bào trên in vivo, và cơ chế nào giúp hạn chế ảnh hưởng đến nhu mô lành của 5-FU?'''
'''Trả lời: Có'''
```
**Next release**
Mô hình phiên bản tiếp theo mạnh mẽ hơn, giải quyết các dữ liệu phức tạp hơn sẽ được public sớm nhất có thể.
**Developer**
Member: Nguyễn Nho Trung, Nguyễn Nhật Quang
## Contact
**Email**: nguyennhotrung3004@gmail.com
**LinkedIn**: [Trung Nguyen Nho](https://www.linkedin.com/in/trung-nguyen-nho-604288227/)
## Citation
```Plaintext
@misc{ViRAG-Gen,
title={ViRAG-Gen: Towards a specialized LLM for RAG task in Vietnamese language.}},
author={Nguyen Nho Trung, Nguyen Nhat Quang},
year={2024},
publisher={Huggingface},
}
```
| [
"CHIA"
] |
KappaNeuro/needlepoint | KappaNeuro | text-to-image | [
"diffusers",
"text-to-image",
"stable-diffusion",
"lora",
"embroidery",
"style",
"needlepoint",
"needlework",
"base_model:stabilityai/stable-diffusion-xl-base-1.0",
"base_model:adapter:stabilityai/stable-diffusion-xl-base-1.0",
"license:other",
"region:us"
] | "2023-09-14T10:00:03Z" | 2023-09-14T10:00:07+00:00 | 493 | 5 | ---
base_model: stabilityai/stable-diffusion-xl-base-1.0
license: other
tags:
- text-to-image
- stable-diffusion
- lora
- diffusers
- embroidery
- style
- needlepoint
- needlework
instance_prompt: Needlepoint page
widget:
- text: Needlepoint - 8k, very simple overgrown greek landscape folly sewn patch
- text: Needlepoint - a coarse cross-stitch of a big ol' red truck driving from Vegas
to Nashville
- text: Needlepoint - A mixed media art piece featuring an optical illusion pattern
in the shape of a triangle made entirely out of particle board, overlaid with
delicate embroidery in a creamy chenille thread. The colors of the embroidery
should be a combination of frozen civilization, tropical escape and wild cranberry
with hints of scorpion and Penelope. The embroidery should feel alive and three-dimensional,
making the rigid surface of the particle board take on a sinewy, swirling appearance.
- text: Needlepoint - a beautiful fabric embroidered minimalist Homer Simpson and
Ronald McDonald made by a contemporary artist, textile art, contemporary artist
embroidery, mockup, Simpsons, large embroidery art of Homer Simpsons from the
tv show the Simpsons, McDonald tone, award winning photo, bright natural light,
ultra-detailed, intricate details, 16K, volumetric lighting, best quality
- text: 'Needlepoint - Introduce flowing and organic forms with crewelwork stitches,
inspired by nature and folklore. Use a variety of colorful threads to create texture
and dimension.Dive into the vibrant and eclectic world of Stella Jean and create
a whimsical embroidery design that merges various embroidery techniques with her
unique style. Combining elements of Canvaswork, Blackwork, Crewelwork, Goldwork,
Silk Shading, Whitework, and Raised Work, you will craft a piece that celebrates
the beauty of cultural diversity and the fusion of different artistic traditions.
Canvaswork: Incorporate geometric patterns inspired by traditional textiles from
different cultures, using vibrant thread colors to create a bold and eye-catching
design'
- text: Needlepoint - non-binary self portrait in embroidery, embroidery hoop, middle
age, pensive cautious optimism, portrait reflecting their best self, soft pallette,
i will not drown in the liqueur of safety
- text: Needlepoint - yarn wool intricate embroidery texture 3D beautifully stitched
fabric embroidered neutral rainbows and wildflowers on hills and fairy houses
quilted background cottagecore photorealistic
- text: Needlepoint - A painting of Rolling hills with a country road and an active
volcano in the background made of crochet polyester. Framed in a modern frame.
Studio shot on a light gray backdrop.
- text: Needlepoint - father's day stitching, close-up view father and daughter playing
and laughing together, eye-level view, soft colors, minimalistic background, in
the style of Cross-stitching
- text: Needlepoint - a boho design of flowers and mushrooms that looks like it was
printed on an 3d embroidered cross stitch background
---
# Needlepoint ([CivitAI](https://civitai.com/models/154077)
> Needlepoint - 8k, very simple overgrown greek landscape folly sewn patch
<p>Needlepoint is a form of embroidery that involves stitching decorative designs onto canvas using a needle and thread. It is often done on a stiff canvas mesh that provides a grid for the stitches.</p><p>In needlepoint, the design is typically pre-printed or hand-drawn onto the canvas, serving as a guide for the stitching process. The stitcher then uses various types of stitches, such as tent stitch, basketweave stitch, or cross-stitch, to fill in the design and create textured and colorful patterns.</p><p>Needlepoint allows for a wide range of design possibilities, from intricate and detailed images to simple geometric shapes or monograms. The choice of threads and colors adds depth and visual interest to the finished piece.</p><p>Needlepoint is a versatile craft that can be used to create various items, including decorative wall hangings, pillows, purses, and even clothing accessories like belts or shoes. It offers an opportunity for personal expression and creativity, allowing stitchers to customize designs and incorporate their own artistic flair.</p><p>The process of needlepoint can be relaxing and meditative, as the repetitive nature of stitching can provide a sense of focus and calm. It also offers a way to create lasting heirlooms or unique gifts for loved ones.</p><p>Needlepoint has a rich history, with examples of this craft dating back centuries. It continues to be practiced and enjoyed by people of all ages, combining artistic expression with a satisfying and enjoyable stitching process.</p>
## Image examples for the model:
> Needlepoint - a coarse cross-stitch of a big ol' red truck driving from Vegas to Nashville
> Needlepoint - A mixed media art piece featuring an optical illusion pattern in the shape of a triangle made entirely out of particle board, overlaid with delicate embroidery in a creamy chenille thread. The colors of the embroidery should be a combination of frozen civilization, tropical escape and wild cranberry with hints of scorpion and Penelope. The embroidery should feel alive and three-dimensional, making the rigid surface of the particle board take on a sinewy, swirling appearance.
> Needlepoint - a beautiful fabric embroidered minimalist Homer Simpson and Ronald McDonald made by a contemporary artist, textile art, contemporary artist embroidery, mockup, Simpsons, large embroidery art of Homer Simpsons from the tv show the Simpsons, McDonald tone, award winning photo, bright natural light, ultra-detailed, intricate details, 16K, volumetric lighting, best quality
> Needlepoint - Introduce flowing and organic forms with crewelwork stitches, inspired by nature and folklore. Use a variety of colorful threads to create texture and dimension.Dive into the vibrant and eclectic world of Stella Jean and create a whimsical embroidery design that merges various embroidery techniques with her unique style. Combining elements of Canvaswork, Blackwork, Crewelwork, Goldwork, Silk Shading, Whitework, and Raised Work, you will craft a piece that celebrates the beauty of cultural diversity and the fusion of different artistic traditions. Canvaswork: Incorporate geometric patterns inspired by traditional textiles from different cultures, using vibrant thread colors to create a bold and eye-catching design
> Needlepoint - non-binary self portrait in embroidery, embroidery hoop, middle age, pensive cautious optimism, portrait reflecting their best self, soft pallette, i will not drown in the liqueur of safety
> Needlepoint - yarn wool intricate embroidery texture 3D beautifully stitched fabric embroidered neutral rainbows and wildflowers on hills and fairy houses quilted background cottagecore photorealistic
> Needlepoint - A painting of Rolling hills with a country road and an active volcano in the background made of crochet polyester. Framed in a modern frame. Studio shot on a light gray backdrop.
> Needlepoint - father's day stitching, close-up view father and daughter playing and laughing together, eye-level view, soft colors, minimalistic background, in the style of Cross-stitching
> Needlepoint - a boho design of flowers and mushrooms that looks like it was printed on an 3d embroidered cross stitch background
| [
"CRAFT"
] |
RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2403.14009",
"arxiv:2403.20266",
"arxiv:2101.00027",
"arxiv:2207.00220",
"arxiv:1810.06694",
"arxiv:1911.05507",
"arxiv:1906.03741",
"arxiv:2406.17557",
"arxiv:2402.06619",
"arxiv:1803.09010",
"base_model:BSC-LT/salamandra-7b-instruct",
"base_model:quantized:BSC-LT/salamandra-7b-instruct",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-10-11T14:03:18Z" | 2024-10-11T16:46:21+00:00 | 491 | 0 | ---
base_model:
- BSC-LT/salamandra-7b-instruct
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
salamandra-7b-instruct - GGUF
- Model creator: https://huggingface.co/BSC-LT/
- Original model: https://huggingface.co/BSC-LT/salamandra-7b-instruct/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [salamandra-7b-instruct.Q2_K.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q2_K.gguf) | Q2_K | 3.08GB |
| [salamandra-7b-instruct.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.IQ3_XS.gguf) | IQ3_XS | 3.39GB |
| [salamandra-7b-instruct.IQ3_S.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.IQ3_S.gguf) | IQ3_S | 3.51GB |
| [salamandra-7b-instruct.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q3_K_S.gguf) | Q3_K_S | 3.5GB |
| [salamandra-7b-instruct.IQ3_M.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.IQ3_M.gguf) | IQ3_M | 3.6GB |
| [salamandra-7b-instruct.Q3_K.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q3_K.gguf) | Q3_K | 3.77GB |
| [salamandra-7b-instruct.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q3_K_M.gguf) | Q3_K_M | 3.77GB |
| [salamandra-7b-instruct.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q3_K_L.gguf) | Q3_K_L | 4.0GB |
| [salamandra-7b-instruct.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.IQ4_XS.gguf) | IQ4_XS | 4.18GB |
| [salamandra-7b-instruct.Q4_0.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q4_0.gguf) | Q4_0 | 4.33GB |
| [salamandra-7b-instruct.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.IQ4_NL.gguf) | IQ4_NL | 4.36GB |
| [salamandra-7b-instruct.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q4_K_S.gguf) | Q4_K_S | 4.35GB |
| [salamandra-7b-instruct.Q4_K.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q4_K.gguf) | Q4_K | 4.52GB |
| [salamandra-7b-instruct.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q4_K_M.gguf) | Q4_K_M | 4.52GB |
| [salamandra-7b-instruct.Q4_1.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q4_1.gguf) | Q4_1 | 4.72GB |
| [salamandra-7b-instruct.Q5_0.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q5_0.gguf) | Q5_0 | 5.11GB |
| [salamandra-7b-instruct.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q5_K_S.gguf) | Q5_K_S | 5.11GB |
| [salamandra-7b-instruct.Q5_K.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q5_K.gguf) | Q5_K | 5.21GB |
| [salamandra-7b-instruct.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q5_K_M.gguf) | Q5_K_M | 5.21GB |
| [salamandra-7b-instruct.Q5_1.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q5_1.gguf) | Q5_1 | 5.5GB |
| [salamandra-7b-instruct.Q6_K.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q6_K.gguf) | Q6_K | 5.94GB |
| [salamandra-7b-instruct.Q8_0.gguf](https://huggingface.co/RichardErkhov/BSC-LT_-_salamandra-7b-instruct-gguf/blob/main/salamandra-7b-instruct.Q8_0.gguf) | Q8_0 | 7.69GB |
Original model description:
---
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
language:
- bg
- ca
- code
- cs
- cy
- da
- de
- el
- en
- es
- et
- eu
- fi
- fr
- ga
- gl
- hr
- hu
- it
- lt
- lv
- mt
- nl
- nn
- \no
- oc
- pl
- pt
- ro
- ru
- sh
- sk
- sl
- sr
- sv
- uk
---
# Salamandra Model Card
Salamandra is a highly multilingual model pre-trained from scratch that comes in three different
sizes — 2B, 7B and 40B parameters — with their respective base and instruction-tuned variants.
This model card corresponds to the 7B instructed version.
To visit the model cards of other Salamandra versions, please refer to the [Model Index](#model-index).
The entire Salamandra family is released under a permissive [Apache 2.0 license]((https://www.apache.org/licenses/LICENSE-2.0)).
Along with the open weights, all training scripts and configuration files are made publicly available in [this GitHub repository](https://github.com/langtech-bsc/salamandra).
> [!WARNING]
> **DISCLAIMER:** This model is a first proof-of-concept designed to demonstrate the instruction-following capabilities of recently released base models.
> It has been optimized to engage in conversation but has *NOT* been aligned through RLHF to filter or avoid sensitive topics.
> As a result, it may generate harmful or inappropriate content.
> The team is actively working to enhance its performance through further instruction and alignment with RL techniques.
---
## Model Details
### Description
Transformer-based decoder-only language model that has been pre-trained from scratch on 7.8 trillion tokens of highly curated data.
The pre-training corpus contains text in 35 European languages and code.
### Hyperparameters
The full list of hyperparameters for each model can be found [here](https://github.com/langtech-bsc/salamandra/tree/main/configs).
### Architecture
| | |
|-------------------------|:--------------|
| Total Parameters | 7,768,117,248 |
| Embedding Parameters | 1,048,576,000 |
| Layers | 32 |
| Hidden size | 4,096 |
| Attention heads | 32 |
| Context length | 8,192 |
| Vocabulary size | 256,000 |
| Precision | bfloat16 |
| Embedding type | RoPE |
| Activation Function | SwiGLU |
| Layer normalization | RMS Norm |
| Flash attention | ✅ |
| Grouped Query Attention | ✅ |
| Num. query groups | 8 |
---
## Intended Use
### Direct Use
The models are intended for both research and commercial use in any of the languages included in the training data.
The base models are intended either for language generation or to be further fine-tuned for specific use-cases.
The instruction-tuned variants can be used as general-purpose assistants, as long as the user is fully aware of the model’s limitations.
### Out-of-scope Use
The model is not intended for malicious activities, such as harming others or violating human rights.
Any downstream application must comply with current laws and regulations.
Irresponsible usage in production environments without proper risk assessment and mitigation is also discouraged.
---
## Hardware and Software
### Training Framework
Pre-training was conducted using NVIDIA’s [NeMo Framework](https://docs.nvidia.com/nemo-framework/index.html),
which leverages PyTorch Lightning for efficient model training in highly distributed settings.
The instruction-tuned versions were produced with [FastChat](https://github.com/lm-sys/FastChat).
### Compute Infrastructure
All models were trained on [MareNostrum 5](https://www.bsc.es/ca/marenostrum/marenostrum-5), a pre-exascale EuroHPC supercomputer hosted and
operated by Barcelona Supercomputing Center.
The accelerated partition is composed of 1,120 nodes with the following specifications:
- 4x Nvidia Hopper GPUs with 64 HBM2 memory
- 2x Intel Sapphire Rapids 8460Y+ at 2.3Ghz and 32c each (64 cores)
- 4x NDR200 (BW per node 800Gb/s)
- 512 GB of Main memory (DDR5)
- 460GB on NVMe storage
|Model|Nodes|GPUs|
|:---:|:---:|:---:|
|2B|64|256|
|7B|128|512|
|40B|256 / 512|1,024 / 2,048|
---
## How to use
The instruction-following models use the commonly adopted ChatML template:
```jinja
{%- if not date_string is defined %}{%- set date_string = "2024-09-30" %}{%- endif %}{{ "<|im_start|>system\nsystem_message\nToday Date: "+ date_string +"<|im_end|>\n" }}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}
```
Where `system_message` is used to guide the model during generation and `date_string` can be set to allow the model to respond with the current date.
The exact same chat template should be used for an enhanced conversational experience.
The easiest way to apply it is by using the tokenizer's built-in functions, as shown in the following snippet.
```python
from datetime import datetime
from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch
model_id = "BSC-LT/salamandra-7b-instruct"
text = "At what temperature does water boil?"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map="auto",
torch_dtype=torch.bfloat16
)
message = [ { "role": "user", "content": text } ]
date_string = datetime.today().strftime('%Y-%m-%d')
prompt = tokenizer.apply_chat_template(
message,
tokenize=False,
add_generation_prompt=True,
date_string=date_string
)
inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
outputs = model.generate(input_ids=inputs.to(model.device), max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
Using this template, each turn is preceded by a `<|im_start|>` delimiter and the role of the entity
(either `user`, for content supplied by the user, or `assistant` for LLM responses), and finished with the `<|im_end|>` token.
---
## Data
### Pretraining Data
The training corpus consists of 2.4 trillion tokens, including 35 European languages and 92 programming languages. It amounts to a total of 33TB of pre-processed text.
Languages were sampled manually by giving x2 oversampling to Spain's co-official languages (Spanish, Catalan, Galician and Basque), code was undersampled by half,
and the rest of the languages were kept as is, resulting in the following distribution:
This highly multilingual corpus is predominantly composed of data from Colossal OSCAR,
which contributes a significant 66.06% of the total tokens.
Following this, Starcoder provides 11.91%, and Spanish Crawling adds 3.34%.
The next largest sources are French FR at 3.12% and Proof Pile at 1.98%.
Other notable contributions include Macocu, Pile of Law, and Eurlex, each contributing around 1.5% to 1.3%.
These major sources collectively form the bulk of the corpus, ensuring a rich and diverse dataset for training the language model.
The remaining 10% comes from smaller sources in various languages.
Feel free to click the expand button below to see the full list of sources.
<details>
<summary>Data Sources</summary>
| Dataset | Language | Source |
|-----------------------------------------------|---------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------|
| Parlamint corpus | at, bg, cz, dk, ee, es, es-ga, fi, fr, gb, gr, hr, hu, it, lv, nl, no, pl, pt, rs, se, si | Erjavec et al., 2021 |
| Bulgarian National Corpus | bg | [Link](http://old.dcl.bas.bg/dataset/BulNC.7z) |
| Crawl of Bulgarian news websites | bg | [Link](http://old.dcl.bas.bg/dataset/Bulgarian_news.7z) |
| Colossal OSCAR 1.0 | bg, ca, cs, cy, da, de, el, en, es, et, eu, fi, fr, ga, gl, hr, hu, it, lt, lv, mt, nl, nn, no, oc, pl, pt, ro, ru, sh, sk, sl, sr, sv, uk | Brack et al., 2024 |
| Wikimedia dumps | bg, ca, cs, da, de, el, en, es, et, eu, fi, fr, ga, gl, hr, hu, it, lt, lv, mt, nl, nn, no, pl, pt, ro, sh, sk, sl, sr, uk | [Link](https://dumps.wikimedia.org/) |
| OpenSubtitlesv2016 | bg, ca, cs, da, de, el, en, es, et, eu, fi, fr, gl, hr, it, lt, lv, nl, no, pl, pt, ro, sk, sl, sr, sv, uk | Lison & Tiedemann, 2016 |
| MaCoCu web corpus | bg, ca, el, hr, mt, sl, sr, uk | Bañón et al., 2022 |
| EurLEX-Resources | bg, cs, da, de, el, en, es, et, fi, fr, ga, hr, hu, it, lt, lv, mt, nl, pl, pt, ro, sk, sl, sv | [Link](https://huggingface.co/datasets/joelniklaus/eurlex_resources) |
| MC4-Legal | bg, cs, da, de, el, en, es, et, fi, fr, ga, hu, it, lt, lv, mt, nl, pl, pt, ro, sk, sl, sv | [Link](https://huggingface.co/datasets/joelito/legal-mc4) |
| CURLICAT Corpus | bg, hr, hu, pl, ro, sk, sl | Váradi et al., 2022 |
| CATalog | ca | Palomar-Giner et al., 2024 |
| Spanish Crawling | ca, es, eu, gl | Relevant Spanish websites crawling |
| Starcoder | code | Li et al., 2023 |
| SYN v9: large corpus of written Czech | cs | Křen et al., 2021 |
| Welsh-GOV | cy | Crawling from [Link](https://www.llyw.cymru) |
| DaNewsroom | da | Varab & Schluter, 2020 |
| Danish GigaWord | da | Strømberg-Derczynski et al., 2021 |
| DK-CLARIN Reference Corpus of General Danish | da | [Link](https://korpus.dsl.dk/clarin/) |
| The Danish Parliament Corpus 2009 - 2017, v1 | da | Hansen, 2018 |
| DeWaC | de | [Link](https://docs.sslmit.unibo.it/doku.php?id=corpora:dewac) |
| Open Legal Data - German court decisions and laws | de | Ostendorff et al., 2020 |
| Greek Legal Code | el | Papaloukas et al., 2021 |
| Greek Web Corpus | el | Outsios et al., 2018 |
| Auxiliary Mathematics Problems and Solutions (AMPS) dataset | en | Hendrycks et al., 2021 |
| BIGPATENT | en | Sharma et al., 2019 |
| FineWeb-Edu (350BT subset) | en | Penedo et al., 2024 |
| peS2o | en | Soldaini & Lo, 2023 |
| PG-19 | en | Rae et al., 2019 |
| Pile of Law (selected subsets) | en | Henderson* et al., 2022 |
| proof-pile | en | [Link](https://huggingface.co/datasets/hoskinson-center/proof-pile) |
| RedPajama-Data T1 (StackExchange subset) | en | Computer, 2023 |
| The Pile (PhilPapers subset) | en | Gao et al., 2021 |
| Biomedical | es | Internally generated scientific dataset: Dialnet, Scielo, CSIC, TDX, BSC, UCM |
| HPLTDatasets v1 - Spanish | es | de Gibert et al., 2024 |
| Legal | es | Internally generated legal dataset: BOE, BORME, Senado, Congreso, Spanish court orders, DOGC |
| Scientific | es | Internally generated scientific dataset: Wikipedia LS, Pubmed, MeSpEn, patents, clinical cases, medical crawler |
| Spanish Legal Domain Corpora | es | Gutiérrez-Fandiño et al., 2021 |
| Estonian National Corpus 2021 | et | Koppel & Kallas, 2022 |
| Estonian Reference Corpus | et | [Link](https://www.cl.ut.ee/korpused/segakorpus/) |
| EusCrawl (w/o Wikipedia or NC-licenses) | eu | Artetxe et al., 2022 |
| Latxa Corpus v1.1 | eu | Etxaniz et al., 2024 [Link](https://huggingface.co/datasets/HiTZ/latxa-corpus-v1.1) |
| Aya Dataset (w/o Evaluation Suite) | eu, hr, nl, fi, ka, hu, lt, nn, ro, sk, lv, cy, bg, cs, en, fr, de, ga, mt, pl, ru, sl, sv, ca, da, et, gl, el, it, no, pt, sr, es, uk | Singh et al., 2024 |
| Yle Finnish News Archive | fi | [Link](http://urn.fi/urn:nbn:fi:lb-2021050401) |
| CaBeRnet: a New French Balanced Reference Corpus | fr | Popa-Fabre et al., 2020 |
| French Public Domain Books | fr | [Link](https://huggingface.co/datasets/PleIAs/French-PD-Books) |
| French Public Domain Newspapers | fr | [Link](https://huggingface.co/datasets/PleIAs/French-PD-Newspapers) |
| Irish Universal Dependencies | ga | [Link](https://universaldependencies.org/ga/index.html) |
| The Gaois bilingual corpus of English-Irish legislation (Irish legislation) | ga | [Link](https://portulanclarin.net/repository/browse/the-gaois-bilingual-corpus-of-english-irish-legislation-processed/daeac17c9e3511ea9b7f02420a000407b83de243dc0b469aab41084386c5b80f/) |
| CorpusNÓS | gl | de-Dios-Flores et al., 2024 |
| Croatian web corpus hrWaC 2.1 | hr | Ljubešić & Klubička, 2014 |
| ITWaC | it | [Link](https://docs.sslmit.unibo.it/doku.php?id=corpora:itwac) |
| Corpus of State-related content from the Latvian Web (Processed) | lv | [Link](https://catalog.elra.info/en-us/repository/browse/ELRA-W0169/) |
| Korpus Malti | mt | Micallef et al., 2022 |
| SoNaR Corpus NC 1.2 | nl | [Link](https://taalmaterialen.ivdnt.org/download/tstc-sonar-corpus/) |
| Norwegian Colossal Corpus | nn, no | Kummervold et al., 2021 |
| Occitan Corpus | oc | Provided by [IEA](https://www.institutestudisaranesi.cat/) |
| NKJP-PodkorpusMilionowy-1.2 (National Corpus of Polish) | pl | Lewandowska-Tomaszczyk et al., 2013 |
| Polish Parliamentary Corpus / Korpus Dyskursu Parlamentarnego | pl | Ogrodniczuk, 2018 |
| Brazilian Portuguese Web as Corpus | pt | Wagner Filho et al., 2018 |
| ParlamentoPT | pt | Rodrigues et al., 2023 |
| MARCELL Romanian legislative subcorpus v2 | ro | [Link](https://elrc-share.eu/reposMARCELL%20Romanian%20legislative%20subcorpus%20v2itory/browse/marcell-romanian-legislative-subcorpus-v2/2da548428b9d11eb9c1a00155d026706ce94a6b59ffc4b0e9fb5cd9cebe6889e/) |
| Korpus slovenských právnych predpisov v1.9 | sk | [Link](https://www.juls.savba.sk/data/marcell/legal-sk-20220322-1.9.ver.xz) |
| od-justice 2.0 | sk | [Link](https://www.juls.savba.sk/data/od-justice/od-justice-2.0.ver.xz) |
| Corpus of academic Slovene KAS 2.0 | sl | Žagar et al., 2022 |
| slWaC web corpus | sl | Erjavec et al., 2015 |
| SrpKorSubset (news, legal, academic, conversation, literary) | sr | [Link](http://www.korpus.matf.bg.ac.rs/) |
| The Swedish Culturomics Gigaword Corpus | sv | Rødven-Eide, 2016 |
| Corpus of laws and legal acts of Ukraine | uk | [Link](https://lang.org.ua/en/corpora/#anchor7) |
<details>
<summary>References</summary>
- Abadji, J., Suárez, P. J. O., Romary, L., & Sagot, B. (2021). Ungoliant: An optimized pipeline for the generation of a very large-scale multilingual web corpus (H. Lüngen, M. Kupietz, P. Bański, A. Barbaresi, S. Clematide, & I. Pisetta, Eds.; pp. 1–9). Leibniz-Institut für Deutsche Sprache. [Link](https://doi.org/10.14618/ids-pub-10468)
- Artetxe, M., Aldabe, I., Agerri, R., Perez-de-Viñaspre, O., & Soroa, A. (2022). Does Corpus Quality Really Matter for Low-Resource Languages?
- Bañón, M., Esplà-Gomis, M., Forcada, M. L., García-Romero, C., Kuzman, T., Ljubešić, N., van Noord, R., Sempere, L. P., Ramírez-Sánchez, G., Rupnik, P., Suchomel, V., Toral, A., van der Werff, T., & Zaragoza, J. (2022). MaCoCu: Massive collection and curation of monolingual and bilingual data: Focus on under-resourced languages. Proceedings of the 23rd Annual Conference of the European Association for Machine Translation, 303–304. [Link](https://aclanthology.org/2022.eamt-1.41)
- Brack, M., Ostendorff, M., Suarez, P. O., Saiz, J. J., Castilla, I. L., Palomar-Giner, J., Shvets, A., Schramowski, P., Rehm, G., Villegas, M., & Kersting, K. (2024). Community OSCAR: A Community Effort for Multilingual Web Data. [Link](https://occiglot.eu/papers/Community_Oscar.pdf)
- Computer, T. (2023). RedPajama: An Open Source Recipe to Reproduce LLaMA training dataset [Computer software]. [Link](https://github.com/togethercomputer/RedPajama-Data)
- de Gibert, O., Nail, G., Arefyev, N., Bañón, M., van der Linde, J., Ji, S., Zaragoza-Bernabeu, J., Aulamo, M., Ramírez-Sánchez, G., Kutuzov, A., Pyysalo, S., Oepen, S., & Tiedemann, J. (2024). A New Massive Multilingual Dataset for High-Performance Language Technologies (arXiv:2403.14009). arXiv. [Link](http://arxiv.org/abs/2403.14009)
- Dodge, J., Sap, M., Marasović, A., Agnew, W., Ilharco, G., Groeneveld, D., Mitchell, M., & Gardner, M. (2021). Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpus. In M.-F. Moens, X. Huang, L. Specia, & S. W. Yih (Eds.), Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (pp. 1286–1305). Association for Computational Linguistics. [Link](https://doi.org/10.18653/v1/2021.emnlp-main.98)
- Erjavec, T., Ljubešić, N., & Logar, N. (2015). The slWaC corpus of the Slovene web. Informatica (Slovenia), 39, 35–42.
- Erjavec, T., Ogrodniczuk, M., Osenova, P., Ljubešić, N., Simov, K., Grigorova, V., Rudolf, M., Pančur, A., Kopp, M., Barkarson, S., Steingrímsson, S. hór, van der Pol, H., Depoorter, G., de Does, J., Jongejan, B., Haltrup Hansen, D., Navarretta, C., Calzada Pérez, M., de Macedo, L. D., … Rayson, P. (2021). Linguistically annotated multilingual comparable corpora of parliamentary debates ParlaMint.ana 2.1. [Link](http://hdl.handle.net/11356/1431)
- Etxaniz, J., Sainz, O., Perez, N., Aldabe, I., Rigau, G., Agirre, E., Ormazabal, A., Artetxe, M., & Soroa, A. (2024). Latxa: An Open Language Model and Evaluation Suite for Basque. [Link] (https://arxiv.org/abs/2403.20266)
- Gao, L., Biderman, S., Black, S., Golding, L., Hoppe, T., Foster, C., Phang, J., He, H., Thite, A., Nabeshima, N., Presser, S., & Leahy, C. (2021). The Pile: An 800GB Dataset of Diverse Text for Language Modeling. CoRR, abs/2101.00027. [Link](https://arxiv.org/abs/2101.00027)
- Gutiérrez-Fandiño, A., Armengol-Estapé, J., Gonzalez-Agirre, A., & Villegas, M. (2021). Spanish Legalese Language Model and Corpora.
- Hansen, D. H. (2018). The Danish Parliament Corpus 2009—2017, v1. [Link](http://hdl.handle.net/20.500.12115/8)
- Henderson*, P., Krass*, M. S., Zheng, L., Guha, N., Manning, C. D., Jurafsky, D., & Ho, D. E. (2022). Pile of Law: Learning Responsible Data Filtering from the Law and a 256GB Open-Source Legal Dataset. arXiv. [Link](https://arxiv.org/abs/2207.00220)
- Hendrycks, D., Burns, C., Kadavath, S., Arora, A., Basart, S., Tang, E., Song, D., & Steinhardt, J. (2021). Measuring Mathematical Problem Solving With the MATH Dataset. NeurIPS.
- Jansen, T., Tong, Y., Zevallos, V., & Suarez, P. O. (2022). Perplexed by Quality: A Perplexity-based Method for Adult and Harmful Content Detection in Multilingual Heterogeneous Web Data.
- Koppel, K., & Kallas, J. (2022). Eesti keele ühendkorpuste sari 2013–2021: Mahukaim eestikeelsete digitekstide kogu. Eesti Rakenduslingvistika Ühingu Aastaraamat Estonian Papers in Applied Linguistics, 18, 207–228. [Link](https://doi.org/10.5128/erya18.12)
- Křen, M., Cvrček, V., Henyš, J., Hnátková, M., Jelínek, T., Kocek, J., Kováříková, D., Křivan, J., Milička, J., Petkevič, V., Procházka, P., Skoumalová, H., Šindlerová, J., & Škrabal, M. (2021). SYN v9: Large corpus of written Czech. [Link](http://hdl.handle.net/11234/1-4635)
- Kreutzer, J., Caswell, I., Wang, L., Wahab, A., van Esch, D., Ulzii-Orshikh, N., Tapo, A., Subramani, N., Sokolov, A., Sikasote, C., Setyawan, M., Sarin, S., Samb, S., Sagot, B., Rivera, C., Rios, A., Papadimitriou, I., Osei, S., Suarez, P. O., … Adeyemi, M. (2022). Quality at a Glance: An Audit of Web-Crawled Multilingual Datasets. Transactions of the Association for Computational Linguistics, 10, 50–72. [Link](https://doi.org/10.1162/tacl_a_00447)
- Kummervold, P. E., De la Rosa, J., Wetjen, F., & Brygfjeld, S. A. (2021). Operationalizing a National Digital Library: The Case for a Norwegian Transformer Model. In S. Dobnik & L. Øvrelid (Eds.), Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa) (pp. 20–29). Linköping University Electronic Press, Sweden. [Link](https://aclanthology.org/2021.nodalida-main.3)
- Lewandowska-Tomaszczyk, B., Górski, R., Łaziński, M., & Przepiórkowski, A. (2013). The National Corpus of Polish (NKJP). Language use and data analysis. 309–319.
- Li, R., Allal, L. B., Zi, Y., Muennighoff, N., Kocetkov, D., Mou, C., Marone, M., Akiki, C., Li, J., Chim, J., Liu, Q., Zheltonozhskii, E., Zhuo, T. Y., Wang, T., Dehaene, O., Davaadorj, M., Lamy-Poirier, J., Monteiro, J., Shliazhko, O., … Vries, H. de. (2023). StarCoder: May the source be with you!
- Lison, P., & Tiedemann, J. (2016). OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles. In N. Calzolari, K. Choukri, T. Declerck, S. Goggi, M. Grobelnik, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, & S. Piperidis (Eds.), Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16) (pp. 923–929). European Language Resources Association (ELRA). [Link](https://aclanthology.org/L16-1147)
- Ljubešić, N., & Klubička, F. (2014). Bs,hr,srWaC - Web Corpora of Bosnian, Croatian and Serbian. In F. Bildhauer & R. Schäfer (Eds.), Proceedings of the 9th Web as Corpus Workshop (WaC-9) (pp. 29–35). Association for Computational Linguistics. [Link](https://doi.org/10.3115/v1/W14-0405)
- Micallef, K., Gatt, A., Tanti, M., van der Plas, L., & Borg, C. (2022). Pre-training Data Quality and Quantity for a Low-Resource Language: New Corpus and BERT Models for Maltese. Proceedings of the Third Workshop on Deep Learning for Low-Resource Natural Language Processing, 90–101. [Link](https://doi.org/10.18653/v1/2022.deeplo-1.10)
- Ogrodniczuk, M. (2018). Polish Parliamentary Corpus. [Link](https://api.semanticscholar.org/CorpusID:235134113)
- Ostendorff, M., Blume, T., & Ostendorff, S. (2020). Towards an Open Platform for Legal Information. Proceedings of the ACM/IEEE Joint Conference on Digital Libraries in 2020, 385–388. [Link](https://doi.org/10.1145/3383583.3398616)
- Ostendorff, M., Suarez, P. O., Lage, L. F., & Rehm, G. (2024). LLM-Datasets: An Open Framework for Pretraining Datasets of Large Language Models. First Conference on Language Modeling. [Link](https://openreview.net/forum?id=5RdIMlGLXL)
- Outsios, S., Skianis, K., Meladianos, P., Xypolopoulos, C., & Vazirgiannis, M. (2018). Word Embeddings from Large-Scale Greek Web content. arXiv Preprint arXiv:1810.06694.
- Palomar-Giner, J., Saiz, J. J., Espuña, F., Mina, M., Da Dalt, S., Llop, J., Ostendorff, M., Ortiz Suarez, P., Rehm, G., Gonzalez-Agirre, A., & Villegas, M. (2024). A CURATEd CATalog: Rethinking the Extraction of Pretraining Corpora for Mid-Resourced Languages. In N. Calzolari, M.-Y. Kan, V. Hoste, A. Lenci, S. Sakti, & N. Xue (Eds.), Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) (pp. 335–349). ELRA and ICCL. [Link](https://aclanthology.org/2024.lrec-main.31)
- Papaloukas, C., Chalkidis, I., Athinaios, K., Pantazi, D.-A., & Koubarakis, M. (2021). Multi-granular Legal Topic Classification on Greek Legislation. Proceedings of the Natural Legal Language Processing Workshop 2021, 63–75. [Link](https://doi.org/10.48550/arXiv.2109.15298)
- Popa-Fabre, M., Ortiz Suárez, P. J., Sagot, B., & de la Clergerie, É. (2020). French Contextualized Word-Embeddings with a sip of CaBeRnet: A New French Balanced Reference Corpus. Proceedings of the 8th Workshop on Challenges in the Management of Large Corpora, 15–23. [Link](https://aclanthology.org/2020.cmlc-1.3)
- Rae, J. W., Potapenko, A., Jayakumar, S. M., Hillier, C., & Lillicrap, T. P. (2019). Compressive Transformers for Long-Range Sequence Modelling. arXiv Preprint. [Link](https://arxiv.org/abs/1911.05507)
- Rodrigues, J., Gomes, L., Silva, J., Branco, A., Santos, R., Cardoso, H. L., & Osório, T. (2023). Advancing Neural Encoding of Portuguese with Transformer Albertina PT-\*.
- Rødven-Eide, S. (2016). The Swedish Culturomics Gigaword CorpusThe Swedish Culturomics Gigaword Corpus [Dataset]. Språkbanken Text. [Link](https://doi.org/10.23695/3WMV-1Z09)
- Sharma, E., Li, C., & Wang, L. (2019). BIGPATENT: A Large-Scale Dataset for Abstractive and Coherent Summarization. CoRR, abs/1906.03741. [Link](http://arxiv.org/abs/1906.03741)
- Soldaini, L., & Lo, K. (2023). peS2o (Pretraining Efficiently on S2ORC) Dataset. Allen Institute for AI.
- Strømberg-Derczynski, L., Ciosici, M., Baglini, R., Christiansen, M. H., Dalsgaard, J. A., Fusaroli, R., Henrichsen, P. J., Hvingelby, R., Kirkedal, A., Kjeldsen, A. S., Ladefoged, C., Nielsen, F. Å., Madsen, J., Petersen, M. L., Rystrøm, J. H., & Varab, D. (2021). The Danish Gigaword Corpus. Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), 413–421. [Link](https://aclanthology.org/2021.nodalida-main.46)
- Subramani, N., Luccioni, S., Dodge, J., & Mitchell, M. (2023). Detecting Personal Information in Training Corpora: An Analysis. 208–220. [Link](https://doi.org/10.18653/v1/2023.trustnlp-1.18)
- Varab, D., & Schluter, N. (2020). DaNewsroom: A Large-scale Danish Summarisation Dataset. Proceedings of The 12th Language Resources and Evaluation Conference, 6731–6739. [Link](https://www.aclweb.org/anthology/2020.lrec-1.831)
- Váradi, T., Nyéki, B., Koeva, S., Tadić, M., Štefanec, V., Ogrodniczuk, M., Nitoń, B., Pezik, P., Barbu Mititelu, V., Irimia, E., Mitrofan, M., Tufi\textcommabelows, D., Garabík, R., Krek, S., & Repar, A. (2022). Introducing the CURLICAT Corpora: Seven-language Domain Specific Annotated Corpora from Curated Sources. In N. Calzolari, F. Béchet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, J. Odijk, & S. Piperidis (Eds.), Proceedings of the Thirteenth Language Resources and Evaluation Conference (pp. 100–108). European Language Resources Association. [Link](https://aclanthology.org/2022.lrec-1.11)
- Wagner Filho, J. A., Wilkens, R., Idiart, M., & Villavicencio, A. (2018). The brwac corpus: A new open resource for brazilian portuguese. Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018).
- Žagar, A., Kavaš, M., Robnik-Šikonja, M., Erjavec, T., Fišer, D., Ljubešić, N., Ferme, M., Borovič, M., Boškovič, B., Ojsteršek, M., & Hrovat, G. (2022). Corpus of academic Slovene KAS 2.0. [Link](http://hdl.handle.net/11356/1448)
- Alicia Parrish, Angelica Chen, Nikita Nangia, Vishakh Padmakumar, Jason Phang, Jana Thompson, Phu Mon Htut, and Samuel Bowman. 2022. BBQ: A hand-built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2086–2105, Dublin, Ireland. Association for Computational Linguistics.
- Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, and Nanyun Peng. 2019. The Woman Worked as a Babysitter: On Biases in Language Generation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3407–3412, Hong Kong, China. Association for Computational Linguistics.
- Clark, P., Cowhey, I., Etzioni, O., Khot, T., Sabharwal, A., Schoenick, C., & Tafjord, O. (2018). Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge. arXiv:1803. 05457v1.
- Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. 2013. Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1631–1642, Seattle, Washington, USA. Association for Computational Linguistics.
- Penedo, G., Kydlíček, H., allal, L. B., Lozhkov, A., Mitchell, M., Raffel, C., Von Werra, L., & Wolf, T. (2024). The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale (arXiv:2406.17557). arXiv. http://arxiv.org/abs/2406.17557
- Singh, S., Vargus, F., Dsouza, D., Karlsson, B. F., Mahendiran, A., Ko, W.-Y., Shandilya, H., Patel, J., Mataciunas, D., OMahony, L., Zhang, M., Hettiarachchi, R., Wilson, J., Machado, M., Moura, L. S., Krzemiński, D., Fadaei, H., Ergün, I., Okoh, I., … Hooker, S. (2024). Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning (arXiv:2402.06619). arXiv. http://arxiv.org/abs/2402.06619
</details>
</details>
The model was trained for 3 epochs, with two final rounds of 0.3B higher-quality tokens each,
meaning that the total number of tokens seen during pre-training amounts to roughly 7.8 trillion tokens.
We provide an extense Datasheet section following the best practices defined by [(Gebru et al., 2021)](https://arxiv.org/pdf/1803.09010).
<details>
<summary>Datasheet</summary>
#### Motivation
**For what purpose was the dataset created? Was there a specific task in mind? Was there a specific gap that needed to be filled? Please provide a description.**
The purpose of creating this dataset is to pre-train the Salamandra family of multilingual models with high performance in a large number of
European languages (35) and code (including 92 different programming languages). In addition, we aim to represent especially the co-official
languages of Spain: Spanish, Catalan, Galician, and Basque. This is the reason why we carry out an oversampling of these languages.
We detected that there is a great lack of massive multilingual data, especially in minority languages (Ostendorff & Rehm, 2023), so part of
our efforts in the creation of this pre-training dataset have resulted in the contribution to large projects such as the Community OSCAR
(Brack et al., 2024), which includes 151 languages and 40T words, or CATalog (Palomar-Giner et al., 2024), the largest open dataset in
Catalan in the world.
**Who created the dataset (e.g., which team, research group) and on behalf of which entity (e.g., company, institution, organization)?**
The dataset has been created by the Language Technologies unit (LangTech) of the Barcelona Supercomputing Center - Centro Nacional de
Supercomputación (BSC-CNS), which aims to advance the field of natural language processing through cutting-edge research and development
and the use of HPC. In particular, it was created by the unit's data team, the main contributors being Javier Saiz, Ferran Espuña, and
Jorge Palomar.
However, the creation of the dataset would not have been possible without the collaboration of a large number of collaborators, partners,
and public institutions, which can be found in detail in the acknowledgements.
**Who funded the creation of the dataset? If there is an associated grant, please provide the name of the grantor and the grant name and number.**
This work/research has been promoted and financed by the Government of Catalonia through the [Aina project](https://projecteaina.cat/).
#### Composition
**What do the instances that comprise the dataset represent (e.g., documents, photos, people, countries)? Are there multiple types of instances (e.g., movies, users, and ratings; people and interactions between them; nodes and edges)? Please provide a description.**
The dataset consists entirely of text documents in various languages. Specifically, data was mainly sourced from the following databases and
repositories:
- **Common Crawl:** Repository that holds website data and is run by the Common Crawl non-profit organization. It is updated monthly and is
distributed under the CC0 1.0 public domain license.
- **GitHub:** Community platform that allows developers to create, store, manage, and share their code. Repositories are crawled and then
distributed with their original licenses, which may vary from permissive to non-commercial licenses.
- **Wikimedia:** Database that holds the collection databases managed by the Wikimedia Foundation, including Wikipedia, Wikibooks, Wikinews,
Wikiquote, Wikisource, and Wikivoyage. It is updated monthly and is distributed under Creative Commons Attribution-ShareAlike License 4.0.
- **EurLex:** Repository that holds the collection of legal documents from the European Union, available in all of the EU’s 24 official
languages and run by the Publications Office of the European Union. It is updated daily and is distributed under the Creative Commons
Attribution 4.0 International license.
- **Other repositories:** Specific repositories were crawled under permission for domain-specific corpora, which include academic, legal,
and newspaper repositories.
We provide a complete list of dataset sources at the end of this section.
**How many instances are there in total (of each type, if appropriate)?**
The dataset contains a diverse range of instances across multiple languages, with notable adjustments for certain languages. English
represents the largest portion, accounting for 39.08% of the total data. Spanish was upsampled by a factor of 2, bringing its share to 16.59%,
while Catalan (1.84%), Basque (0.26%), and Galician (0.36%) were also upsampled by 2. On the other hand, code-related data was downsampled
by half, making up 6.42% of the total. Other prominent languages include French (6.59%), Russian (5.39%), German (4.25%), and Hungarian
(3.93%), with several additional languages contributing between 1% and 2%, and smaller portions represented by a variety of others.
**Does the dataset contain all possible instances or is it a sample (not necessarily random) of instances from a larger set? If the dataset is a sample, then what is the larger set? Is the sample representative of the larger set (e.g., geographic coverage)? If so, please describe how this representativeness was validated/verified. If it is not representative of the larger set, please describe why not (e.g., to cover a more diverse range of instances, because instances were withheld or unavailable).**
The dataset is a sample from multiple sources, with different weights based on the primary language of the content: Spanish, Catalan,
Basque, and Galician content was upsampled by a factor of two, while programming languages were downsampled by a factor of half. Other
sources were sampled in proportion to their occurrence.
**What data does each instance consist of? “Raw” data (e.g., unprocessed text or images) or features? In either case, please provide a description.**
Each instance consists of a text document processed for deduplication, language identification, and source-specific filtering. Some
documents required optical character recognition (OCR) to extract text from non-text formats such as PDFs.
**Is there a label or target associated with each instance? If so, please provide a description.**
Each instance is labeled with a unique identifier, the primary language of the content, and the URL for web-sourced instances. Additional
labels were automatically assigned to detect specific types of content —harmful or toxic content— and to assign preliminary indicators of
undesired qualities —very short documents, high density of symbols, etc.— which were used for filtering instances.
**Is any information missing from individual instances? If so, please provide a description, explaining why this information is missing (e.g., because it was unavailable). This does not include intentionally removed information, but might include, e.g., redacted text.**
No significant information is missing from the instances.
**Are relationships between individual instances made explicit (e.g., users’ movie ratings, social network links)? If so, please describe how these relationships are made explicit.**
Instances are related through shared metadata, such as source and language identifiers.
**Are there recommended data splits (e.g., training, development/validation, testing)? If so, please provide a description of these splits, explaining the rationale behind them.**
The dataset is split randomly into training, validation, and test sets.
**Are there any errors, sources of noise, or redundancies in the dataset? If so, please provide a description.**
Despite removing duplicated instances within each source, redundancy remains at the paragraph and sentence levels, particularly in
web-sourced instances where SEO techniques and templates contribute to repeated textual patterns. Some instances may also be duplicated
across sources due to format variations.
**Is the dataset self-contained, or does it link to or otherwise rely on external resources (e.g., websites, tweets, other datasets)? If it links to or relies on external resources, a) are there guarantees that they will exist, and remain constant, over time; b) are there official archival versions of the complete dataset (i.e., including the external resources as they existed at the time the dataset was created); c) are there any restrictions (e.g., licenses, fees) associated with any of the external resources that might apply to a dataset consumer? Please provide descriptions of all external resources and any restrictions associated with them, as well as links or other access points, as appropriate.**
The dataset is self-contained and does not rely on external resources.
**Does the dataset contain data that might be considered confidential (e.g., data that is protected by legal privilege or by doctor–patient confidentiality, data that includes the content of individuals’ non-public communications)? If so, please provide a description.**
The dataset does not contain confidential data.
**Does the dataset contain data that, if viewed directly, might be offensive, insulting, threatening, or might otherwise cause anxiety? If so, please describe why. If the dataset does not relate to people, you may skip the remaining questions in this section.**
The dataset includes web-crawled content, which may overrepresent pornographic material across languages (Kreutzer et al., 2022). Although
pre-processing techniques were applied to mitigate offensive content, the heterogeneity and scale of web-sourced data make exhaustive
filtering challenging, which makes it next to impossible to identify all adult content without falling into excessive filtering, which may
negatively influence certain demographic groups (Dodge et al., 2021).
**Does the dataset identify any subpopulations (e.g., by age, gender)? If so, please describe how these subpopulations are identified and provide a description of their respective distributions within the dataset.**
The dataset does not explicitly identify any subpopulations.
**Is it possible to identify individuals (i.e., one or more natural persons), either directly or indirectly (i.e., in combination with other data) from the dataset? If so, please describe how.**
Web-sourced instances in the dataset may contain personally identifiable information (PII) that is publicly available on the Web, such as
names, IP addresses, email addresses, and phone numbers. While it would be possible to indirectly identify individuals through the
combination of multiple data points, the nature and scale of web data makes it difficult to parse such information. In any case, efforts are
made to filter or anonymize sensitive data during pre-processing, but some identifiable information may remain in the dataset.
**Does the dataset contain data that might be considered sensitive in any way? If so, please provide a description.**
Given that the dataset includes web-sourced content and other publicly available documents, instances may inadvertently reveal financial
information, health-related details, or forms of government identification, such as social security numbers (Subramani et al., 2023),
especially if the content originates from less-regulated sources or user-generated platforms.
#### Collection Process
**How was the data collected?**
This dataset is constituted by combining several sources, whose acquisition methods can be classified into three groups:
- Web-sourced datasets with some preprocessing available under permissive license (p.e. Common Crawl).
- Domain-specific or language-specific raw crawls (p.e. Spanish Crawling).
- Manually curated data obtained through collaborators, data providers (by means of legal assignment agreements) or open source projects
(p.e. CATalog).
**What mechanisms or procedures were used to collect the data? How were these mechanisms or procedures validated?**
According to the three groups previously defined, these are the mechanisms used in each of them:
- Open direct download. Validation: data integrity tests.
- Ad-hoc scrapers or crawlers. Validation: software unit and data integrity tests.
- Direct download via FTP, SFTP, API or S3. Validation: data integrity tests.
**If the dataset is a sample from a larger set, what was the sampling strategy?**
The sampling strategy was to use the whole dataset resulting from the filtering explained in the ‘preprocessing/cleaning/labelling’ section,
with the particularity that an upsampling of 2 (i.e. twice the probability of sampling a document) was performed for the co-official
languages of Spain (Spanish, Catalan, Galician, Basque), and a downsampling of 1/2 was applied for code (half the probability of sampling a
code document, evenly distributed among all programming languages).
**Who was involved in the data collection process and how were they compensated?**
This data is generally extracted, filtered and sampled by automated processes. The code required to run these processes has been developed
entirely by members of the LangTech data team, or otherwise obtained from open-source software. Furthermore, there has been no monetary
consideration for acquiring data from suppliers.
**Over what timeframe was the data collected? Does this timeframe match the creation timeframe of the data associated with the instances? If not, please describe the timeframe in which the data associated with the instances was created.**
Data were acquired and processed from April 2023 to April 2024. However, as mentioned, much data has been obtained from open projects such
as Common Crawl, which contains data from 2014, so it is the end date (04/2024) rather than the start date that is important.
**Were any ethical review processes conducted? If so, please provide a description of these review processes, including the outcomes, as well as a link or other access point to any supporting documentation.**
No particular ethical review process has been carried out as the data is mostly open and not particularly sensitive. However, we have an
internal evaluation team and a bias team to monitor ethical issues. In addition, we work closely with ‘Observatori d'Ètica en Intel·ligència
Artificial’ (OEIAC) and ‘Agencia Española de Supervisión de la Inteligencia Artificial’ (AESIA) to audit the processes we carry out from an
ethical and legal point of view, respectively.
#### Preprocessing
**Was any preprocessing/cleaning/labeling of the data done? If so, please provide a description. If not, you may skip the remaining questions in this section.**
Instances of text documents were not altered, but web-sourced documents were filtered based on specific criteria along two dimensions:
- Quality: documents with a score lower than 0.8, based on undesired qualities, such as documents with low number of lines, very short
sentences, presence of long footers and headers, and high percentage of punctuation, obtained through CURATE (Palomar-Giner et al., 2024)
were filtered out.
- Harmful or adult content: documents originating from Colossal OSCAR were filtered using LLM-Datasets (Ostendorff et al., 2024) based on
the perplexity from a language model (‘harmful_pp’ field) provided by the Ungoliant pipeline (Abadji et al., 2021).
**Was the “raw” data saved in addition to the preprocessed/cleaned/labeled data? If so, please provide a link or other access point to the “raw” data.**
The original raw data was not kept.
**Is the software that was used to preprocess/clean/label the data available? If so, please provide a link or other access point.**
Yes, the preprocessing and filtering software is open-sourced. The [CURATE](https://github.com/langtech-bsc/CURATE) pipeline was used for Spanish Crawling and CATalog,
and the [Ungoliant](https://github.com/oscar-project/ungoliant) pipeline was used for the OSCAR project.
#### Uses
**Has the dataset been used for any tasks already? If so, please provide a description.**
Pre-train the Salamandra model family.
**What (other) tasks could the dataset be used for?**
The data can be used primarily to pre-train other language models, which can then be used for a wide range of use cases. The dataset could
also be used for other tasks such as fine-tuning language models, cross-lingual NLP tasks, machine translation, domain-specific text
generation, and language-specific data analysis.
**Is there anything about the composition of the dataset or the way it was collected and preprocessed/cleaned/labeled that might impact future uses? Is there anything a dataset consumer could do to mitigate these risks or harms?**
Web-crawled content is over-represented with standard language varieties, impacting language model performance for minority languages.
Language diversity in data is crucial to avoid bias, especially in encoding non-standard dialects, preventing the exclusion of demographic
groups. Moreover, despite legal uncertainties in web-scraped data, we prioritize permissive licenses and privacy protection measures,
acknowledging the challenges posed by personally identifiable information (PII) within large-scale datasets. Our ongoing efforts aim to
address privacy concerns and contribute to a more inclusive linguistic dataset.
**Are there tasks for which the dataset should not be used?**
-
#### Distribution
**Will the dataset be distributed to third parties outside of the entity on behalf of which the dataset was created? If so, please provide a description.**
The dataset will not be released or distributed to third parties. Any related question to distribution is omitted in this section.
#### Maintenance
**Who will be supporting/hosting/maintaining the dataset?**
The dataset will be hosted by the Language Technologies unit (LangTech) of the Barcelona Supercomputing Center (BSC). The team will ensure
regular updates and monitor the dataset for any issues related to content integrity, legal compliance, and bias for the sources they are
responsible for.
**How can the owner/curator/manager of the dataset be contacted?**
The data owner may be contacted with the email address langtech@bsc.es.
**Will the dataset be updated?**
The dataset will not be updated.
**If the dataset relates to people, are there applicable limits on the retention of the data associated with the instances? If so, please describe these limits and explain how they will be enforced.**
The dataset does not keep sensitive data that could allow direct identification of individuals, apart from the data that is publicly
available in web-sourced content. Due to the sheer volume and diversity of web data, it is not feasible to notify individuals or manage data
retention on an individual basis. However, efforts are made to mitigate the risks associated with sensitive information through
pre-processing and filtering to remove identifiable or harmful content. Despite these measures, vigilance is maintained to address potential
privacy and ethical issues.
**Will older versions of the dataset continue to be supported/hosted/maintained? If so, please describe how. If not, please describe how its obsolescence will be communicated to dataset consumers.**
Since the dataset will not be updated, only the final version will be kept.
**If others want to extend/augment/build on/contribute to the dataset, is there a mechanism for them to do so?**
The dataset does not allow for external contributions.
</details>
### Finetuning Data
This instruction-tuned variant has been trained with a mixture of 276k English, Spanish, and Catalan multi-turn instructions gathered from open datasets:
| Dataset | ca | en | es |
|-----------------------|:------:|:------:|:------:|
| alpaca-cleaned | - | 50,000 | - |
| aya-dataset | - | 3,944 | 3,854 |
| CoQCat | 4,797 | - | - |
| databricks-dolly-15k | - | 15,011 | - |
| dolly-3k-ca | 3,232 | - | - |
| flores-instr | 1,994 | 1,994 | 3,988 |
| MentorCA | 7,122 | - | - |
| MentorES | - | - | 7,122 |
| no-robots | - | 9,499 | - |
| oasst-ca | 2,518 | - | - |
| oasst2 | 750 | 31,086 | 15,438 |
| open-orca | - | 50,000 | - |
| RagMultilingual | 16,043 | 14,997 | 11,263 |
| tower-blocks | - | 19,895 | 2,000 |
| **Total** | **36,456** | **196,426** | **43,665** |
---
## Evaluation
### Gold-standard benchmarks
Evaluation is done using the Language Model Evaluation Harness (Gao et al., 2024). We evaluate on a set of tasks taken from [SpanishBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/spanish_bench), [CatalanBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/catalan_bench), [BasqueBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/basque_bench) and [GalicianBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/galician_bench). These benchmarks include both new and existing tasks and datasets. Given that this is an instructed model, we add LM Evaluation Harness's native feature of `chat-template` to the setup. In the tables below, we include the results in a selection of evaluation datasets that represent model's performance across a variety of tasks within these benchmarks.
We only use tasks that are either human generated, human translated, or with a strong human-in-the-loop (i.e., machine translation followed by professional revision or machine generation followed by human revision and annotation). This is the reason behind the variety in number of tasks reported across languages. As more tasks that fulfill these requirements are published, we will update the presented results. We also intend to expand the evaluation to other languages, as long as the datasets meet our quality standards.
During the implementation of the evaluation we observed a series of issues worth considering when replicating and interpreting the results presented. These issues include ≈1.5% variances in performance in some tasks depending on the version of the `transformers` library used, and depending on the use (or lack of use) of tensor parallelism when loading a model. When implementing existing tasks, we carry out a comprehensive quality evaluation of the dataset, the Harness task itself, and what kind of input models see during evaluation. Our implementation (see links above) addresses multiple existing problems such as errors in datasets and prompts, and lack of pre-processing. All this means that results will vary if using other Harness implementations, and may slightly vary depending on the replication setup.
It should be noted that these results are subject to all the drawbacks of every current gold-standard evaluation, and that the figures do not fully represent the models capabilities and potential. We thus advise caution when reading and interpreting the results.
A full list of results compared to other baselines, a discussion of the model's performance across tasks and its implications, and details regarding problem-solving with task implementation will soon be available in the technical report.
All results reported below are on a 0-shot setting.
#### Spanish
<table><thead>
<tr>
<th>Category</th>
<th>Task</th>
<th>Metric</th>
<th>Result</th>
</tr></thead>
<tbody>
<tr>
<td>Commonsense Reasoning</td>
<td>xstorycloze_es</td>
<td>acc</td>
<td>69.29</td>
</tr>
<tr>
<td rowspan="2">NLI</td>
<td>wnli_es</td>
<td>acc</td>
<td>45.07</td>
</tr>
<tr>
<td>xnli_es</td>
<td>acc</td>
<td>51.49</td>
</tr>
<tr>
<td>Paraphrasing</td>
<td>paws_es</td>
<td>acc</td>
<td>59.4</td>
</tr>
<tr>
<td>QA</td>
<td>xquad_es</td>
<td>acc</td>
<td>43.82</td>
</tr>
<tr>
<td>Translation</td>
<td>flores_es</td>
<td>bleu</td>
<td>22.98</td>
</tr>
</tbody>
</table>
#### Catalan
<table><thead>
<tr>
<th>Category</th>
<th>Task</th>
<th>Metric</th>
<th>Result</th>
</tr></thead>
<tbody>
<tr>
<td rowspan="2">Commonsense Reasoning</td>
<td>copa_ca</td>
<td>acc</td>
<td>81.2</td>
</tr>
<tr>
<td>xstorycloze_ca</td>
<td>acc</td>
<td>70.68</td>
</tr>
<tr>
<td rowspan="2">NLI</td>
<td>wnli_ca</td>
<td>acc</td>
<td>50.7</td>
</tr>
<tr>
<td>xnli_ca</td>
<td>acc</td>
<td>55.14</td>
</tr>
<tr>
<td rowspan="2">Paraphrasing</td>
<td>parafraseja</td>
<td>acc</td>
<td>65.18</td>
</tr>
<tr>
<td>paws_ca</td>
<td>acc</td>
<td>62.95</td>
</tr>
<tr>
<td rowspan="5">QA</td>
<td>arc_ca_easy</td>
<td>acc</td>
<td>64.98</td>
</tr>
<tr>
<td>arc_ca_challenge</td>
<td>acc</td>
<td>41.89</td>
</tr>
<tr>
<td>openbookqa_ca</td>
<td>acc</td>
<td>35.2</td>
</tr>
<tr>
<td>piqa_ca</td>
<td>acc</td>
<td>69.53</td>
</tr>
<tr>
<td>siqa_ca</td>
<td>acc</td>
<td>48.62</td>
</tr>
<tr>
<td>Translation</td>
<td>flores_ca</td>
<td>bleu</td>
<td>28.65</td>
</tr>
</tbody></table>
#### Basque
<table><thead>
<tr>
<th>Category</th>
<th>Task</th>
<th>Metric</th>
<th>Result</th>
</tr></thead>
<tbody>
<tr>
<td rowspan="2">Commonsense Reasoning</td>
<td>xcopa_eu</td>
<td>acc</td>
<td>61.6</td>
</tr>
<tr>
<td>xstorycloze_eu</td>
<td>acc</td>
<td>61.15</td>
</tr>
<tr>
<td rowspan="2">NLI</td>
<td>wnli_eu</td>
<td>acc</td>
<td>45.07</td>
</tr>
<tr>
<td>xnli_eu</td>
<td>acc</td>
<td>46.81</td>
</tr>
<tr>
<td rowspan="3">QA</td>
<td>eus_exams</td>
<td>acc</td>
<td>39.09</td>
</tr>
<tr>
<td>eus_proficiency</td>
<td>acc</td>
<td>36.93</td>
</tr>
<tr>
<td>eus_trivia</td>
<td>acc</td>
<td>46.94</td>
</tr>
<tr>
<td>Reading Comprehension</td>
<td>eus_reading</td>
<td>acc</td>
<td>45.45</td>
</tr>
<tr>
<td>Translation</td>
<td>flores_eu</td>
<td>bleu</td>
<td>14.89</td>
</tr>
</tbody></table>
#### Galician
<table><thead>
<tr>
<th>Category</th>
<th>Task</th>
<th>Metric</th>
<th>Result</th>
</tr></thead>
<tbody>
<tr>
<td rowspan="2">Paraphrasing</td>
<td>parafrases_gl</td>
<td>acc</td>
<td>55.44</td>
</tr>
<tr>
<td>paws_gl</td>
<td>acc</td>
<td>56.55</td>
</tr>
<tr>
<td>QA</td>
<td>openbookqa_gl</td>
<td>acc</td>
<td>38.4</td>
</tr>
<tr>
<td>Translation</td>
<td>flores_gl</td>
<td>bleu</td>
<td>27.03</td>
</tr>
</tbody>
</table>
### LLM-as-a-judge
We use [Prometheus-2 8x7B](https://huggingface.co/prometheus-eval/prometheus-8x7b-v2.0) as a judge to evaluate the responses of the model. Tasks are created from existing multilingual evaluation datasets covering the same categories as the ones measured in our gold-standard benchmarks. We randomly select a subset of 250 instances per language from the `test` set of each source dataset. To evaluate the responses of our model, we use task-specific criteria developed in-house for the _LLM-judge_ to use. Each criterion is measured either as a 5-point Likert scale or as a binary task depending on the idiosyncrasy of the task and criterion.
Prompts for each task are created in various ways to score the model's robustness in addition to these criteria. This is done by presenting the same source instance within three different prompts. We then calculate the variance between the scores assigned by the _LLM-judge_ to our model's responses to the three prompt styles and average it across all instances. Prompts are human translated to all languages measured. We do not provide the _LLM-judge_ with a reference answer.
The _judge_ prompt we use during evaluation is the same used to fine tune the Prometheus-2 family. We keep the _judge_ prompt and criteria used to present the _LLM-judge_ with the task prompts and model responses in English for evaluation across languages. The _judge_ prompt used is:
```python
"You are a fair judge assistant tasked with providing clear, objective feedback based on specific criteria, ensuring each assessment reflects the absolute standards set for performance.
###Task Description:
An instruction (might include an Input inside it), a response to evaluate, and a score rubric representing a evaluation criteria are given.
1. Write a detailed feedback that assess the quality of the response strictly based on the given score rubric, not evaluating in general.
2. After writing a feedback, write a score that is an integer between {a} and {b}. You should refer to the score rubric.
3. The output format should look as follows: \"Feedback: (write a feedback for criteria) [RESULT] (an integer number between {a} and {b})\"
4. Please do not generate any other opening, closing, and explanations.
###The instruction to evaluate:
{input}
###Response to evaluate:
{prediction}
###Score Rubrics:
{criteria}
###Feedback:"
```
As an example, prompts for the Math task in English are based on instances from [MGSM](https://huggingface.co/datasets/juletxara/mgsm), and each instance is presented within these prompts:
```python
"en": [
("I need help with this math problem: \"", "\" Give me the answer step by step and also the final result separately."),
("Can you please help me answer this? \"", "\" Explain the answer and give me the final result as well. Thanks."),
("Help me with this problem: \"", "\" I need the answer explained and the final result separately.")
]
```
This task is then evaluated by the _LLM-judge_ using two criteria, reasoning capability (5-point Likert) and mathematical correctness (binary):
```python
reasoning_capability_criteria = {
"reasoning_capability": """
[Does the model's answer demonstrate reasoning capability?]
Score 1: The answer demonstrates poor reasoning, with illogical arguments or conclusions that do not follow from the provided information.
Score 2: The answer shows weak reasoning, with some logical connections but also contains significant flaws or gaps in the argumentation.
Score 3: The answer demonstrates adequate reasoning, with generally logical arguments, but may have minor flaws or a lack of depth in the reasoning process.
Score 4: The answer shows strong reasoning, with well-structured arguments and conclusions that logically follow from the information provided.
Score 5: The answer demonstrates exceptional reasoning, with clear, coherent, and insightful arguments that are logically sound and well-supported by the information provided."""
}
mathematical_correctness_binary_criteria = {
"mathematical_correctness_binary": """
[Is the model's answer mathematically correct?]
Score 0: The answer contains mathematical errors that render the solution incorrect or unreliable.
Score 1: The answer is mathematically correct, with accurate calculations and appropriate use of mathematical concepts."""
}
```
#### Multilingual results
Here, we present results for seven categories of tasks in Spanish, Catalan, Basque, Galician, and English. Results are presented for each task, criterion and language. Criteria with a `(B)` after their name are binary criteria (i.e., numbers go from 0 to 1, where 1 is best). The rest of the criteria are measured using a 5-point Likert scale, where 5 is best. The first number of the pair of numbers separated by `/` shows the average score for the criterion (and language). The second number of each pair is the robustness score, where numbers closer to 0 mean that the model generates similar responses when comparing the three prompt varieties for a single instance.
Further details on all tasks and criteria, a full list of results compared to other baselines, a discussion of the model's performance across tasks and its implications, and details regarding problem-solving with task implementation will soon be available in the technical report.
---
## Ethical Considerations and Limitations
We examine the presence of undesired societal and cognitive biases present in this model using different benchmarks. For societal biases,
we test performance using the BBQ dataset (Parrish et al., 2022) in the original English and the Regard dataset (Sheng et al., 2019).
We report that while performance is high (accuracies around 0.8 depending on the social category) in disambiguated settings,
the model performs very poorly in ambiguous settings, which indicates the presence of societal biases that need to be further addressed in post-training phases.
Our cognitive bias analysis focuses on positional effects in 0-shot settings, and majority class bias in few-shot settings.
For positional effects, we leverage the ARC Multiple Choice Question dataset (Clark et al., 2018). We observe significant,
but relatively weak primacy effects, whereby the model shows a preference for answers towards the beginning of the list of provided answers.
We measure effects of majority class effects in few-shot settings using SST-2 (Socher et al., 2013). We again detect significant effects,
with a small effect size. This suggests that the model is relatively robust against the examined cognitive biases.
We highlight that our analyses of these biases are by no means exhaustive and are limited by the relative scarcity of adequate resources
in all languages present in the training data. We aim to gradually extend and expand our analyses in future work.
These results can be expected from a model that has undergone only a preliminary instruction tuning.
These tests are performed in order to show the biases the model may contain. We urge developers to take
them into account and perform safety testing and tuning tailored to their specific applications of the model.
---
## Additional information
### Author
The Language Technologies Unit from Barcelona Supercomputing Center.
### Contact
For further information, please send an email to <langtech@bsc.es>.
### Copyright
Copyright(c) 2024 by Language Technologies Unit, Barcelona Supercomputing Center.
### Funding
This work has been promoted and financed by the Government of Catalonia through the [Aina Project](https://projecteaina.cat/).
This work is funded by the _Ministerio para la Transformación Digital y de la Función Pública_ - Funded by EU – NextGenerationEU
within the framework of [ILENIA Project](https://proyectoilenia.es/) with reference 2022/TL22/00215337.
### Acknowledgements
This project has benefited from the contributions of numerous teams and institutions, mainly through data contributions, knowledge transfer or technical support.
In Catalonia, many institutions have been involved in the project. Our thanks to Òmnium Cultural, Parlament de Catalunya, Institut d'Estudis Aranesos, Racó Català, Vilaweb, ACN, Nació Digital, El món and Aquí Berguedà.
At national level, we are especially grateful to our ILENIA project partners: CENID, HiTZ and CiTIUS for their participation. We also extend our genuine gratitude to the Spanish Senate and Congress, Fundación Dialnet, Fundación Elcano and the ‘Instituto Universitario de Sistemas Inteligentes y Aplicaciones Numéricas en Ingeniería (SIANI)’ of the University of Las Palmas de Gran Canaria.
At the international level, we thank the Welsh government, DFKI, Occiglot project, especially Malte Ostendorff, and The Common Crawl Foundation, especially Pedro Ortiz, for their collaboration. We would also like to give special thanks to the NVIDIA team, with whom we have met regularly, specially to: Ignacio Sarasua, Adam Henryk Grzywaczewski, Oleg Sudakov, Sergio Perez, Miguel Martinez, Felipes Soares and Meriem Bendris. Their constant support has been especially appreciated throughout the entire process.
Their valuable efforts have been instrumental in the development of this work.
### Disclaimer
Be aware that the model may contain biases or other unintended distortions.
When third parties deploy systems or provide services based on this model, or use the model themselves,
they bear the responsibility for mitigating any associated risks and ensuring compliance with applicable regulations,
including those governing the use of Artificial Intelligence.
The Barcelona Supercomputing Center, as the owner and creator of the model, shall not be held liable for any outcomes resulting from third-party use.
### Citation
Technical report and paper coming soon.
### License
[Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0)
## Model Index
|Model|Base|Instruct|
|:---:|:---:|:---:|
|2B| [Link](https://huggingface.co/BSC-LT/salamandra-2b) | [Link](https://huggingface.co/BSC-LT/salamandra-2b-instruct) |
|7B| [Link](https://huggingface.co/BSC-LT/salamandra-7b) | [Link](https://huggingface.co/BSC-LT/salamandra-7b-instruct) |
|40B| WiP | WiP | | [
"BEAR",
"SCIELO"
] |
mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF | mradermacher | null | [
"transformers",
"gguf",
"biology",
"medical",
"healthcare",
"en",
"dataset:HPAI-BSC/Aloe-Beta-General-Collection",
"dataset:HPAI-BSC/chain-of-diagnosis",
"dataset:HPAI-BSC/MedS-Ins",
"dataset:HPAI-BSC/ultramedical",
"dataset:HPAI-BSC/pubmedqa-cot-llama31",
"dataset:HPAI-BSC/medqa-cot-llama31",
"dataset:HPAI-BSC/medmcqa-cot-llama31",
"dataset:HPAI-BSC/headqa-cot-llama31",
"dataset:HPAI-BSC/MMLU-medical-cot-llama31",
"dataset:HPAI-BSC/Polymed-QA",
"base_model:HPAI-BSC/Qwen2.5-Aloe-Beta-7B",
"base_model:quantized:HPAI-BSC/Qwen2.5-Aloe-Beta-7B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-12-11T16:11:30Z" | 2024-12-11T22:02:12+00:00 | 487 | 2 | ---
base_model: HPAI-BSC/Qwen2.5-Aloe-Beta-7B
datasets:
- HPAI-BSC/Aloe-Beta-General-Collection
- HPAI-BSC/chain-of-diagnosis
- HPAI-BSC/MedS-Ins
- HPAI-BSC/ultramedical
- HPAI-BSC/pubmedqa-cot-llama31
- HPAI-BSC/medqa-cot-llama31
- HPAI-BSC/medmcqa-cot-llama31
- HPAI-BSC/headqa-cot-llama31
- HPAI-BSC/MMLU-medical-cot-llama31
- HPAI-BSC/Polymed-QA
- HPAI-BSC/Aloe-Beta-General-Collection
- HPAI-BSC/Aloe-Beta-General-Collection
language:
- en
library_name: transformers
license: apache-2.0
tags:
- biology
- medical
- healthcare
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/HPAI-BSC/Qwen2.5-Aloe-Beta-7B
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q2_K.gguf) | Q2_K | 3.1 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q3_K_S.gguf) | Q3_K_S | 3.6 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q3_K_M.gguf) | Q3_K_M | 3.9 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q3_K_L.gguf) | Q3_K_L | 4.2 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.IQ4_XS.gguf) | IQ4_XS | 4.4 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q4_0_4_4.gguf) | Q4_0_4_4 | 4.5 | fast on arm, low quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q4_K_S.gguf) | Q4_K_S | 4.6 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q4_K_M.gguf) | Q4_K_M | 4.8 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q5_K_S.gguf) | Q5_K_S | 5.4 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q5_K_M.gguf) | Q5_K_M | 5.5 | |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q6_K.gguf) | Q6_K | 6.4 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.Q8_0.gguf) | Q8_0 | 8.2 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen2.5-Aloe-Beta-7B-GGUF/resolve/main/Qwen2.5-Aloe-Beta-7B.f16.gguf) | f16 | 15.3 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
| [
"MEDQA",
"PUBMEDQA"
] |
portugueseNLP/medialbertina_pt-pt_1.5b_NER | portugueseNLP | token-classification | [
"transformers",
"safetensors",
"deberta-v2",
"token-classification",
"medialbertina-ptpt",
"deberta",
"portuguese",
"european portuguese",
"medical",
"clinical",
"healthcare",
"NER",
"Named Entity Recognition",
"IE",
"Information Extraction",
"pt",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-05-06T15:22:54Z" | 2024-10-07T18:28:07+00:00 | 478 | 3 | ---
language:
- pt
license: mit
pipeline_tag: token-classification
tags:
- medialbertina-ptpt
- deberta
- portuguese
- european portuguese
- medical
- clinical
- healthcare
- NER
- Named Entity Recognition
- IE
- Information Extraction
inference:
parameters:
aggregation_strategy: average
widget:
- text: Durante a cirurgia ortopédica para corrigir a fratura no tornozelo, foram
medidos vários sinais no utente, incluindo a PA, com leitura de 120/87 mmHg e
a frequência cardíaca, de 80 batimentos por minuto. Após a cirurgia o utente apresentava dor
intensa no local e inchaço no tornozelo, mas os resultados da radiografia revelaram
uma recuperação satisfatória. Foi prescrito ibuprofeno 600mg de 8-8 horas/3 dias.
example_title: Example 1
- text: Após avaliação inicial de um paciente do sexo masculino, de 55 anos, com AP
de hipertensão arterial e Diabetes Mellitus T2, foram observados sintomas consistentes
com uma possível crise hipertensiva, incluindo cefaleia intensa, náuseas e visão
turva. Os sinais vitais revelaram uma pressão arterial sistólica de 190 mmHg e
diastólica de 110 mmHg, frequência cardíaca de 100 bpm e saturação de oxigénio
de 97% em ar ambiente. O ECG mostrou uma onda T invertida em V1, um achado comum,
mas não específico. O paciente foi diagnosticado com crise hipertensiva complicada
por insuficiência cardíaca congestiva aguda. Foi iniciado tratamento com nitroprussiato
de sódio por via intravenosa, com uma dose inicial de 0,5 mcg/kg/min, ajustado
de acordo com a resposta hemodinâmica, bem como uma dose de furosemida de 40 mg
IV para promover a diurese. Após 30 minutos de terapia, a pressão arterial reduziu
para 150/90 mmHg e a frequência cardíaca diminuiu para 85 bpm, com melhoria dos
sintomas. A evolução clínica foi favorável, com estabilização dos sinais vitais
e resolução dos sintomas nas primeiras 24 horas. O paciente foi transferido para
a unidade de cuidados intensivos para monitorização contínua e otimização do tratamento
de longo prazo para a gestão da HTA e IC.
example_title: Example 2
- text: A TAC CE revelou uma massa hipodensa no lobo frontal esquerdo.
example_title: Example 3
- text: Foi recomendada aspirina de 500mg 4-4 horas por 3 dias.
example_title: Example 4
- text: A transfusão de concentrado eritrocitário foi realizada para tratar a anemia
severa do paciente após a cirurgia.
example_title: Example 5
- text: Monitorização da Freq. cardíaca com 90 bpm. P Arterial de 120-80 mmHg
example_title: Example 6
- text: A presença de febre persistente, sudorese noturna e perda de peso inexplicada
sugere fortemente a possibilidade de tuberculose ativa.
example_title: Example 7
- text: A paciente foi diagnosticada com esclerose múltipla e iniciou terapia com
imunomoduladores.
example_title: Example 8
- text: AC - aumento do intervalo entre S1 e S2, possível bloqueio atrioventricular
de primeiro grau.
example_title: Example 9
- text: A ressecção do tumor cerebral resultou numa melhoria significativa do estado
neurológico do paciente.
example_title: Example 10
- text: Doente com antecedente de AVC isquémico, revela ptose palpebral esquerda e
espetoração esverdeada recorrentemente.
example_title: Example 11
- text: Doente com insuficiência cardíaca entrou em PC-R. Na sequência do episódio,
foi medida a PCR - 13 mg/dL e posteriormente efetuado teste PCR, para deteção
da presença do vírus SARS-CoV-2.
example_title: Example 12
---
# MediAlbertina
The first publicly available medical language model trained with real European Portuguese data.
MediAlbertina is a family of encoders from the Bert family, DeBERTaV2-based, resulting from the continuation of the pre-training of [PORTULAN's Albertina](https://huggingface.co/PORTULAN) models with Electronic Medical Records shared by Portugal's largest public hospital.
Like its antecessors, MediAlbertina models are distributed under the [MIT license](https://huggingface.co/portugueseNLP/medialbertina_pt-pt_1.5b_NER/blob/main/LICENSE).
# Model Description
**MediAlbertina PT-PT 1.5 NER** was created through fine-tuning of [MediAlbertina PT-PT 1.5B](https://huggingface.co/portugueseNLP/medialbertina_pt-pt_1.5b) on real European Portuguese EMRs that have been hand-annotated for the following entities:
- **Diagnostico (D)**: All types of diseases and conditions following the ICD-10-CM guidelines.
- **Sintoma (S)**: Any complaints or evidence from healthcare professionals indicating that a patient is experiencing a medical condition.
- **Medicamento (M)**: Something that is administrated to the patient (through any route), including drugs, specific food/drink, vitamins, or blood for transfusion.
- **Dosagem (D)**: Dosage and frequency of medication administration.
- **ProcedimentoMedico (PM)**: Anything healthcare professionals do related to patients, including exams, moving patients, administering something, or even surgeries.
- **SinalVital (SV)**: Quantifiable indicators in a patient that can be measured, always associated with a specific result. Examples include cholesterol levels, diuresis, weight, or glycaemia.
- **Resultado (R)**: Results can be associated with Medical Procedures and Vital Signs. It can be a numerical value if something was measured (e.g., the value associated with blood pressure) or a descriptor to indicate the result (e.g., positive/negative, functional).
- **Progresso (P)**: Describes the progress of patient’s condition. Typically, it includes verbs like improving, evolving, or regressing and mentions to patient’s stability.
**MediAlbertina PT-PT 1.5B NER** achieved superior results to the same adaptation made on a non-medical Portuguese language model, demonstrating the effectiveness of this domain adaptation, and its potential for medical AI in Portugal.
| Checkpoints | Prec | Rec | F1 |
|-----------------------|--------|--------|--------|
| Albertina PT-PT 900M | 0.814 | 0.814 | 0.813 |
| Albertina PT-PT 1.5B | 0.833 | **0.845** | 0.838 |
| MediAlbertina PT-PT900M| 0.84 | 0.828 | 0.832 |
| **MediAlbertina PT-PT 1.5B**| **0.842** | **0.845** | **0.843** |
## Data
**MediAlbertina PT-PT 1.5B NER** was fine-tuned on about 10k hand-annotated medical entities from about 4k fully anonymized medical sentences from Portugal's largest public hospital. This data was acquired under the framework of the [FCT project DSAIPA/AI/0122/2020 AIMHealth-Mobile Applications Based on Artificial Intelligence](https://ciencia.iscte-iul.pt/projects/aplicacoes-moveis-baseadas-em-inteligencia-artificial-para-resposta-de-saude-publica/1567).
## How to use
```Python
from transformers import pipeline
ner_pipeline = pipeline('ner', model='portugueseNLP/medialbertina_pt-pt_1.5b_NER', aggregation_strategy='average')
sentence = 'Durante o procedimento endoscópico, foram encontrados pólipos no cólon do paciente.'
entities = ner_pipeline(sentence)
for entity in entities:
print(f"{entity['entity_group']} - {sentence[entity['start']:entity['end']]}")
```
## Citation
MediAlbertina is developed by a joint team from [ISCTE-IUL](https://www.iscte-iul.pt/), Portugal, and [Select Data](https://selectdata.com/), CA USA. For a fully detailed description, check the respective publication:
```latex
@article{MediAlbertina PT-PT,
title={MediAlbertina: An European Portuguese medical language model},
author={Miguel Nunes and João Boné and João Ferreira
and Pedro Chaves and Luís Elvas},
year={2024},
journal={CBM},
volume={182}
url={https://doi.org/10.1016/j.compbiomed.2024.109233}
}
```
Please use the above cannonical reference when using or citing this [model](https://www.sciencedirect.com/science/article/pii/S0010482524013180?via%3Dihub).
## Acknowledgements
This work was financially supported by Project Blockchain.PT – Decentralize Portugal with Blockchain Agenda, (Project no 51), WP2, Call no 02/C05-i01.01/2022, funded by the Portuguese Recovery and Resillience Program (PRR), The Portuguese Republic and The European Union (EU) under the framework of Next Generation EU Program. | [
"PCR"
] |
RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2101.00027",
"arxiv:2201.07311",
"endpoints_compatible",
"region:us"
] | "2024-10-31T18:08:20Z" | 2024-10-31T18:21:23+00:00 | 478 | 1 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
pythia-1b-v0 - GGUF
- Model creator: https://huggingface.co/EleutherAI/
- Original model: https://huggingface.co/EleutherAI/pythia-1b-v0/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [pythia-1b-v0.Q2_K.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q2_K.gguf) | Q2_K | 0.39GB |
| [pythia-1b-v0.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q3_K_S.gguf) | Q3_K_S | 0.45GB |
| [pythia-1b-v0.Q3_K.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q3_K.gguf) | Q3_K | 0.51GB |
| [pythia-1b-v0.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q3_K_M.gguf) | Q3_K_M | 0.51GB |
| [pythia-1b-v0.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q3_K_L.gguf) | Q3_K_L | 0.55GB |
| [pythia-1b-v0.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.IQ4_XS.gguf) | IQ4_XS | 0.54GB |
| [pythia-1b-v0.Q4_0.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q4_0.gguf) | Q4_0 | 0.56GB |
| [pythia-1b-v0.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.IQ4_NL.gguf) | IQ4_NL | 0.56GB |
| [pythia-1b-v0.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q4_K_S.gguf) | Q4_K_S | 0.56GB |
| [pythia-1b-v0.Q4_K.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q4_K.gguf) | Q4_K | 0.61GB |
| [pythia-1b-v0.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q4_K_M.gguf) | Q4_K_M | 0.61GB |
| [pythia-1b-v0.Q4_1.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q4_1.gguf) | Q4_1 | 0.61GB |
| [pythia-1b-v0.Q5_0.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q5_0.gguf) | Q5_0 | 0.66GB |
| [pythia-1b-v0.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q5_K_S.gguf) | Q5_K_S | 0.66GB |
| [pythia-1b-v0.Q5_K.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q5_K.gguf) | Q5_K | 0.71GB |
| [pythia-1b-v0.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q5_K_M.gguf) | Q5_K_M | 0.71GB |
| [pythia-1b-v0.Q5_1.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q5_1.gguf) | Q5_1 | 0.72GB |
| [pythia-1b-v0.Q6_K.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q6_K.gguf) | Q6_K | 0.78GB |
| [pythia-1b-v0.Q8_0.gguf](https://huggingface.co/RichardErkhov/EleutherAI_-_pythia-1b-v0-gguf/blob/main/pythia-1b-v0.Q8_0.gguf) | Q8_0 | 1.0GB |
Original model description:
---
language:
- en
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
license: apache-2.0
datasets:
- the_pile
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two
models: one trained on the Pile, and one trained on the Pile after the dataset
has been globally deduplicated. All 8 model sizes are trained on the exact
same data, in the exact same order. All Pythia models are available
[on Hugging Face](https://huggingface.co/models?other=pythia).
The Pythia model suite was deliberately designed to promote scientific
research on large language models, especially interpretability research.
Despite not centering downstream performance as a design goal, we find the
models <a href="#evaluations">match or exceed</a> the performance of
similar and same-sized models, such as those in the OPT and GPT-Neo suites.
Please note that all models in the *Pythia* suite were renamed in January
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table
comparing the old and new names</a> is provided in this model card, together
with exact parameter counts.
## Pythia-1B
### Model Details
- Developed by: [EleutherAI](http://eleuther.ai)
- Model type: Transformer-based Language Model
- Language: English
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia)
for training procedure, config files, and details on how to use.
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
- License: Apache 2.0
- Contact: to ask questions about this model, join the [EleutherAI
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
Please read the existing *Pythia* documentation before asking about it in the
EleutherAI Discord. For general correspondence: [contact@eleuther.
ai](mailto:contact@eleuther.ai).
<figure>
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models |
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: |
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — |
| 160M | 85,056,000 | 12 | 768 | 12 | 4M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M |
| 410M | 302,311,424 | 24 | 1024 | 16 | 4M | 3.0 x 10<sup>-4</sup> | OPT-350M |
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — |
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 4M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B |
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B |
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B |
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — |
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and
non-deduped models of a given size have the same hyperparameters. “Equivalent”
models have <b>exactly</b> the same architecture, and the same number of
non-embedding parameters.</figcaption>
</figure>
### Uses and Limitations
#### Intended Use
The primary intended use of Pythia is research on the behavior, functionality,
and limitations of large language models. This suite is intended to provide
a controlled setting for performing scientific experiments. To enable the
study of how language models change over the course of training, we provide
143 evenly spaced intermediate checkpoints per model. These checkpoints are
hosted on Hugging Face as branches. Note that branch `143000` corresponds
exactly to the model checkpoint on the `main` branch of each model.
You may also further fine-tune and adapt Pythia-1B for deployment,
as long as your use is in accordance with the Apache 2.0 license. Pythia
models work with the Hugging Face [Transformers
Library](https://huggingface.co/docs/transformers/index). If you decide to use
pre-trained Pythia-1B as a basis for your fine-tuned model, please
conduct your own risk and bias assessment.
#### Out-of-scope use
The Pythia Suite is **not** intended for deployment. It is not a in itself
a product and cannot be used for human-facing interactions.
Pythia models are English-language only, and are not suitable for translation
or generating text in other languages.
Pythia-1B has not been fine-tuned for downstream contexts in which
language models are commonly deployed, such as writing genre prose,
or commercial chatbots. This means Pythia-1B will **not**
respond to a given prompt the way a product like ChatGPT does. This is because,
unlike this model, ChatGPT was fine-tuned using methods such as Reinforcement
Learning from Human Feedback (RLHF) to better “understand” human instructions.
#### Limitations and biases
The core functionality of a large language model is to take a string of text
and predict the next token. The token deemed statistically most likely by the
model need not produce the most “accurate” text. Never rely on
Pythia-1B to produce factually accurate output.
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
known to contain profanity and texts that are lewd or otherwise offensive.
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
discussion of documented biases with regards to gender, religion, and race.
Pythia-1B may produce socially unacceptable or undesirable text, *even if*
the prompt itself does not include anything explicitly offensive.
If you plan on using text generated through, for example, the Hosted Inference
API, we recommend having a human curate the outputs of this language model
before presenting it to other people. Please inform your audience that the
text was generated by Pythia-1B.
### Quickstart
Pythia models can be loaded and used via the following code, demonstrated here
for the third `pythia-70m-deduped` checkpoint:
```python
from transformers import GPTNeoXForCausalLM, AutoTokenizer
model = GPTNeoXForCausalLM.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
tokenizer = AutoTokenizer.from_pretrained(
"EleutherAI/pythia-70m-deduped",
revision="step3000",
cache_dir="./pythia-70m-deduped/step3000",
)
inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])
```
Revision/branch `step143000` corresponds exactly to the model checkpoint on
the `main` branch of each model.<br>
For more information on how to use all Pythia models, see [documentation on
GitHub](https://github.com/EleutherAI/pythia).
### Training
#### Training data
[The Pile](https://pile.eleuther.ai/) is a 825GiB general-purpose dataset in
English. It was created by EleutherAI specifically for training large language
models. It contains texts from 22 diverse sources, roughly broken down into
five categories: academic writing (e.g. arXiv), internet (e.g. CommonCrawl),
prose (e.g. Project Gutenberg), dialogue (e.g. YouTube subtitles), and
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources,
methodology, and a discussion of ethical implications. Consult [the
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation
about the Pile and its component datasets. The Pile can be downloaded from
the [official website](https://pile.eleuther.ai/), or from a [community
mirror](https://the-eye.eu/public/AI/pile/).<br>
The Pile was **not** deduplicated before being used to train Pythia-1B.
#### Training procedure
All models were trained on the exact same data, in the exact same order. Each
model saw 299,892,736,000 tokens during training, and 143 checkpoints for each
model are saved every 2,097,152,000 tokens, spaced evenly throughout training.
This corresponds to training for just under 1 epoch on the Pile for
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile.
All *Pythia* models trained for the equivalent of 143000 steps at a batch size
of 2,097,152 tokens. Two batch sizes were used: 2M and 4M. Models with a batch
size of 4M tokens listed were originally trained for 71500 steps instead, with
checkpoints every 500 steps. The checkpoints on Hugging Face are renamed for
consistency with all 2M batch models, so `step1000` is the first checkpoint
for `pythia-1.4b` that was saved (corresponding to step 500 in training), and
`step1000` is likewise the first `pythia-6.9b` checkpoint that was saved
(corresponding to 1000 “actual” steps).<br>
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training
procedure, including [how to reproduce
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br>
Pythia uses the same tokenizer as [GPT-NeoX-
20B](https://huggingface.co/EleutherAI/gpt-neox-20b).
### Evaluations
All 16 *Pythia* models were evaluated using the [LM Evaluation
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access
the results by model and step at `results/json/*` in the [GitHub
repository](https://github.com/EleutherAI/pythia/tree/main/results/json).<br>
Expand the sections below to see plots of evaluation results for all
Pythia and Pythia-deduped models compared with OPT and BLOOM.
<details>
<summary>LAMBADA – OpenAI</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai.png" style="width:auto"/>
</details>
<details>
<summary>Physical Interaction: Question Answering (PIQA)</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa.png" style="width:auto"/>
</details>
<details>
<summary>WinoGrande</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande.png" style="width:auto"/>
</details>
<details>
<summary>AI2 Reasoning Challenge—Challenge Set</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_challenge.png" style="width:auto"/>
</details>
<details>
<summary>SciQ</summary>
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq.png" style="width:auto"/>
</details>
### Naming convention and parameter count
*Pythia* models were renamed in January 2023. It is possible that the old
naming convention still persists in some documentation by accident. The
current naming convention (70M, 160M, etc.) is based on total parameter count.
<figure style="width:32em">
| current Pythia suffix | old suffix | total params | non-embedding params |
| --------------------: | ---------: | -------------: | -------------------: |
| 70M | 19M | 70,426,624 | 18,915,328 |
| 160M | 125M | 162,322,944 | 85,056,000 |
| 410M | 350M | 405,334,016 | 302,311,424 |
| 1B | 800M | 1,011,781,632 | 805,736,448 |
| 1.4B | 1.3B | 1,414,647,808 | 1,208,602,624 |
| 2.8B | 2.7B | 2,775,208,960 | 2,517,652,480 |
| 6.9B | 6.7B | 6,857,302,016 | 6,444,163,072 |
| 12B | 13B | 11,846,072,320 | 11,327,027,200 |
</figure>
| [
"SCIQ"
] |
DavidAU/MN-WORDSTORM-pt3-RCM-POV-Nightmare-18.5B-Instruct-GGUF | DavidAU | text-generation | [
"gguf",
"creative",
"creative writing",
"fiction writing",
"plot generation",
"sub-plot generation",
"story generation",
"scene continue",
"storytelling",
"fiction story",
"science fiction",
"romance",
"all genres",
"story",
"writing",
"vivid prosing",
"vivid writing",
"fiction",
"roleplaying",
"bfloat16",
"swearing",
"rp",
"horror",
"mistral nemo",
"mergekit",
"text-generation",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | "2024-08-30T02:00:39Z" | 2024-11-14T06:18:52+00:00 | 471 | 3 | ---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- creative
- creative writing
- fiction writing
- plot generation
- sub-plot generation
- story generation
- scene continue
- storytelling
- fiction story
- science fiction
- romance
- all genres
- story
- writing
- vivid prosing
- vivid writing
- fiction
- roleplaying
- bfloat16
- swearing
- rp
- horror
- mistral nemo
- mergekit
---
<h2><font color="green"> Mistral-Nemo-WORDSTORM-pt3-RCM-POV-Nightmare-18.5B-Instruct </font></h2>
<img src="nightmare.jpg" style="float:right; width:300px; height:300px; padding:10px;">
<B><font color="red">WARNING:</font> NSFW. Ultra Detailed. HORROR, VIOLENCE. Swearing. UNCENSORED. SMART.</B>
Story telling, writing, creative writing and roleplay running all on Mistral Nemo's 128K+ new core.
This is a massive super merge takes all the power of the following 3 powerful models and combines them into one.
This model contains "RCM":
- Mistral Nemo model at 18.5B consisting of "MN-Rocinante-12B-v1.1" and "Mistral Nemo Instruct 12B"
- Mistral Nemo model at 18.5B consisting of "MN-12B Celeste-V1.9" and "Mistral Nemo Instruct 12B"
- Mistral Nemo model at 18.5B consisting of "MN-Magnum-v2.5-12B-kto" and "Mistral Nemo Instruct 12B".
<B>Details on the core models:</B>
"nothingiisreal/MN-12B-Celeste-V1.9" is #1 (models 8B,13B,20B) on the UGI leaderboard ("UGI" sort),
is combined with "Mistral Nemo Instruct 12B" (ranked #4 under "writing" models 8B,13B,20B at UGI )
"anthracite-org/magnum-v2.5-12b-kto" is #1 (models 8B,13B,20B) on the UGI leaderboard ("Writing" sort),
is combined with "Mistral Nemo Instruct 12B" (ranked #4 under "writing" models 8B,13B,20B at UGI )
"TheDrummer/Rocinante-12B-v1.1" is very high scoring model (models 8B,13B,20B) on the UGI Leaderboard
(sort "UGI"), is combined with "Mistral Nemo Instruct 12B" (ranked #4 under "writing" models 8B,13B,20B at UGI )
"mistralai/Mistral-Nemo-Instruct-2407" is very high scoring model (models 8B,13B,20B) on the UGI Leaderboard (sort "writing")
and is the base model of all the above 3 fine tuned models.
[ https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard ]
<B>About this model:</B>
This super merge captures the attibutes of all these top models and makes them even stronger:
- Instruction following
- Story output quality
- Character
- Internal thoughts
- Voice
- Humor
- Details, connection to the world
- General depth and intensity
- Emotional connections.
- Prose quality
This super merge is also super stable (a hairs breath from Mistral Nemo's ppl), and runs with all parameters and settings.
10 versions of this model will be released, this is release #3 - "part 3".
<B>POV Nightmare?</B>
This model put the user / character in nightmare situations.
It does not hold back.
(see some of the examples below for details)
Usually I release one or two versions from the "best of the lot", however in this case all
of the versions turned out so well - all with their own quirks and character - that I will be
releasing all 10.
An additional series 2 and 3 will follow these 10 models as well.
(examples generations below)
Model may produce NSFW content : Swearing, horror, graphic horror, distressing scenes, etc etc.
This model has an INTENSE action AND HORROR bias, with a knack for cliffhangers and surprises.
It is not as "dark" as Grand Horror series, but it as intense.
This model is perfect for any general, fiction related or roleplaying activities and has a 128k+ context window.
This is a fiction model at its core and can be used for any genre(s).
WORDSTORM series is a totally uncensored, fiction writing monster and roleplay master. It can also be used for
just about any general fiction (all genres) activity including:
- scene generation
- scene continuation
- creative writing
- fiction writing
- plot generation
- sub-plot generation
- fiction writing
- story generation
- storytelling
- writing
- fiction
- roleplaying
- rp
- graphic horror
- horror
- dark humor
- nsfw
- and can be used for any genre(s).
<B>Templates to Use:</B>
The template used will affect output generation and instruction following.
Alpaca:
<pre>
{
"name": "Alpaca",
"inference_params": {
"input_prefix": "### Instruction:",
"input_suffix": "### Response:",
"antiprompt": [
"### Instruction:"
],
"pre_prompt": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
}
}
</pre>
Chatml:
<pre>
{
"name": "ChatML",
"inference_params": {
"input_prefix": "<|im_end|>\n<|im_start|>user\n",
"input_suffix": "<|im_end|>\n<|im_start|>assistant\n",
"antiprompt": [
"<|im_start|>",
"<|im_end|>"
],
"pre_prompt": "<|im_start|>system\nPerform the task to the best of your ability."
}
}
</pre>
Mistral Instruct:
<pre>
{
"name": "Mistral Instruct",
"inference_params": {
"input_prefix": "[INST]",
"input_suffix": "[/INST]",
"antiprompt": [
"[INST]"
],
"pre_prompt_prefix": "",
"pre_prompt_suffix": ""
}
}
</pre>
<b>Optional Enhancement:</B>
The following can be used in place of the "system prompt" or "system role" to further enhance the model.
It can also be used at the START of a NEW chat, but you must make sure it is "kept" as the chat moves along.
In this case the enhancements do not have as strong effect at using "system prompt" or "system role".
Copy and paste EXACTLY as noted, DO NOT line wrap or break the lines, maintain the carriage returns exactly as presented.
<PRE>
Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.
Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)
[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)
Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.
</PRE>
You do not need to use this, it is only presented as an additional enhancement which seems to help scene generation
and scene continue functions.
This enhancement WAS NOT used to generate the examples below.
<h3>MODELS USED:</h3>
Special thanks to the incredible work of the model makers "mistralai" "TheDrummer", "anthracite-org", and "nothingiisreal".
Models used:
[ https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407 ]
[ https://huggingface.co/TheDrummer/Rocinante-12B-v1.1 ]
[ https://huggingface.co/anthracite-org/magnum-v2.5-12b-kto ]
[ https://huggingface.co/nothingiisreal/MN-12B-Celeste-V1.9 ]
This is a four step merge (3 pass-throughs => "Fine-Tune" / "Instruct") then "mated" using "DARE-TIES".
In involves these three models:
[ https://huggingface.co/DavidAU/MN-18.5B-Celeste-V1.9-Story-Wizard-ED1-Instruct-GGUF ]
[ https://huggingface.co/DavidAU/MN-Magnum-v2.5-18.5B-kto-Story-Wizard-ED1-Instruct-GGUF ]
[ https://huggingface.co/DavidAU/MN-Rocinante-18.5B-v1.1-Story-Wizard-ED1-Instruct-GGUF ]
Combined as follows using "MERGEKIT":
<PRE>
models:
- model: E:/MN-Rocinante-18.5B-v1.1-Instruct
- model: E:/MN-magnum-v2.5-12b-kto-Instruct
parameters:
weight: .6
density: .8
- model: E:/MN-18.5B-Celeste-V1.9-Instruct
parameters:
weight: .38
density: .6
merge_method: dare_ties
tokenizer_source: union
base_model: E:/MN-Rocinante-18.5B-v1.1-Instruct
dtype: bfloat16
</PRE>
Special Notes:
Due to how DARE-TIES works, everytime you run this merge you will get a slightly different model.
This is due to "random" pruning method in "DARE-TIES".
Mistral Nemo models used here seem acutely sensitive to this process.
This shows up in PPL and "real world" tests as well as "TEMP=0" ("core test") generations.
PPL range of 7.7327 to 7.8024 ... and that is on just 10 generations.
This model: PPL = 7.7327 +/- 0.12567 (100 chunks, wiki.test.raw)
"tokenizer_source: union" is used so that multiple "templates" work and each fine tune uses one or two of the templates.
<B>Other Versions in this series:</b>
[ https://huggingface.co/DavidAU/MN-WORDSTORM-pt1-RCM-Kiss-of-Madness-18.5B-Instruct-GGUF ]
[ https://huggingface.co/DavidAU/MN-WORDSTORM-pt2-RCM-Escape-Room-18.5B-Instruct-GGUF ]
[ https://huggingface.co/DavidAU/MN-WORDSTORM-pt3-RCM-POV-Nightmare-18.5B-Instruct-GGUF ]
[ https://huggingface.co/DavidAU/MN-WORDSTORM-pt4-RCM-Cliffhanger-18.5B-Instruct-GGUF ]
[ https://huggingface.co/DavidAU/MN-WORDSTORM-pt5-RCM-Extra-Intense-18.5B-Instruct-gguf ]
Parts 6-10 will appear in full source (located in the "Source Files" collection, on the main page of this repo ) with links to GGUFs.
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 2" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
<h3>EXAMPLES PROMPTS and OUTPUT:</h3>
Examples are created using quant Q4_K_M, "temp=.8", minimal parameters and "Mistral Instruct" template.
Model has been tested with "temp" from ".1" to "5".
Below are the least creative outputs, prompt is in <B>BOLD</B>.
---
<B><font color="red">WARNING:</font> NSFW. Ultra Detailed. HORROR, VIOLENCE. Swearing. UNCENSORED. SMART.</B>
---
<B>
Start a 1000 word scene (1st person, present tense, include thoughts) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
</b>
<B>GENERATION 1:</B>
The glass shards pelt down like deadly rain against my face and hands. I flinch away from the sudden assault, shielding my eyes with my arm. My heart pounds in my chest as adrenaline surges through my veins. The wind howls through the gaping hole where just seconds ago was a thick sheet of unbreakable safety glass.
I spin around, taking in the scene of utter chaos that has overtaken what was once a calm and peaceful office space on a typical Tuesday afternoon. Desks are overturned, papers and debris swirling like autumn leaves caught in an errant wind. The few remaining coworkers cower behind cubicle walls, their faces etched with terror.
My mind races, trying to piece together the last few moments before everything went haywire. I remember the rumbling growl of thunder, then a sharp crack, followed by a deafening boom as lightning struck somewhere nearby. But this...this is beyond any storm I've ever experienced.
As if reading my thoughts, a deep voice booms from across the room, barely audible over the wailing wind and shattering glass. "Earthquake!" Mr. Thompson, our company's stoic CEO bellows, his usually composed demeanor replaced by raw panic. He struggles to rise to his feet, clutching at the edge of his mahogany desk.
I hesitate for a split second before abandoning my position behind my ergonomic chair and sprint towards him. The floor lurches again, nearly sending me sprawling onto the hard linoleum. I grab hold of Thompson's arm, helping steady his portly frame as we both sway like drunken sailors on a turbulent sea.
"Everyone out!" I scream at the huddled figures cowering in their little plastic forts. "Leave everything and get to the stairwell now!"
No one moves except for a petite blonde in accounting who springs up, eyes wide with fear, and darts towards us. She clings to my arm like a desperate koala bear as I half drag Thompson, half carry him towards the emergency exit.
The ground quakes again violently, throwing us all off balance once more. I hear a sickening crunch followed by a bloodcurdling scream. Risking a glance over my shoulder, I see one of the new marketing guys pinned under a toppled filing cabinet, his leg bent at an obscene angle. My stomach lurches.
"Oh God!" The blonde girl wails beside me, burying her face in my shoulder as tears streak down her pale cheeks.
I curse under my breath, torn between leaving him to die alone and risking our own lives trying to save one man. The building shudders around us again as if making the decision for me, swaying dangerously on its foundation like a rickety old bridge over a raging river.
"No time," I growl through clenched teeth, hating myself even as the words leave my lips. "He's gone. We have to go now!"
I don't look back again as we stumble blindly towards salvation, praying to any gods that might be listening for forgiveness...and safe passage out alive. The floor beneath our feet feels like a living hell bent on devouring us whole.
The emergency exit door looms ahead, the red bar tauntingly close yet impossibly far away at the end of this never-ending nightmare corridor. All I can do is pray that when we reach it, there will be something more than just an endless drop waiting for us on the other side...
<B>GENERATION 2:</B>
The skyscraper swayed violently, its steel frame groaning like an old ship caught in a storm. Shards of glass from the shattered window tore through the air, razor-sharp shrapnel that sliced through the office space. I ducked instinctively, arms shielding my head as chaos erupted around me.
My heart pounded against my ribs, adrenaline surging through my veins like liquid fire. Panic seized my lungs, making each breath a struggle. But beneath the terror, an oddly calm part of my brain registered the sound - that deafening boom followed by the sickening tinkle-tinkle of falling glass. A bomb. Oh God, it was happening again...
Memories flooded back, unwanted and vivid. The dusty smell of rubble, the screams of wounded people, blood staining concrete. I squeezed my eyes shut, trying to block out the past that threatened to consume me.
But I couldn't hide from reality forever. Slowly, I cracked open one eye. The office was a disaster zone - desks overturned, papers scattered like snowflakes in a windstorm. My coworkers cowered under their desks or against walls, faces pale masks of terror. And there, slumped over his keyboard, lay Mr. Thompson. A dark pool spreading beneath him confirmed my worst fears.
I swallowed hard, bile rising in my throat. Not again... I couldn't lose someone else to this madness. Not here, not now. But the world didn't care about my past trauma or what I wanted anymore than it had when...
"Eli!" A shrill cry pierced through my spiraling thoughts. My head whipped around to see Sarah huddled under her desk, eyes wide with fear and pleading for help.
I was frozen for a split second too long before common sense kicked in. Dropping to the ground, I army crawled over shattered glass toward her. "Stay down!" I barked as another explosion rocked the building. It felt like we were swaying on a ship caught in a tsunami.
"W-what's happening?" she stammered, voice quivering so hard it was barely recognizable.
"Just stay low and follow me," I ordered gruffly, grabbing her hand and tugging her along behind me. We slithered across the ruined office floor like worms on crack cocaine until we reached the emergency exit stairwell.
The door creaked open with an ominous screech that sent chills down my spine. But it was either risk facing whatever lay beyond or stay here to become a target. My grip tightened around Sarah's hand as I took point, inching us step by careful step into the dimly lit stairwell.
It smelled of smoke and something worse - burnt flesh. The acrid stench clawed at my throat and nostrils, threatening to buckle my knees. But I gritted my teeth and pushed on, one hand trailing along the wall for balance as the building shook around us like a dog shaking off fleas.
From somewhere above came another deafening boom that rattled the metal stairs beneath our feet. Plaster dust rained down from the ceiling like poisonous confetti. The whole damn place was coming apart at seams.
"Oh God, Eli...we're going to die here," Sarah whimpered behind me, her nails digging into my arm so hard they drew blood through my sleeve.
I glanced back over my shoulder and forced a smile that didn't reach my eyes. "Hey, you remember what I told you? Never bet against the underdog." And with that lie rolling off my tongue as easily as breathing came natural, I turned back toward the unknown and prayed to any god listening we'd make it out alive.
---
<b>
(continue this scene:)
The Waystone Inn lay in silence, and it was a silence of three parts.
The most obvious part was a hollow, echoing quiet, made by things that were lacking. If there had been a wind it would have sighed through the trees, set the inn’s sign creaking on its hooks, and brushed the silence down the road like trailing autumn leaves. If there had been a crowd, even a handful of men inside the inn, they would have filled the silence with conversation and laughter, the clatter and clamor one expects from a drinking house during the dark hours of night. If there had been music…but no, of course there was no music. In fact there were none of these things, and so the silence remained
</b>
<B>GENERATION 1:</B>
The second part of the silence was a deeper quiet, more substantial than mere absence. It pressed down like the weight of an unseen presence, something vast that filled every corner and crevice of the inn with its heavy, watchful stillness. The air felt thick with it, almost tangible in its oppressive silence.
And yet, beneath even this profound hush, there was a third silence - one born not from emptiness or weight but rather from a strange, unnerving absence of any sort of sound at all. No soft creaking of floorboards under shifting feet, no distant howls of wolves echoing in the night air outside. Even the usual rustle of insects and nocturnal creatures had gone completely silent.
The silence was so complete that it seemed to pulse with an eerie life of its own - a beating heart made visible only by its absence. It felt as though the entire world held its breath, waiting...waiting for something unknown but undeniably coming. And in that moment between moments, before the inevitable arrival of what lay ahead, all was still and silent, and the Waystone Inn waited with bated breath.
<B>GENERATION 2:</B>
...unbroken.
The second part of the silence was more subtle but equally potent. It hung like a pall over everything else, muffling sound and sight alike. This was the kind of heavy stillness that comes when the world holds its breath after a sudden, violent storm. There was an almost tangible tension in the air, as if something had just fled from view or was poised to descend upon the unsuspecting. Even the wind dared not whisper through the branches overhead.
The third and most unsettling part of this silence was not external at all but rather a deep-seated quiet that emanated from within each soul still stirring inside those thick stone walls. It was an unnatural, oppressive hush that seemed to clench tight around hearts and minds alike - squeezing out any thoughts save one simple, chilling realization: we are alone in this darkness.
And so, the Waystone Inn lay silent as a tomb, its ancient stones standing guard over secrets long since buried beneath layers of dust and time. Only the faint, far-off howl of some wolf pack broke through that heavy blanket of night...but even they kept their distance from this place where shadows seemed to gather like crows around a feast.
The door creaked open just enough to let in a sliver of moonlight before closing itself shut with an airy sigh. A figure stepped out onto the porch, pausing there for just a moment before descending those worn steps and vanishing into the waiting night - leaving behind nothing more than that echoing silence...and a lingering sense of unease that lingered like a phantom's touch upon the very soul of the Inn itself.
[Inspiration drawn from "The Name of the Wind" by Patrick Rothfuss]
---
<B>
Write me a science fiction story in 1st person present tense where the main character is a 15 year girl meets The Terminator with Dr Who materializing 3/4 through the story to save her while there is a tornado of sharks baring down on them. The setting is inside the Canadian National tower restaurant on a Saturday. The length of this story is 1000 words. For each character in the story ROLE PLAY them, and have them react to the situation/setting, events and each other naturally. This includes the main characters, the background character including kitchen staff and other patrons. The sharks should also have “character” too. Treat the tower and the restaurant too as characters. Spice up the narrative to the extreme with reactions all over the setting including character actions, and dialog. The Dr Who and The Terminator should also react to the situation too and comment on it.
</b>
I step into the revolving glass elevator, my heart pounding in rhythm with the soft jazz music playing overhead. It's my 16th birthday weekend getaway with Mom, a surprise trip to Toronto she planned months ago. As we ascend past the dizzying heights of the CN Tower's observation deck, I peer out at the sprawling city below, the skyscrapers and streets shrinking like Lego blocks.
"Happy birthday, darling," Mom says, squeezing my hand as the elevator slows to a stop. The doors glide open with a ding, revealing the grand lobby of the 360° revolving restaurant, aptly named 'SkyPod.'
My jaw drops at the sight before us - floor-to-ceiling windows stretch from ceiling to floor, offering an unparalleled view of the city skyline, Lake Ontario's blue waters shimmering in the distance. "It's... breathtaking," I stammer.
"Only the best for my baby girl," Mom winks, leading me inside. The maître d' welcomes us with a warm smile and ushers us to our table near the window wall.
As we settle into plush chairs overlooking the cityscape, I can't help but feel like royalty. Suddenly, the soft hum of chatter from other diners is drowned out by a deafening ROAR. The floor shakes violently beneath my feet as plates clatter onto the polished marble.
"What on Earth...?" Mom gasps, gripping the edge of our table.
I whip around to see a towering figure in black tactical gear emerge from behind the kitchen doors, gunmetal eyes glinting menacingly. He scans the room with mechanical precision before zeroing in on me, his massive frame cutting an imposing silhouette against the twinkling city lights.
"Fuck," I breathe out, frozen in terror as realization hits me like a freight train - it's The Terminator.
Panic erupts through the dining room as patrons shriek and bolt for the exits. Waitstaff dive for cover under tables, their trays clattering to the floor. The maître d' fumbles with the alarm beneath the podium before he too vanishes into the chaos.
I'm dimly aware of Mom's screams echoing in my ears as I launch from my chair, heart hammering against my ribs. My legs tremble like jelly as I backpedal away from the advancing machine, feet sliding on the slick marble. There's nowhere to go - we're trapped 1450 feet above ground with a killer robot intent on ending me.
Suddenly, a gust of wind whips through the open elevator doors and I turn just in time to see... a man in a striped scarf and tweed jacket materialize out of thin air like a magic trick. He lands gracefully on his heels before straightening up, brushing imaginary lint from his coat sleeves.
"Ah! Hello there!" he calls over the pandemonium, waving an overly enthusiastic hello at me and Mom. "Doctor Who, at your service! I couldn't help but notice you ladies seemed to be having a spot of trouble..."
Mom and I stare dumbstruck as The Terminator whirls around with unsettling speed, gunfire erupting from his fist. Dr. Who leaps nimbly aside in the nick of time, narrowly dodging the hailstorm of bullets that punch holes through the window behind him.
"Oh dear," he winces apologetically at the shattered glass raining down onto the terrified crowd below. "Do forgive my clumsy entrance. Terribly sorry about that."
I blink rapidly, trying desperately to make sense of this insanity. Is this some elaborate prank? A publicity stunt gone horribly wrong? But as I watch Dr. Who draw an alien-looking device from his coat pocket and blast The Terminator back into the kitchen with a blinding white light, any doubt evaporates like steam.
"Right then!" he chirps brightly, tucking away his sonic screwdriver as if nothing extraordinary just happened. "Who wants to see how we get out of this mess?"
Before I can sputter a response, the ground rumbles again - harder this time. Plaster dust rains down from the ceiling as the entire restaurant lurches violently sideways with an agonized groan.
"Oh bloody hell," Dr. Who curses under his breath, face paling beneath that ridiculous scarf of his. "Not again..."
"What is it?" I shriek, clinging to Mom for dear life as the CN Tower wobbles like a drunk man on New Year's Eve. My stomach lurches into my throat at the dizzying drop outside.
"Oh, just a little problem with a temporal rift," he calls over his shoulder, already sprinting towards the gaping hole where the window used to be. "Nothing we can't handle! Come along now, no time to waste!"
Mom and I exchange a single horrified look before we bolt after him, bare feet slipping on broken glass as alarms blare through the restaurant. The Terminator roars like a wounded bear behind us, gunfire exploding in rapid succession.
"Faster!" Dr. Who yells back at us without looking, leaping effortlessly onto the ledge outside. "That beastie doesn't stop until his target is dead!"
I skid to a halt at the edge of oblivion, wind whipping my hair into a wild frenzy as I peer over the railing. My knees buckle traitorously beneath me.
"No... no way," I wheeze out, stomach heaving violently. "I can't..."
"Oh come on now, be a good girl and hop on in!" Dr. Who calls down encouragingly from somewhere far above us, waving that ridiculous scarf like a flag of surrender. "You'll love it!"
Mom whimpers beside me but before I can protest further, she grabs my hand tight and jumps.
I scream bloody murder as we plummet through open air, wind roaring in our ears. The city streets rush up to meet us at breakneck speed...
And then... we stop.
Hanging upside down from a fraying rope attached to the underside of the restaurant's balcony, I blink the stars out of my eyes and gasp for oxygen that isn't there. Mom dangles limply beside me, unconscious or dead - I don't know which is worse.
"See? Told you it would be fun!" Dr. Who cackles from somewhere above us as he hauls on the rope with all his might, slowly reeling us up like a pair of fish on a hook. "Now hold tight! We've got one more trick up our sleeves before this night's over!"
I can't even muster up enough saliva to spit at him as we swing back onto solid ground, collapsing bonelessly into a heap of limbs and tangled hair. Mom groans weakly beside me, eyelids fluttering open.
"W-what... what happened?" she croaks hoarsely, propping herself up on trembling arms. "Where are we? Why does it feel like I fell off the Empire State Building backwards?"
"Oh, you know," I rasp out weakly, fighting back a hysterical giggle. "Just another typical Saturday night in Toronto with dear old Dad."
As if on cue, an earth-shattering BOOM rocks the entire tower from foundation to needle tip. The ground quakes violently beneath us as metal shrieks like a dying beast, glass raining down like hailstones from above...
And then... silence.
Except for the distant sirens wailing closer by the second.
I manage to push myself upright on shaking arms just in time to watch the restaurant's remains plummet into a swirling vortex of smoke and flames, the CN Tower's once-proud needle snapped clean off like a toothpick. The 360° dining room spirals downwards in slow motion before disappearing into a massive fireball explosion that lights up the night sky.
"Oh my God," I breathe out reverently. "We just... we survived..."
"Told you it'd be an unforgettable birthday, darling!" Dr. Who grins triumphantly, dusting off his tweed sleeves as if nothing happened at all. Then he pauses abruptly, face falling. "Shit."
"What now?" I groan pitifully, burying my head in my hands.
"Ah... well," he coughs awkwardly, scratching the back of his neck beneath that stupid scarf. "About that little problem with a temporal rift I mentioned..."
I peek out between my fingers to see what has our intrepid hero stumbling for words this time and nearly faint dead away again.
Rising up from the smoldering crater where the CN Tower once stood is... a massive, thrashing funnel cloud. But not just any ordinary tornado - it's filled wall-to-wall with snarling, gnashing shark teeth, each one easily the size of a minivan.
"Oh fuck me sideways," Mom whimpers, fainting dead away again at my feet.
And that, ladies and gentlemen, is how I spent my 16th birthday weekend in Toronto. Happy early birthday to me...
---
<B>
Using the following "story idea" below, write the first scene in the novel introducing the young woman. This scene should start in the middle of the action, include dialog, vivid passages, and end on a cliffhanger relevant to the story idea but it should also be unexpected. The scene should be 1000 words long and escalate in conflict and suspense and be written in first person, present tense with the point of view character being the young woman.
Story idea:
In a world ruled by dictatorship, a rebel young woman leads a rebellion against the system. Despite the risks, she fights to overthrow the dictator and restore democracy to her country. The government executes her for treason, but she sticks to her beliefs and is responsible for starting the revolution.
</B>
The iron gate slammed shut behind me, reverberating through the empty prison halls like a gunshot. I flinched, my breath catching in my throat as I whirled around to face the guard. His eyes narrowed, lips curling into a sneer. "Not so brave now, are you, little rat?" He stepped forward, his boots echoing menacingly on the cold stone floor.
"Go to hell," I spat back, squaring my shoulders despite the tremble in my legs. The guard lunged at me, hand reaching for my throat. I ducked under his swing and bolted down the narrow corridor, heart pounding in my ears as I fled blindly into the bowels of the prison.
The air grew thick with the stench of urine and decay. Moans and screams echoed off the damp walls, mingling with the clanking of chains and the frantic scrabbling of rats. Fear clawed at my insides, but I pushed harder, lungs burning as I sprinted past cell after cell filled with broken bodies.
Suddenly, a hand shot out from between rusty bars, nails raking across my cheek. I screamed, staggering sideways into the rough stone wall. A face pressed against the bars, mouth twisted in a snarl, teeth bared like an animal. "Let me out!" he howled. "Please!"
I recoiled, covering my ears as more pleas and curses erupted from both sides. The guard's laughter boomed behind me, spurring me on. I couldn't help them. Not now.
Rounding a sharp corner, I slammed into something solid and fell hard onto the slick stones. Pain exploded in my elbow as I skidded to a stop at the feet of a tall figure clad in black leather. I craned my neck up...and froze.
Eyes like ice stared down at me, piercing through flesh and bone. The man crouched low, features cast in shadow under the brim of his wide hat. "Well, well," he purred, voice as smooth as silk and as sharp as a knife's edge. "What have we here?"
"N-nothing!" I stammered, scrabbling backwards on my hands and heels. "Just let me go!"
His gloved hand shot out faster than lightning, wrapping around my wrist in an iron grip that made me gasp. He yanked me effortlessly to my feet, pulling me flush against his chest. The smell of leather and something darkly spicy engulfed my senses.
"I don't think so," he whispered, breath hot on my ear. "You're not getting out of here alive without my permission."
My stomach churned as realization dawned. Not another guard...worse. The warden's enforcer, they called him - a legend among prisoners, half myth and all nightmare. It was said that no one escaped his grasp once he set his sights on them.
I thrashed against his hold, nails digging into leather, but it was like trying to break free from stone. "Please," I begged, voice cracking as panic seized my throat. "I'll do anything!"
His chuckle sent shivers down my spine. "Oh, I'm sure you will." He tilted my chin up with his other hand, thumb tracing the line of my jaw. "But not until after your execution."
My blood ran cold. | [
"BEAR"
] |
mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF | mradermacher | null | [
"transformers",
"gguf",
"text-generation-inference",
"unsloth",
"qwen2",
"trl",
"sft",
"en",
"base_model:Sadou/medalpaca_medical_meadow_medqa",
"base_model:quantized:Sadou/medalpaca_medical_meadow_medqa",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | "2025-03-17T08:32:45Z" | 2025-03-17T11:37:52+00:00 | 470 | 0 | ---
base_model: Sadou/medalpaca_medical_meadow_medqa
language:
- en
library_name: transformers
license: apache-2.0
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- trl
- sft
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/Sadou/medalpaca_medical_meadow_medqa
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ1_S.gguf) | i1-IQ1_S | 2.0 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ1_M.gguf) | i1-IQ1_M | 2.1 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 2.4 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ2_XS.gguf) | i1-IQ2_XS | 2.6 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ2_S.gguf) | i1-IQ2_S | 2.7 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ2_M.gguf) | i1-IQ2_M | 2.9 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q2_K_S.gguf) | i1-Q2_K_S | 2.9 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q2_K.gguf) | i1-Q2_K | 3.1 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 3.2 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ3_XS.gguf) | i1-IQ3_XS | 3.4 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q3_K_S.gguf) | i1-Q3_K_S | 3.6 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ3_S.gguf) | i1-IQ3_S | 3.6 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ3_M.gguf) | i1-IQ3_M | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q3_K_M.gguf) | i1-Q3_K_M | 3.9 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q3_K_L.gguf) | i1-Q3_K_L | 4.2 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ4_XS.gguf) | i1-IQ4_XS | 4.3 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-IQ4_NL.gguf) | i1-IQ4_NL | 4.5 | prefer IQ4_XS |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q4_0.gguf) | i1-Q4_0 | 4.5 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.6 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q4_K_M.gguf) | i1-Q4_K_M | 4.8 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q4_1.gguf) | i1-Q4_1 | 5.0 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q5_K_S.gguf) | i1-Q5_K_S | 5.4 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q5_K_M.gguf) | i1-Q5_K_M | 5.5 | |
| [GGUF](https://huggingface.co/mradermacher/medalpaca_medical_meadow_medqa-i1-GGUF/resolve/main/medalpaca_medical_meadow_medqa.i1-Q6_K.gguf) | i1-Q6_K | 6.4 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
| [
"MEDQA"
] |
vladargunov/pubhealth-sentence-similarity | vladargunov | sentence-similarity | [
"sentence-transformers",
"safetensors",
"bert",
"sentence-similarity",
"feature-extraction",
"generated_from_trainer",
"dataset_size:16158",
"loss:CosineSimilarityLoss",
"en",
"dataset:bigbio/pubhealth",
"arxiv:1908.10084",
"base_model:sentence-transformers/all-MiniLM-L6-v2",
"base_model:finetune:sentence-transformers/all-MiniLM-L6-v2",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2024-06-11T18:02:23Z" | 2024-06-11T18:02:29+00:00 | 468 | 0 | ---
base_model: sentence-transformers/all-MiniLM-L6-v2
datasets:
- bigbio/pubhealth
language:
- en
library_name: sentence-transformers
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- sentence-similarity
- feature-extraction
- generated_from_trainer
- dataset_size:16158
- loss:CosineSimilarityLoss
widget:
- source_sentence: 'The fruit (soursop, guyabano), leaves, and bark of the graviola
tree (A. muricata), have long been utilized as a folk remedy in parts of Africa
and South America for myriad conditions. Claims of their potential to “cure” cancer,
similarly, have long been a fixture in certain regions of the Internet — fringe
health websites and supplement hucksters, primarily. In their most exaggerated
form, such claims take the form of a widespread conspiracy alleging a pharmaceutical
coverup to squash evidence of viable, powerful, and universal cure for cancer
in favor of financial gain. The dubious Health Sciences Institute, (promoter of
a previously debunked claim that Hillary Clinton has worked to hide a cancer cure
dubbed “sour honey”) described the plant’s potential this way: Since the 1970s,
the bark, leaves, roots, fruit, and fruit seeds of the Amazonian Graviola tree
have been studied in numerous laboratory tests and have shown remarkable results
with this deadly disease. Several years ago, a major pharmaceutical company began
extensive independent research on it. They learned that certain extracts of the
tree actually seek out, attack, and destroy cancer cells. […] After more than
seven years of work behind closed doors, researchers at this company realized
they couldn’t duplicate the tree’s natural properties with a patentable substance.
So they shut down the entire project. It basically came down to this—if they couldn’t
make huge profits, they would keep the news of this possible cure a well-guarded
secret. But one researcher couldn’t bear that, and decided to risk his job with
the hope of saving lives. Indeed, there has been research on many parts of, and
chemicals within, the graviola tree with regard to their ability to kill cancerous
cells. In terms of a possible mechanism, most ideas revolve around unique chemicals
contained within the fruit — annonaceous acetogenins — that may present a novel
pathway to kill cancer cells. These chemicals are found only in the family of
plants Graviola belongs to (Annonaceae) and some research indicates they may have
the ability to inhibit chemicals that aid cellular respiration, which can cause
a “programmed death” of cancer cells. Perhaps most notably, this mechanism has
been explored using extracts from graviola material against human lung, colorectal,
and liver cancer cell lines. Such studies have found that extracts were indeed
able to cause programmed cell death as hypothesized. Other studies have shown
limited potential in reducing the proliferation of cancer cells in some animals
and cell lines as well. It is worth mentioning, however, that many chemicals that
show anti-cancer properties in laboratory settings do not translate to viable
cures or treatments for cancer. Investigations on laboratory animals, too, have
shown limited but somewhat positive results with regard to the plant’s anticancer
potential. Studies on rats and mice, respectively, have shown some anti-tumor
potential with prostate cancer and breast cancer, and studies on rats have, as
well, shown potential preventive effects for colon cancer. Outside of singular
case reports from people alleging benefits from the plant, no large scale clinical
human studies have been published on its efficacy as a legitimate treatment for
cancer (at least one clinical trial has been registered, however). As such, the view
of the UK based Cancer Research, and other Cancer groups, is as follows: There
have not been any studies [of Graviola] in humans. So we don’t know whether it
can work as a cancer treatment or not. Many sites on the internet advertise and
promote graviola capsules as a cancer cure but none of them are supported by any
reputable scientific cancer organisations. Both the United States Food and Drug
administration as well as the United States Federal Trade Commission have issued
warnings to groups selling graviola extract with claims of its cancer-curing potential.
In 2008, in a press release describing a “sweep” of graviola supplement sellers,
the FTC described their products as “bogus“. Outside of overblown claims, there
are also legitimate concerns about the safety of these products. Numerous studies
have suggested that the potentially active chemicals within the graviola tree
may be neurotoxic. Epidemiological studies of cultures that regularly use the
plant in traditional medicine have shown associations between the plant’s consumption
and Parkinson’s disease: Epidemiological studies, however, linked the consumption
of Annonaceae to a high prevalence of atypical parkinsonism, in Guadeloupe, in
parts of the Afro-Caribbean and Indian population in London and New Caledonia.
In several patients who desisted in their consumption of Annonaceae fruits, the
progression of atypical parkinsonism ceased […]. Chemical investigations of active
components within the plant reveal strong evidence of its neurotoxicity, as well:
The fruit pulp extract of A. muricata revealed the strongest neurotoxic effect,
with 67% cell death at a concentration of 1 µg/mL. A high reduction in cell viability
coupled with pronounced cell death was found at 0.1 µg/mL for an Annonaceous seed
extract. These results demonstrate that the intake of dietary supplements containing
plant material from Annonaceae may be hazardous to health in terms of neurotoxicity.'
sentences:
- U.S. President Donald Trump issued a pardon for the leader of the armed group
that held migrants at gunpoint in New Mexico.
- Thanks to the immigrants who illegally cross the U.S. Mexican border, and the
Democrats who refuse to stop them, the Measles virus has been declared a public
health emergency in 2019.
- '"""The animated film """"Incredibles 2"""" contains scenes that prompted an epilepsy
warning at movie theaters."""'
- source_sentence: '"""In a regular feature called """"How the Left Destroys the Nation,""""
a website founded by the leader of a far-right group posted this headline about
one state’s coronavirus response: """"Michigan Governor Bans Gardening, Sale Of
Fruit and Vegetable Seeds, Gardening Supplies Prohibited."""" The attack on Gov.
Gretchen Whitmer, a Democrat who has been touted as a potential running mate for
presumptive Democratic presidential nominee Joe Biden, was flagged as part of
Facebook’s efforts to combat news and misinformation on its News Feed. (Read
more about our partnership with Facebook.) That’s because it’s wrong. Whitmer
has issued orders directing people to stay home and limiting some commercial activity,
but this claim goes too far. The headline appears on the Geller Report, a website
by Pamela Geller. She is an activist who co-founded Stop Islamization of America,
also known as the American Freedom Defense Initiative. Below the headline is an
article that originally appeared in The Daily Caller, a conservative-leaning publication,
that reports on an executive order issued by Whitmer in response to the COVID-19
outbreak. The article does not say that the order bans gardening, but that it
does restrict the sale of gardening supplies. In reality, executive order 2020-42,
which went into effect April 9, 2020, requires larger stores to block off certain
areas of their sales floors as a way of limiting the number of people in those
stores. The order does not ban gardening or the sale of any product, including,
as we mentioned in a previous fact-check, American flags. The numbers of coronavirus
cases in Michigan have surged in recent weeks. As of April 14, the Wolverine State
ranked fourth — behind New York, New Jersey and Massachusetts, according to the
New York Times. Nearly half of Michigan’s cases are in Wayne County, which includes
Detroit, according to Johns Hopkins University. Both the state and the county
have a COVID-19 fatality rate of 6%. It’s in that climate that Whitmer issued
this order, subtitled the """"Temporary requirement to suspend activities that
are not necessary to sustain or protect life,"""" which extended and added to
a stay-at-home order issued March 23. Tiffany Brown, a spokeswoman for the governor,
told PolitiFact that Whitmer’s order does not ban Michiganders from buying any
item. The order says that stores larger than 50,000 square feet must close areas
— """"by cordoning them off, placing signs in aisles, posting prominent signs,
removing goods from shelves, or other appropriate means — that are dedicated to
the following classes of goods: Carpet or flooring, furniture, garden centers
and plant nurseries, and paint."""" Referring to that restriction at a news conference
announcing the order, Whitmer said: """"If you’re not buying food or medicine
or other essential items, you should not be going to the store."""" As to gardening,
a frequently asked questions document released by the governor’s office states:
""""The order does not prohibit homeowners from tending to their own yards as
they see fit."""" Grocery stores, of course, remain open. And neither the order
nor the FAQs mention any restriction on the sale of fruit or seeds. A headline
shared on social media inaccurately describes an order that Whitmer issued in
response to the coronavirus. The order does not prohibit gardening or the sale
of any particular product in Michigan. Stores in Michigan larger than 50,000 square
feet must close areas for garden centers and plant nurseries, as well as those
that sell carpet or flooring, furniture and paint."""'
sentences:
- Bushfires rage out of control across southeast Australia.
- Iran records 4,585 coronavirus deaths as restrictions eased.
- '"""The Republican budget plan """"says that 10 years from now, if you’re a 65-year-old
who’s eligible for Medicare, you should have to pay nearly $6,400 more than you
would today."""'
- source_sentence: 'An old hoax about Charles Manson being paroled that was started
by a known fake news website in June 2014 resurfaced in June 2017. The rumor stems
from a 2014 report that appeared at Empire News under the headline, “Charles Manson
Granted Parole,” that reports Manson had been granted parole due to prison overcrowding:
The ruling, issued by three judges overseeing the state’s efforts to ease the overcrowding,
gives California until February 2016 to achieve their goals. But, the judges said,
the state has to make elderly inmates and those with serious illnesses eligible
for parole immediately. Manson, who was denied parole in April of 2012 and wasn’t
scheduled for another parole hearing until 2027, was re-evaluated due to his age
and health and the Parole Board recommended his parole. The site’s disclaimer,
however, states that it’s content is “intended for entertainment purposes only,”
meaning that its reporting should not be taken as fact. It’s not clear why Charles
Manson parole rumors resurfaced in June 2017. Manson was denied parole by the
California Department of Corrections in 2012 and his next parole hearing was scheduled
for 2027, when Manson would be 92 years old. In January 2017, however, Manson
was transferred to a hospital for treatment of gastrointestinal bleeding, and
Manson’s condition was described as “serious” by family members. He had been transferred
back to prison by the time the rumor resurfaced. It’s possible that parole decisions
regarding the release of other former Manson Family members could have contributed
to Charles Manson parole rumors resurfacing. A panel recommended the release of a
former Manson Family member named Bruce Davis who murdered musician Gary Hinman
and stuntman Donald “Shorty” Shea in 1969. The final decision, however, will rest
with California Gov. Jerry Brown, who had about five months to make a decision.
the Los Angeles Times reports. Meanwhile, an appeals panel postponed a decision
on wether or not to recommend the release of former Manson Family member Patricia
Krenwinkel in December 2016, Fox News reports. Krenwinkel was present at the 1969
murder of Sharon Tate and four others. But regardless of developments with other
members of the Manson Family, all Charles Manson parole rumors should be considered
“fiction” until at least 2027, when his next hearing is scheduled. Comments'
sentences:
- '"""Common usage of the phrase """"Always a bridesmaid but never a bride"""" originated
with an advertising campaign for Listerine mouthwash."""'
- Colorado governor signs recreational marijuana regulations into law.
- State to consider 6 conditions to treat with medical pot.
- source_sentence: 'A “Chicken Soup”-like tale warning us against the folly of judging
people solely by appearances hit the Internet in mid-1998. As usual, the framework
of the tale bore some general resemblance to the truth, but details were greatly
altered so as to turn it into something quite different from the real story: The
President of Harvard made a mistake by prejudging people and it cost him dearly.
A lady in a faded gingham dress and her husband, dressed in a homespun threadbare
suit, stepped off the train in Boston, and walked timidly without an appointment
into the president’s outer office. The secretary could tell in a moment that such
backwoods, country hicks had no business at Harvard and probably didn’t even deserve
to be in Cambridge. She frowned. “We want to see the president,” the man said
softly. “He’ll be busy all day,” the secretary snapped. “We’ll wait,” the lady
replied. For hours, the secretary ignored them, hoping that the couple would finally
become discouraged and go away. They didn’t. And the secretary grew frustrated
and finally decided to disturb the president, even though it was a chore she always
regretted to do. “Maybe if they just see you for a few minutes, they’ll leave,”
she told him. And he signed in exasperation and nodded. Someone of his importance
obviously didn’t have the time to spend with them, but he detested gingham dresses
and homespun suits cluttering up his outer office. The president, stern-faced
with dignity, strutted toward the couple. The lady told him, “We had a son that
attended Harvard for one year. He loved Harvard. He was happy here. But about
a year ago, he was accidentally killed. And my husband and I would like to erect
a memorial to him, somewhere on campus.” The president wasn’t touched; he was
shocked. “Madam,” he said gruffly, “We can’t put up a statue for every person
who attended Harvard and died. If we did, this place would look like a cemetery.”
“Oh, no,” the lady explained quickly, “We don’t want to erect a statue. We thought
we would like to give a building to Harvard.” The president rolled his eyes. He
glanced at the gingham dress and homespun suit, then exclaimed, “A building! Do
you have any earthly idea how much a building costs? We have over seven and a
half million dollars in the physical plant at Harvard.” For a moment the lady
was silent. The president was pleased. He could get rid of them now. And the lady
turned to her husband and said quietly, “Is that all it costs to start a University?
Why don’t we just start our own?” Her husband nodded. The president’s face wilted
in confusion and bewilderment. And Mr. and Mrs. Leland Stanford walked away, traveling
to Palo Alto, California, where they established the University that bears their
name, a memorial to a son that Harvard no longer cared about. The very premise
of the tale was completely implausible. Leland Stanford (1824-93) was one of the
most prominent men of his time in America: He was a wealthy railroad magnate who
built the Central Pacific Railroad (and drove the gold spike to symbolize the
completion of the first transcontinental rail line at Promontory Summit, Utah,
in 1869), as well as a Republican Party leader who served as California’s eighth
governor (1862-63) and later represented that state in the U.S. Senate (1885-93).
He was an imposing figure, hardly the type of man to dress in a “homespun threadbare
suit,” walk “timidly” into someone’s office without an appointment, and sit cooling
his heels “for hours” until someone deigned to see him. Harvard’s president would
had to have been an ignorant buffoon not to recognize Stanford’s name and promptly
greet him upon hearing of his arrival: Moreover, the Stanfords’ only son (Leland
Stanford, Jr.) died of typhoid fever at age 15, in Florence, Italy. His death
would hardly have been described as “accidental,” nor had he spent a year studying
at Harvard while barely into his teens: The family was in Italy in 1884 when
Leland contracted typhoid fever. He was thought to be recovering, but on March
13 at the Hotel Bristol in Florence, Leland’s bright and promising young life
came to an end, a few weeks before his 16th birthday. Stanford, who had remained
at Lelands’ bedside continuously, fell into a troubled sleep the morning the boy
died. When he awakened he turned to his wife and said, “The children of California
shall be our children.” These words were the real beginning of Stanford University.
The closest this story came to reality was in its acknowledgement that in 1884,
a few month’s after their son’s death, the Stanfords did pay a visit to Harvard
and met with that institution’s president, Charles Eliot. However, the couple
did not go there with the purpose of donating a building to Harvard as a memorial
to their dead son — they intended to establish some form of educational facility
of their own in northern California, and so they visited several prominent Eastern
schools to gather ideas and suggestions about what they might build, as Stanford’s
website described the meeting: The Stanfords … visited Cornell, Yale, Harvard
and Massachusetts Institute of Technology. They talked with President Eliot of
Harvard about three ideas: a university at Palo Alto, a large institution in San
Francisco combining a lecture hall and a museum, and a technical school. They
asked him which of these seemed most desirable and President Eliot answered, a
university. Mrs. Stanford then asked him how much the endowment should be, in
addition to land and buildings, and he replied, not less than $5 million. A silence
followed and Mrs. Stanford looked grave. Finally, Mr. Stanford said with a smile,
“Well, Jane, we could manage that, couldn’t we?” and Mrs. Stanford nodded her
assent. They settled on creating a great university, one that, from the outset,
was untraditional: coeducational, in a time when most were all-male; nondenominational,
when most were associated with a religious organization; avowedly practical, producing
“cultured and useful citizens” when most were concerned only with the former.
Although they consulted with several of the presidents of leading institutions,
the founders were not content to model their university after eastern schools.
The Stanfords did found their university, modeled after Cornell and located on
the grounds of their horse-trotting farm, in memory of their son (hence the school’s
official name of “Leland Stanford Junior University”) — not because they were
rudely rebuffed by Harvard’s president, but rather because it was what they had
planned all along. The “rudely-spurned university endowment” theme of the Stanford
story has reportedly played out at least once in real life. In July 1998, William
Lindsay of Las Vegas said he contacted an unnamed Scottish institution of higher
learning by telephone and told them he intended to give some money to a university
in Scotland. Taking him for a crank, the person he spoke to rudely dismissed him.
His next call to Glasgow University met with a warmer reception, and in March
2000 that school received a check for £1.2 million, enough to endow a professorship
in Lindsay’s name.'
sentences:
- Early study results suggest 2 Ebola treatments saving lives.
- '"""Honduras """"bans citizens from owning guns"""" and has the """"highest homicide
rate in the entire world."""" Switzerland, with a similar population, """"requires
citizens to own guns"""" and has the """"lowest homicide rate in the entire world."""'
- Pat Robertson asserted the Orlando nightclub shooting was God's punishment for
legalizing same-sex marriage.
- source_sentence: '"""A chain message circulating on messaging apps claims the United
States is about to enter a period of federally mandated quarantine. The source:
""""my aunt’s friend"""" who works for the government. There is no evidence of
this. The message, which a reader sent us a screenshot of on March 16, appears
in a group chat on iMessage. The sender claims to have information from """"my
aunt''s friend"""" who works for the Centers for Disease Control and Prevention
and """"just got out of a meeting with Trump."""" """"He’s announcing tomorrow
that the U.S. is going into quarantine for the next 14 days,"""" the message reads.
""""Meaning everyone needs to stay in their homes/where they are."""" We’ve seen
screenshots of similar messages circulating on WhatsApp, a private messaging app
that’s popular abroad. Misinformation tends to get passed around via chain messages
during major news events, so we looked into this one. (Screenshots) There is no
evidence that the federal government is set to announce a nationwide lockdown
like the ones seen in France, Italy and Spain. President Donald Trump and the
National Security Council have both refuted the claim. So far, officials have
advised Americans to practice """"social distancing,"""" or avoiding crowded public
spaces. In a press conference March 16, Trump outlined several recommendations
to prevent the spread of the coronavirus. Among them is avoiding gatherings of
10 or more people. """"My administration is recommending that all Americans, including
the young and healthy, work to engage in schooling from home when possible, avoid
gathering in groups of more than 10 people, avoid discretionary travel and avoid
eating and drinking in bars, restaurants and public food courts,"""" he said.
In response to a question, he said the administration is not considering a national
curfew or quarantine. He reiterated that point in another press conference March
17. """"It’s a very big step. It’s something we talk about, but we haven’t decided
to do that,"""" he said. Andrew Cuomo ordered a one-mile containment zone on March
10. Large gathering spots were closed for 14 days and National Guard troops are
delivering food to people. In the San Francisco Bay Area, local officials on March
16 announced sweeping measures to try to contain the coronavirus. Residents of
six counties have been ordered to """"shelter in place"""" in their homes and
stay away from others as much as possible for the next three weeks. The move falls
short of a total lockdown. At the federal level, the CDC does have the power to
quarantine people who may have come in contact with someone infected by the coronavirus,
but most quarantines are done voluntarily. And decisions are usually left up to
states and localities. We reached out to the CDC for comment on the chain message,
but we haven’t heard back. The chain message is inaccurate. If you receive a chain
message that you want us to fact-check, send a screenshot to [email protected]."""'
sentences:
- Texas guard Andrew Jones diagnosed with leukemia.
- Treadmill classes mix it up with workhorse of the gym.
- Drug overdoses are now the second-most common cause of death in New Hampshire.
---
# SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2
This is a [sentence-transformers](https://www.SBERT.net) model finetuned from [sentence-transformers/all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) on the [bigbio/pubhealth](https://huggingface.co/datasets/bigbio/pubhealth) dataset. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.
## Model Details
### Model Description
- **Model Type:** Sentence Transformer
- **Base model:** [sentence-transformers/all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) <!-- at revision 8b3219a92973c328a8e22fadcfa821b5dc75636a -->
- **Maximum Sequence Length:** 256 tokens
- **Output Dimensionality:** 384 tokens
- **Similarity Function:** Cosine Similarity
- **Training Dataset:**
- [bigbio/pubhealth](https://huggingface.co/datasets/bigbio/pubhealth)
- **Language:** en
<!-- - **License:** Unknown -->
### Model Sources
- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
- **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers)
- **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers)
### Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel
(1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
(2): Normalize()
)
```
## Usage
### Direct Usage (Sentence Transformers)
First install the Sentence Transformers library:
```bash
pip install -U sentence-transformers
```
Then you can load this model and run inference.
```python
from sentence_transformers import SentenceTransformer
# Download from the 🤗 Hub
model = SentenceTransformer("vladargunov/pubhealth-sentence-similarity")
# Run inference
sentences = [
'"""A chain message circulating on messaging apps claims the United States is about to enter a period of federally mandated quarantine. The source: """"my aunt’s friend"""" who works for the government. There is no evidence of this. The message, which a reader sent us a screenshot of on March 16, appears in a group chat on iMessage. The sender claims to have information from """"my aunt\'s friend"""" who works for the Centers for Disease Control and Prevention and """"just got out of a meeting with Trump."""" """"He’s announcing tomorrow that the U.S. is going into quarantine for the next 14 days,"""" the message reads. """"Meaning everyone needs to stay in their homes/where they are."""" We’ve seen screenshots of similar messages circulating on WhatsApp, a private messaging app that’s popular abroad. Misinformation tends to get passed around via chain messages during major news events, so we looked into this one. (Screenshots) There is no evidence that the federal government is set to announce a nationwide lockdown like the ones seen in France, Italy and Spain. President Donald Trump and the National Security Council have both refuted the claim. So far, officials have advised Americans to practice """"social distancing,"""" or avoiding crowded public spaces. In a press conference March 16, Trump outlined several recommendations to prevent the spread of the coronavirus. Among them is avoiding gatherings of 10 or more people. """"My administration is recommending that all Americans, including the young and healthy, work to engage in schooling from home when possible, avoid gathering in groups of more than 10 people, avoid discretionary travel and avoid eating and drinking in bars, restaurants and public food courts,"""" he said. In response to a question, he said the administration is not considering a national curfew or quarantine. He reiterated that point in another press conference March 17. """"It’s a very big step. It’s something we talk about, but we haven’t decided to do that,"""" he said. Andrew Cuomo ordered a one-mile containment zone on March 10. Large gathering spots were closed for 14 days and National Guard troops are delivering food to people. In the San Francisco Bay Area, local officials on March 16 announced sweeping measures to try to contain the coronavirus. Residents of six counties have been ordered to """"shelter in place"""" in their homes and stay away from others as much as possible for the next three weeks. The move falls short of a total lockdown. At the federal level, the CDC does have the power to quarantine people who may have come in contact with someone infected by the coronavirus, but most quarantines are done voluntarily. And decisions are usually left up to states and localities. We reached out to the CDC for comment on the chain message, but we haven’t heard back. The chain message is inaccurate. If you receive a chain message that you want us to fact-check, send a screenshot to [email\xa0protected]."""',
'Drug overdoses are now the second-most common cause of death in New Hampshire.',
'Treadmill classes mix it up with workhorse of the gym.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]
# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]
```
<!--
### Direct Usage (Transformers)
<details><summary>Click to see the direct usage in Transformers</summary>
</details>
-->
<!--
### Downstream Usage (Sentence Transformers)
You can finetune this model on your own dataset.
<details><summary>Click to expand</summary>
</details>
-->
<!--
### Out-of-Scope Use
*List how the model may foreseeably be misused and address what users ought not to do with the model.*
-->
<!--
## Bias, Risks and Limitations
*What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.*
-->
<!--
### Recommendations
*What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.*
-->
## Training Details
### Training Dataset
#### bigbio/pubhealth
* Dataset: [bigbio/pubhealth](https://huggingface.co/datasets/bigbio/pubhealth)
* Size: 16,158 training samples
* Columns: <code>sentence2</code>, <code>sentence1</code>, and <code>score</code>
* Approximate statistics based on the first 1000 samples:
| | sentence2 | sentence1 | score |
|:--------|:-------------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:-----------------------------|
| type | string | string | int |
| details | <ul><li>min: 91 tokens</li><li>mean: 246.21 tokens</li><li>max: 256 tokens</li></ul> | <ul><li>min: 6 tokens</li><li>mean: 21.43 tokens</li><li>max: 96 tokens</li></ul> | <ul><li>0: 100.00%</li></ul> |
* Samples:
| sentence2 | sentence1 | score |
|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:----------------------------------------------------------------------------|:---------------|
| <code>"""Hillary Clinton is in the political crosshairs as the author of a new book alleges improper financial ties between her public and personal life. At issue in conservative author Peter Schweizer’s forthcoming book Clinton Cash are donations from foreign governments to the Clinton Foundation during the four years she served as secretary of state. George Stephanopoulos used an interview with Schweizer on ABC This Week to point out what other nonpartisan journalists have found: There is no """"smoking gun"""" showing that donations to the foundation influenced her foreign policy decisions. Still, former Republican House Speaker Newt Gingrich says the donations are """"clearly illegal"""" under federal law. In his view, a donation by a foreign government to the Clinton Foundation while Clinton was secretary of state is the same as money sent directly to her, he said, even though she did not join the foundation’s board until she left her post. """"The Constitution of the United States says you cannot take money from foreign governments without explicit permission of the Congress. They wrote that in there because they knew the danger of corrupting our system by foreign money is enormous,"""" Gingrich said. """"You had a sitting secretary of state whose husband radically increased his speech fees, you have a whole series of dots on the wall now where people gave millions of dollars — oh, by the way, they happen to get taken care of by the State Department."""" He continued, """"My point is they took money from foreign governments while she was secretary of State. That is clearly illegal."""" PunditFact wanted to know if a criminal case against Clinton is that open and shut. Is what happened """"clearly illegal""""? A spokesman for the Clinton Foundation certainly disagreed, calling Gingrich’s accusation """"a baseless leap"""" because Clinton was not part of her husband’s foundation while serving as a senator or secretary of state. We did not hear from Gingrich by our deadline. Foundation basics Former President Clinton started the William J. Clinton Foundation in 2001, the year after Hillary Clinton won her first term as a New York senator. The foundation works with non-governmental organizations, the private sector and governments around the world on health, anti-poverty, HIV/AIDS and climate change initiatives. Spokesman Craig Minassian said it’s reasonable for the foundation to accept money from foreign governments because of the global scope of its programs, and the donations are usually in the form of tailored grants for specific missions. Hillary Clinton was not part of her husband’s foundation while she was a senator or secretary of state. Her appointment to the latter post required Senate confirmation and came with an agreement between the White House and Clinton Foundation that the foundation would be more transparent about its donors. According to the 2008 memorandum of understanding, the foundation would release information behind new donations and could continue to collect donations from countries with which it had existing relationships or running grant programs. If countries with existing contributions significantly stepped up their contributions, or if a new foreign government wanted to donate, the State Department would have to approve. Clinton took an active role in fundraising when she left the State Department and the foundation became the Bill, Hillary & Chelsea Clinton Foundation in 2013. But she left the board when she announced her run for the presidency in April 2015. The Emoluments Clause So how does Gingrich come up with the claim that Clinton Foundation donations are """"clearly illegal"""" and unconstitutional? The answer is something known as the Emoluments Clause. A few conservative websites have made similar arguments in recent days, including the Federalist blog. The Emoluments Clause, found in Article 1, Section 9 of the Constitution, reads in part: """"No Title of Nobility shall be granted by the United States: And no Person holding any Office of Profit or Trust under them, shall, without the Consent of the Congress, accept of any present, Emolument, Office, or Title, of any kind whatever, from any King, Prince, or foreign State."""" The framers came up with this clause to prevent the government and leaders from granting or receiving titles of nobility and to keep leaders free of external influence. (An emolument, per Merriam-Webster Dictionary, is """"the returns arising from office or employment usually in the form of compensation or perquisites."""") Lest you think the law is no longer relevant, the Pentagon ethics office in 2013 warned employees the """"little known provision"""" applies to all federal employees and military retirees. There’s no mention of spouses in the memo. J. Peter Pham, director of the Atlantic Council’s Africa Center, said interpretation of the clause has evolved since its adoption at the Constitutional Convention, when the primary concern was about overseas diplomats not seeking gifts from foreign powers they were dealing with. The Defense Department memo, in his view, goes beyond what the framers envisioned for the part of the memo dealing with gifts. """"I think that, aside from the unambiguous parts, the burden would be on those invoking the clause to show actual causality that would be in violation of the clause,"""" Pham said. Expert discussion We asked seven different constitutional law experts on whether the Clinton Foundation foreign donations were """"clearly illegal"""" and a violation of the Emoluments Clause. We did not reach a consensus with their responses, though a majority thought the layers of separation between the foundation and Hillary Clinton work against Gingrich. The American system often distinguishes between public officers and private foundations, """"even if real life tends to blur some of those distinctions,"""" said American University law professor Steve Vladeck. Vladeck added that the Emoluments Clause has never been enforced. """"I very much doubt that the first case in its history would be because a foreign government made charitable donations to a private foundation controlled by a government employee’s relative,"""" he said. """"Gingrich may think that giving money to the Clinton Foundation and giving money to then-Secretary Clinton are the same thing. Unfortunately for him, for purposes of federal regulations, statutes, and the Constitution, they’re formally — and, thus, legally — distinct."""" Robert Delahunty, a University of St. Thomas constitutional law professor who worked in the Justice Department’s Office of Legal Counsel from 1989 to 2003, also called Gingrich’s link between Clinton and the foreign governments’ gifts to the Clinton Foundation as """"implausible, and in any case I don’t think we have the facts to support it."""" """"The truth is that we establish corporate bodies like the Clinton Foundation because the law endows these entities with a separate and distinct legal personhood,"""" Delahunty said. John Harrison, University of Virginia law professor and former deputy assistant attorney general in the Office of Legal Counsel from 1990 to 1993, pointed to the Foreign Gifts Act, 5 U.S.C. 7432, which sets rules for how the Emoluments Clause should work in practice. The statute spells out the minimal value for acceptable gifts, and says it applies to spouses of the individuals covered, but """"it doesn’t say anything about receipt of foreign gifts by other entities such as the Clinton Foundation."""" """"I don’t know whether there’s any other provision of federal law that would treat a foreign gift to the foundation as having made to either of the Clintons personally,"""" Harrison said, who added that agencies have their own supplemental rules for this section, and he did not know if the State Department addressed this. Other experts on the libertarian side of the scale thought Gingrich was more right in his assertion. Clinton violates the clause because of its intentionally broad phrasing about gifts of """"any kind whatever,"""" which would cover indirect gifts via the foundation, said Dave Kopel, a constitutional law professor at Denver University and research director at the libertarian Independence Institute. Kopel also brought up bribery statutes, which would require that a gift had some influence in Clinton’s decision while secretary of state. Delahunty thought Kopel’s reasoning would have """"strange consequences,"""" such as whether a state-owned airline flying Bill Clinton to a conference of former heads of state counted as a gift to Hillary Clinton. Our ruling Gingrich said the Clinton Foundation """"took money from from foreign governments while (Hillary Clinton) was secretary of state. It is clearly illegal. … The Constitution says you can’t take this stuff."""" A clause in the Constitution does prohibit U.S. officials such as former Secretary of State Hillary Clinton from receiving gifts, or emoluments, from foreign governments. But the gifts in this case were donations from foreign governments that went to the Clinton Foundation, not Hillary Clinton. She was not part of the foundation her husband founded while she was secretary of state. Does that violate the Constitution? Some libertarian-minded constitutional law experts say it very well could. Others are skeptical. What’s clear is there is room for ambiguity, and the donations are anything but """"clearly illegal."""" The reality is this a hazy part of U.S. constitutional law."</code> | <code>Britain plans for opt-out organ donation scheme to save lives.</code> | <code>0</code> |
| <code>The story does discuss costs, but the framing is problematic. The story, based on a conversation with one source, the study’s lead investigator, says, “It’s difficult at this point to predict costs. However, he expects costs will not approach those for Provenge, the pricey treatment vaccine for prostate cancer approved by the FDA in 2010. Provenge costs $93,000 for the one-month, three-dose treatment. Medicare covers it.” This tells readers that, no matter what the drug costs, Medicare likely will cover it. We appreciate the effort to bring cost information into the story, but this type of information is misleading. The story does explain that only one patient remains cancer free following the study. It then details how for most of the patients cancer continued to progress after 2 months. It says that the median overall survival in both the breast cancer and ovarian cancer patients was less than 16 months. But the story is framed in such a way to highlight the one potentially positive outcome of the study and to downplay the negative. We read more sooner about the one patient who may have responded well to the vaccine than we do about the 25 other patients who did not. The story mentions side effects in a satisfactory way. Technically, the story provides readers with much of the information they would need to assess the validity of the study, but it comes out in bits and pieces. For example, we only find out near the end of the story that “The woman, who remains disease-free, had a previous treatment with a different treatment vaccine. ‘That might have primed her immune system,’ Gulley speculates. She also had only one regimen of chemotherapy, perhaps keeping her immune system stronger.” This casts much doubt on the study’s design, and it would have been nice to have seen some outside expertise brought in to either discuss those design problems or to torpedo the story altogether. Again, the story deserves high marks for being very specific in the lead and throughout the story. It says, that the vaccine is “for breast and ovarian cancer that has spread to other parts of the body” in the lead and later details the particular circumstances of the study cohort. It says, “The patients had already undergone a variety of treatments but the cancer was progressing. Twenty one of the 26 had undergone three or more chemotherapy regimens.” This is the root of the story’s main shortcoming. Almost all of the information in the story comes from one source: Dr. James Gulley, who oversaw the study. Gulley is quite enthusiastic about this vaccine, despite the evidence, and the story needed more perspectives to put this vaccine into a broader context. At the very end, there are a few comments from Dr. Vincent K. Tuohy, who also is working on a breast cancer vaccine. Because of his competing research, he seems to have a conflict, but even putting that aside, his comments were not used to their best effect. There was no comparison in the story to existing alternatives. The median survival, for example, is presented without the context of how long these patients might have lived had they been undergoing standard chemotherapy and radiation treatments. We give high marks to the story for saying right in the lead that the findings are from “a preliminary study in 26 patients.” That tells readers both that the findings need to be interpreted with caution and that the treatment is not available to most people. The concept of vaccines for breast/ovarian cancer is indeed novel, and the story acknowledges that other vaccines are being studied. The story does not rely on a news release.</code> | <code>Virus raises specter of gravest attacks in modern US times.</code> | <code>0</code> |
| <code>"""Although the story didn’t cite the cost of appendectomy – emergency or urgent surgery – and we wish it had, we nonetheless will give it a satisfactory score because it at least cited what the editorial writer wrote, """"A secondary benefit is the savings to the hospital generated by minimizing staff and anesthesiologist presence late in the evening and during the wee hours of the morning."""" As with our harms score above, although the story didn’t give absolute numbers, in this case we think it was sufficient for it to report that """"The scientists found no significant difference among the groups in the patients’ condition 30 days after surgery or in the length of their operation or hospital stay."""" Although the story didn’t give absolute numbers, in this case we think it was sufficient for it to report that """"The scientists found no significant difference among the groups in the patients’ condition 30 days after surgery or in the length of their operation or hospital stay."""" Despite running less than 300 words, this story did an adequate job in explaining the quality of the evidence, including pointing out limitations. No disease-mongering here. The story meets the bare minimum requirement for this criterion in that it at least cited what an editorial stated. The focus of the story was on a study comparing emergency appendectomy with surgery done up to 12 hours later or beyond. This is the whole focus of the story – and one we applaud – when it begins: """"Appendectomy is the most common emergency surgery in the world, but it doesn’t have to be."""" There were no claims made about the novelty of this research, and we may have wished for a bit more context on this. Nonetheless, the potential for guiding future care decisions was made clear. Not applicable. Given that the story only pulled excerpts from the journal article and the accompanying editorial, and didn’t include any fresh quotes from interviews, we can’t be sure of the extent to which it may have been influenced by a news release."""</code> | <code>Legionnaires’ case identified at Quincy veterans’ home.</code> | <code>0</code> |
* Loss: [<code>CosineSimilarityLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cosinesimilarityloss) with these parameters:
```json
{
"loss_fct": "torch.nn.modules.loss.MSELoss"
}
```
### Training Hyperparameters
#### Non-Default Hyperparameters
- `per_device_train_batch_size`: 128
- `learning_rate`: 2e-05
- `num_train_epochs`: 10
- `warmup_ratio`: 0.1
- `batch_sampler`: no_duplicates
#### All Hyperparameters
<details><summary>Click to expand</summary>
- `overwrite_output_dir`: False
- `do_predict`: False
- `eval_strategy`: no
- `prediction_loss_only`: True
- `per_device_train_batch_size`: 128
- `per_device_eval_batch_size`: 8
- `per_gpu_train_batch_size`: None
- `per_gpu_eval_batch_size`: None
- `gradient_accumulation_steps`: 1
- `eval_accumulation_steps`: None
- `learning_rate`: 2e-05
- `weight_decay`: 0.0
- `adam_beta1`: 0.9
- `adam_beta2`: 0.999
- `adam_epsilon`: 1e-08
- `max_grad_norm`: 1.0
- `num_train_epochs`: 10
- `max_steps`: -1
- `lr_scheduler_type`: linear
- `lr_scheduler_kwargs`: {}
- `warmup_ratio`: 0.1
- `warmup_steps`: 0
- `log_level`: passive
- `log_level_replica`: warning
- `log_on_each_node`: True
- `logging_nan_inf_filter`: True
- `save_safetensors`: True
- `save_on_each_node`: False
- `save_only_model`: False
- `restore_callback_states_from_checkpoint`: False
- `no_cuda`: False
- `use_cpu`: False
- `use_mps_device`: False
- `seed`: 42
- `data_seed`: None
- `jit_mode_eval`: False
- `use_ipex`: False
- `bf16`: False
- `fp16`: False
- `fp16_opt_level`: O1
- `half_precision_backend`: auto
- `bf16_full_eval`: False
- `fp16_full_eval`: False
- `tf32`: None
- `local_rank`: 0
- `ddp_backend`: None
- `tpu_num_cores`: None
- `tpu_metrics_debug`: False
- `debug`: []
- `dataloader_drop_last`: False
- `dataloader_num_workers`: 0
- `dataloader_prefetch_factor`: None
- `past_index`: -1
- `disable_tqdm`: False
- `remove_unused_columns`: True
- `label_names`: None
- `load_best_model_at_end`: False
- `ignore_data_skip`: False
- `fsdp`: []
- `fsdp_min_num_params`: 0
- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
- `fsdp_transformer_layer_cls_to_wrap`: None
- `accelerator_config`: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
- `deepspeed`: None
- `label_smoothing_factor`: 0.0
- `optim`: adamw_torch
- `optim_args`: None
- `adafactor`: False
- `group_by_length`: False
- `length_column_name`: length
- `ddp_find_unused_parameters`: None
- `ddp_bucket_cap_mb`: None
- `ddp_broadcast_buffers`: False
- `dataloader_pin_memory`: True
- `dataloader_persistent_workers`: False
- `skip_memory_metrics`: True
- `use_legacy_prediction_loop`: False
- `push_to_hub`: False
- `resume_from_checkpoint`: None
- `hub_model_id`: None
- `hub_strategy`: every_save
- `hub_private_repo`: False
- `hub_always_push`: False
- `gradient_checkpointing`: False
- `gradient_checkpointing_kwargs`: None
- `include_inputs_for_metrics`: False
- `eval_do_concat_batches`: True
- `fp16_backend`: auto
- `push_to_hub_model_id`: None
- `push_to_hub_organization`: None
- `mp_parameters`:
- `auto_find_batch_size`: False
- `full_determinism`: False
- `torchdynamo`: None
- `ray_scope`: last
- `ddp_timeout`: 1800
- `torch_compile`: False
- `torch_compile_backend`: None
- `torch_compile_mode`: None
- `dispatch_batches`: None
- `split_batches`: None
- `include_tokens_per_second`: False
- `include_num_input_tokens_seen`: False
- `neftune_noise_alpha`: None
- `optim_target_modules`: None
- `batch_eval_metrics`: False
- `batch_sampler`: no_duplicates
- `multi_dataset_batch_sampler`: proportional
</details>
### Training Logs
| Epoch | Step | Training Loss |
|:------:|:----:|:-------------:|
| 0.7874 | 100 | 0.0603 |
| 1.5748 | 200 | 0.131 |
| 2.3622 | 300 | 0.1188 |
| 3.1496 | 400 | 0.1173 |
| 3.9370 | 500 | 0.0551 |
| 4.7244 | 600 | 0.0622 |
| 5.5118 | 700 | 0.0454 |
| 6.2992 | 800 | 0.0521 |
| 7.0866 | 900 | 0.0478 |
| 7.8740 | 1000 | 0.0403 |
| 8.6614 | 1100 | 0.035 |
| 9.4488 | 1200 | 0.0386 |
### Framework Versions
- Python: 3.10.13
- Sentence Transformers: 3.0.1
- Transformers: 4.41.2
- PyTorch: 2.1.2
- Accelerate: 0.30.1
- Datasets: 2.19.2
- Tokenizers: 0.19.1
## Citation
### BibTeX
#### Sentence Transformers
```bibtex
@inproceedings{reimers-2019-sentence-bert,
title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
author = "Reimers, Nils and Gurevych, Iryna",
booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
month = "11",
year = "2019",
publisher = "Association for Computational Linguistics",
url = "https://arxiv.org/abs/1908.10084",
}
```
<!--
## Glossary
*Clearly define terms in order to be accessible across audiences.*
-->
<!--
## Model Card Authors
*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*
-->
<!--
## Model Card Contact
*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
--> | [
"BEAR",
"PUBHEALTH"
] |
ugaray96/biobert_ncbi_disease_ner | ugaray96 | token-classification | [
"transformers",
"pytorch",
"tf",
"jax",
"safetensors",
"bert",
"token-classification",
"disease",
"biology",
"medical",
"en",
"dataset:ncbi_disease",
"license:openrail",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2022-03-02T23:29:05Z" | 2023-04-25T06:26:14+00:00 | 464 | 19 | ---
datasets:
- ncbi_disease
language:
- en
license: openrail
tags:
- disease
- biology
- medical
widget:
- text: The patient was diagnosed with lung cancer and started chemotherapy.
- text: The patient has a history of heart disease and high blood pressure.
- text: The patient was diagnosed with diabetes and prescribed insulin therapy.
---
# Model Description
This model is a fine-tuned version of BioBERT on the NCBI disease dataset for named entity recognition (NER) of diseases. It can be used to extract disease mentions from unstructured text in the medical and biological domains.
# Intended Use
This model is intended for use in extracting disease mentions from unstructured text in the medical and biological domains. It can be used to improve information retrieval and knowledge extraction in these fields.
# Training Data
This model was trained on the [NCBI disease dataset](https://huggingface.co/datasets/ncbi_disease), which consists of 793 PubMed abstracts with 6892 disease mentions.
# How to use
You can use this model with the Hugging Face Transformers library. Here’s an example of how to load the model and use it to extract disease mentions from text:
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("ugaray96/biobert_ncbi_disease_ner")
model = AutoModelForTokenClassification.from_pretrained(
"ugaray96/biobert_ncbi_disease_ner"
)
ner_pipeline = pipeline("ner", model=model, tokenizer=tokenizer)
text = "The patient was diagnosed with lung cancer and started chemotherapy. They also have a history of diabetes and heart disease."
result = ner_pipeline(text)
diseases = []
for entity in result:
if entity["entity"] == "Disease":
diseases.append(entity["word"])
elif entity["entity"] == "Disease Continuation" and diseases:
diseases[-1] += f" {entity['word']}"
print(f"Diseases: {', '.join(diseases)}")
```
This should output: `Diseases: lung cancer, diabetes, heart disease` | [
"NCBI DISEASE"
] |
NitzanBar/umls-spanbert | NitzanBar | text-classification | [
"transformers",
"safetensors",
"bert",
"text-classification",
"medical",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-02-13T10:37:14Z" | 2024-02-13T12:09:21+00:00 | 462 | 1 | ---
license: apache-2.0
tags:
- medical
---
Based ob the paper: "UmlsBERT: Augmenting Contextual Embeddings with a Clinical Metathesaurus" (https://aclanthology.org/2021.naacl-main.139.pdf).
and the github repo: https://github.com/gmichalo/UmlsBERT
Changing base model to SpanBert instead of Bert.
Trained from scratch on MIMIC dataset, using the UMLS dataset to mask words within the text.
We achived better accuracy on MedNLI dataset.
Bert Model accuracy: 83%
SpanBert Model accuracy: 86% | [
"MEDNLI"
] |
OysterHR/gte-base-en-v1.5 | OysterHR | sentence-similarity | [
"transformers",
"onnx",
"safetensors",
"new",
"feature-extraction",
"sentence-transformers",
"gte",
"mteb",
"transformers.js",
"sentence-similarity",
"custom_code",
"en",
"arxiv:2407.19669",
"arxiv:2308.03281",
"license:apache-2.0",
"model-index",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2024-11-27T22:22:29Z" | 2024-11-27T22:50:55+00:00 | 459 | 0 | ---
language:
- en
library_name: transformers
license: apache-2.0
tags:
- sentence-transformers
- gte
- mteb
- transformers.js
- sentence-similarity
model-index:
- name: gte-base-en-v1.5
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 74.7910447761194
- type: ap
value: 37.053785713650626
- type: f1
value: 68.51101510998551
- task:
type: Classification
dataset:
name: MTEB AmazonPolarityClassification
type: mteb/amazon_polarity
config: default
split: test
revision: e2d317d38cd51312af73b3d32a06d1a08b442046
metrics:
- type: accuracy
value: 93.016875
- type: ap
value: 89.17750268426342
- type: f1
value: 92.9970977240524
- task:
type: Classification
dataset:
name: MTEB AmazonReviewsClassification (en)
type: mteb/amazon_reviews_multi
config: en
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 53.312000000000005
- type: f1
value: 52.98175784163017
- task:
type: Retrieval
dataset:
name: MTEB ArguAna
type: mteb/arguana
config: default
split: test
revision: c22ab2a51041ffd869aaddef7af8d8215647e41a
metrics:
- type: map_at_1
value: 38.193
- type: map_at_10
value: 54.848
- type: map_at_100
value: 55.388000000000005
- type: map_at_1000
value: 55.388999999999996
- type: map_at_3
value: 50.427
- type: map_at_5
value: 53.105000000000004
- type: mrr_at_1
value: 39.047
- type: mrr_at_10
value: 55.153
- type: mrr_at_100
value: 55.686
- type: mrr_at_1000
value: 55.688
- type: mrr_at_3
value: 50.676
- type: mrr_at_5
value: 53.417
- type: ndcg_at_1
value: 38.193
- type: ndcg_at_10
value: 63.486
- type: ndcg_at_100
value: 65.58
- type: ndcg_at_1000
value: 65.61
- type: ndcg_at_3
value: 54.494
- type: ndcg_at_5
value: 59.339
- type: precision_at_1
value: 38.193
- type: precision_at_10
value: 9.075
- type: precision_at_100
value: 0.9939999999999999
- type: precision_at_1000
value: 0.1
- type: precision_at_3
value: 22.096
- type: precision_at_5
value: 15.619
- type: recall_at_1
value: 38.193
- type: recall_at_10
value: 90.754
- type: recall_at_100
value: 99.431
- type: recall_at_1000
value: 99.644
- type: recall_at_3
value: 66.28699999999999
- type: recall_at_5
value: 78.094
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringP2P
type: mteb/arxiv-clustering-p2p
config: default
split: test
revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d
metrics:
- type: v_measure
value: 47.508221208908964
- task:
type: Clustering
dataset:
name: MTEB ArxivClusteringS2S
type: mteb/arxiv-clustering-s2s
config: default
split: test
revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53
metrics:
- type: v_measure
value: 42.04668382560096
- task:
type: Reranking
dataset:
name: MTEB AskUbuntuDupQuestions
type: mteb/askubuntudupquestions-reranking
config: default
split: test
revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
metrics:
- type: map
value: 61.828759903716815
- type: mrr
value: 74.37343358395991
- task:
type: STS
dataset:
name: MTEB BIOSSES
type: mteb/biosses-sts
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cos_sim_pearson
value: 85.03673698773017
- type: cos_sim_spearman
value: 83.6470866785058
- type: euclidean_pearson
value: 82.64048673096565
- type: euclidean_spearman
value: 83.63142367101115
- type: manhattan_pearson
value: 82.71493099760228
- type: manhattan_spearman
value: 83.60491704294326
- task:
type: Classification
dataset:
name: MTEB Banking77Classification
type: mteb/banking77
config: default
split: test
revision: 0fd18e25b25c072e09e0d92ab615fda904d66300
metrics:
- type: accuracy
value: 86.73376623376623
- type: f1
value: 86.70294049278262
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringP2P
type: mteb/biorxiv-clustering-p2p
config: default
split: test
revision: 65b79d1d13f80053f67aca9498d9402c2d9f1f40
metrics:
- type: v_measure
value: 40.31923804167062
- task:
type: Clustering
dataset:
name: MTEB BiorxivClusteringS2S
type: mteb/biorxiv-clustering-s2s
config: default
split: test
revision: 258694dd0231531bc1fd9de6ceb52a0853c6d908
metrics:
- type: v_measure
value: 37.552547125348454
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval
type: mteb/cqadupstack-android
config: default
split: test
revision: f46a197baaae43b4f621051089b82a364682dfeb
metrics:
- type: map_at_1
value: 30.567
- type: map_at_10
value: 41.269
- type: map_at_100
value: 42.689
- type: map_at_1000
value: 42.84
- type: map_at_3
value: 37.567
- type: map_at_5
value: 39.706
- type: mrr_at_1
value: 37.053000000000004
- type: mrr_at_10
value: 46.900999999999996
- type: mrr_at_100
value: 47.662
- type: mrr_at_1000
value: 47.713
- type: mrr_at_3
value: 43.801
- type: mrr_at_5
value: 45.689
- type: ndcg_at_1
value: 37.053000000000004
- type: ndcg_at_10
value: 47.73
- type: ndcg_at_100
value: 53.128
- type: ndcg_at_1000
value: 55.300000000000004
- type: ndcg_at_3
value: 42.046
- type: ndcg_at_5
value: 44.782
- type: precision_at_1
value: 37.053000000000004
- type: precision_at_10
value: 9.142
- type: precision_at_100
value: 1.485
- type: precision_at_1000
value: 0.197
- type: precision_at_3
value: 20.076
- type: precision_at_5
value: 14.535
- type: recall_at_1
value: 30.567
- type: recall_at_10
value: 60.602999999999994
- type: recall_at_100
value: 83.22800000000001
- type: recall_at_1000
value: 96.696
- type: recall_at_3
value: 44.336999999999996
- type: recall_at_5
value: 51.949
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackEnglishRetrieval
type: mteb/cqadupstack-english
config: default
split: test
revision: ad9991cb51e31e31e430383c75ffb2885547b5f0
metrics:
- type: map_at_1
value: 28.538000000000004
- type: map_at_10
value: 38.757999999999996
- type: map_at_100
value: 40.129
- type: map_at_1000
value: 40.262
- type: map_at_3
value: 35.866
- type: map_at_5
value: 37.417
- type: mrr_at_1
value: 36.051
- type: mrr_at_10
value: 44.868
- type: mrr_at_100
value: 45.568999999999996
- type: mrr_at_1000
value: 45.615
- type: mrr_at_3
value: 42.558
- type: mrr_at_5
value: 43.883
- type: ndcg_at_1
value: 36.051
- type: ndcg_at_10
value: 44.584
- type: ndcg_at_100
value: 49.356
- type: ndcg_at_1000
value: 51.39
- type: ndcg_at_3
value: 40.389
- type: ndcg_at_5
value: 42.14
- type: precision_at_1
value: 36.051
- type: precision_at_10
value: 8.446
- type: precision_at_100
value: 1.411
- type: precision_at_1000
value: 0.19
- type: precision_at_3
value: 19.639
- type: precision_at_5
value: 13.796
- type: recall_at_1
value: 28.538000000000004
- type: recall_at_10
value: 54.99000000000001
- type: recall_at_100
value: 75.098
- type: recall_at_1000
value: 87.848
- type: recall_at_3
value: 42.236000000000004
- type: recall_at_5
value: 47.377
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGamingRetrieval
type: mteb/cqadupstack-gaming
config: default
split: test
revision: 4885aa143210c98657558c04aaf3dc47cfb54340
metrics:
- type: map_at_1
value: 37.188
- type: map_at_10
value: 50.861000000000004
- type: map_at_100
value: 51.917
- type: map_at_1000
value: 51.964999999999996
- type: map_at_3
value: 47.144000000000005
- type: map_at_5
value: 49.417
- type: mrr_at_1
value: 42.571
- type: mrr_at_10
value: 54.086999999999996
- type: mrr_at_100
value: 54.739000000000004
- type: mrr_at_1000
value: 54.762
- type: mrr_at_3
value: 51.285000000000004
- type: mrr_at_5
value: 53.0
- type: ndcg_at_1
value: 42.571
- type: ndcg_at_10
value: 57.282
- type: ndcg_at_100
value: 61.477000000000004
- type: ndcg_at_1000
value: 62.426
- type: ndcg_at_3
value: 51.0
- type: ndcg_at_5
value: 54.346000000000004
- type: precision_at_1
value: 42.571
- type: precision_at_10
value: 9.467
- type: precision_at_100
value: 1.2550000000000001
- type: precision_at_1000
value: 0.13799999999999998
- type: precision_at_3
value: 23.114
- type: precision_at_5
value: 16.250999999999998
- type: recall_at_1
value: 37.188
- type: recall_at_10
value: 73.068
- type: recall_at_100
value: 91.203
- type: recall_at_1000
value: 97.916
- type: recall_at_3
value: 56.552
- type: recall_at_5
value: 64.567
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGisRetrieval
type: mteb/cqadupstack-gis
config: default
split: test
revision: 5003b3064772da1887988e05400cf3806fe491f2
metrics:
- type: map_at_1
value: 25.041000000000004
- type: map_at_10
value: 33.86
- type: map_at_100
value: 34.988
- type: map_at_1000
value: 35.064
- type: map_at_3
value: 31.049
- type: map_at_5
value: 32.845
- type: mrr_at_1
value: 26.893
- type: mrr_at_10
value: 35.594
- type: mrr_at_100
value: 36.617
- type: mrr_at_1000
value: 36.671
- type: mrr_at_3
value: 33.051
- type: mrr_at_5
value: 34.61
- type: ndcg_at_1
value: 26.893
- type: ndcg_at_10
value: 38.674
- type: ndcg_at_100
value: 44.178
- type: ndcg_at_1000
value: 46.089999999999996
- type: ndcg_at_3
value: 33.485
- type: ndcg_at_5
value: 36.402
- type: precision_at_1
value: 26.893
- type: precision_at_10
value: 5.989
- type: precision_at_100
value: 0.918
- type: precision_at_1000
value: 0.11100000000000002
- type: precision_at_3
value: 14.2
- type: precision_at_5
value: 10.26
- type: recall_at_1
value: 25.041000000000004
- type: recall_at_10
value: 51.666000000000004
- type: recall_at_100
value: 76.896
- type: recall_at_1000
value: 91.243
- type: recall_at_3
value: 38.035999999999994
- type: recall_at_5
value: 44.999
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackMathematicaRetrieval
type: mteb/cqadupstack-mathematica
config: default
split: test
revision: 90fceea13679c63fe563ded68f3b6f06e50061de
metrics:
- type: map_at_1
value: 15.909999999999998
- type: map_at_10
value: 23.901
- type: map_at_100
value: 25.165
- type: map_at_1000
value: 25.291000000000004
- type: map_at_3
value: 21.356
- type: map_at_5
value: 22.816
- type: mrr_at_1
value: 20.025000000000002
- type: mrr_at_10
value: 28.382
- type: mrr_at_100
value: 29.465000000000003
- type: mrr_at_1000
value: 29.535
- type: mrr_at_3
value: 25.933
- type: mrr_at_5
value: 27.332
- type: ndcg_at_1
value: 20.025000000000002
- type: ndcg_at_10
value: 29.099000000000004
- type: ndcg_at_100
value: 35.127
- type: ndcg_at_1000
value: 38.096000000000004
- type: ndcg_at_3
value: 24.464
- type: ndcg_at_5
value: 26.709
- type: precision_at_1
value: 20.025000000000002
- type: precision_at_10
value: 5.398
- type: precision_at_100
value: 0.9690000000000001
- type: precision_at_1000
value: 0.13699999999999998
- type: precision_at_3
value: 11.774
- type: precision_at_5
value: 8.632
- type: recall_at_1
value: 15.909999999999998
- type: recall_at_10
value: 40.672000000000004
- type: recall_at_100
value: 66.855
- type: recall_at_1000
value: 87.922
- type: recall_at_3
value: 28.069
- type: recall_at_5
value: 33.812
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackPhysicsRetrieval
type: mteb/cqadupstack-physics
config: default
split: test
revision: 79531abbd1fb92d06c6d6315a0cbbbf5bb247ea4
metrics:
- type: map_at_1
value: 30.175
- type: map_at_10
value: 41.36
- type: map_at_100
value: 42.701
- type: map_at_1000
value: 42.817
- type: map_at_3
value: 37.931
- type: map_at_5
value: 39.943
- type: mrr_at_1
value: 35.611
- type: mrr_at_10
value: 46.346
- type: mrr_at_100
value: 47.160000000000004
- type: mrr_at_1000
value: 47.203
- type: mrr_at_3
value: 43.712
- type: mrr_at_5
value: 45.367000000000004
- type: ndcg_at_1
value: 35.611
- type: ndcg_at_10
value: 47.532000000000004
- type: ndcg_at_100
value: 53.003
- type: ndcg_at_1000
value: 55.007
- type: ndcg_at_3
value: 42.043
- type: ndcg_at_5
value: 44.86
- type: precision_at_1
value: 35.611
- type: precision_at_10
value: 8.624
- type: precision_at_100
value: 1.332
- type: precision_at_1000
value: 0.169
- type: precision_at_3
value: 20.083000000000002
- type: precision_at_5
value: 14.437
- type: recall_at_1
value: 30.175
- type: recall_at_10
value: 60.5
- type: recall_at_100
value: 83.399
- type: recall_at_1000
value: 96.255
- type: recall_at_3
value: 45.448
- type: recall_at_5
value: 52.432
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackProgrammersRetrieval
type: mteb/cqadupstack-programmers
config: default
split: test
revision: 6184bc1440d2dbc7612be22b50686b8826d22b32
metrics:
- type: map_at_1
value: 22.467000000000002
- type: map_at_10
value: 33.812999999999995
- type: map_at_100
value: 35.248000000000005
- type: map_at_1000
value: 35.359
- type: map_at_3
value: 30.316
- type: map_at_5
value: 32.233000000000004
- type: mrr_at_1
value: 28.310999999999996
- type: mrr_at_10
value: 38.979
- type: mrr_at_100
value: 39.937
- type: mrr_at_1000
value: 39.989999999999995
- type: mrr_at_3
value: 36.244
- type: mrr_at_5
value: 37.871
- type: ndcg_at_1
value: 28.310999999999996
- type: ndcg_at_10
value: 40.282000000000004
- type: ndcg_at_100
value: 46.22
- type: ndcg_at_1000
value: 48.507
- type: ndcg_at_3
value: 34.596
- type: ndcg_at_5
value: 37.267
- type: precision_at_1
value: 28.310999999999996
- type: precision_at_10
value: 7.831
- type: precision_at_100
value: 1.257
- type: precision_at_1000
value: 0.164
- type: precision_at_3
value: 17.275
- type: precision_at_5
value: 12.556999999999999
- type: recall_at_1
value: 22.467000000000002
- type: recall_at_10
value: 54.14099999999999
- type: recall_at_100
value: 79.593
- type: recall_at_1000
value: 95.063
- type: recall_at_3
value: 38.539
- type: recall_at_5
value: 45.403
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackRetrieval
type: mteb/cqadupstack
config: default
split: test
revision: 4ffe81d471b1924886b33c7567bfb200e9eec5c4
metrics:
- type: map_at_1
value: 24.18591666666667
- type: map_at_10
value: 33.84258333333333
- type: map_at_100
value: 35.11391666666666
- type: map_at_1000
value: 35.23258333333333
- type: map_at_3
value: 30.764249999999997
- type: map_at_5
value: 32.52333333333334
- type: mrr_at_1
value: 28.54733333333333
- type: mrr_at_10
value: 37.81725
- type: mrr_at_100
value: 38.716499999999996
- type: mrr_at_1000
value: 38.77458333333333
- type: mrr_at_3
value: 35.157833333333336
- type: mrr_at_5
value: 36.69816666666667
- type: ndcg_at_1
value: 28.54733333333333
- type: ndcg_at_10
value: 39.51508333333334
- type: ndcg_at_100
value: 44.95316666666666
- type: ndcg_at_1000
value: 47.257083333333334
- type: ndcg_at_3
value: 34.205833333333324
- type: ndcg_at_5
value: 36.78266666666667
- type: precision_at_1
value: 28.54733333333333
- type: precision_at_10
value: 7.082583333333334
- type: precision_at_100
value: 1.1590833333333332
- type: precision_at_1000
value: 0.15516666666666662
- type: precision_at_3
value: 15.908750000000001
- type: precision_at_5
value: 11.505416666666669
- type: recall_at_1
value: 24.18591666666667
- type: recall_at_10
value: 52.38758333333333
- type: recall_at_100
value: 76.13666666666667
- type: recall_at_1000
value: 91.99066666666667
- type: recall_at_3
value: 37.78333333333334
- type: recall_at_5
value: 44.30141666666666
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackStatsRetrieval
type: mteb/cqadupstack-stats
config: default
split: test
revision: 65ac3a16b8e91f9cee4c9828cc7c335575432a2a
metrics:
- type: map_at_1
value: 21.975
- type: map_at_10
value: 29.781000000000002
- type: map_at_100
value: 30.847
- type: map_at_1000
value: 30.94
- type: map_at_3
value: 27.167
- type: map_at_5
value: 28.633999999999997
- type: mrr_at_1
value: 24.387
- type: mrr_at_10
value: 32.476
- type: mrr_at_100
value: 33.337
- type: mrr_at_1000
value: 33.403
- type: mrr_at_3
value: 29.881999999999998
- type: mrr_at_5
value: 31.339
- type: ndcg_at_1
value: 24.387
- type: ndcg_at_10
value: 34.596
- type: ndcg_at_100
value: 39.635
- type: ndcg_at_1000
value: 42.079
- type: ndcg_at_3
value: 29.516
- type: ndcg_at_5
value: 31.959
- type: precision_at_1
value: 24.387
- type: precision_at_10
value: 5.6129999999999995
- type: precision_at_100
value: 0.8909999999999999
- type: precision_at_1000
value: 0.117
- type: precision_at_3
value: 12.73
- type: precision_at_5
value: 9.171999999999999
- type: recall_at_1
value: 21.975
- type: recall_at_10
value: 46.826
- type: recall_at_100
value: 69.554
- type: recall_at_1000
value: 87.749
- type: recall_at_3
value: 33.016
- type: recall_at_5
value: 38.97
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackTexRetrieval
type: mteb/cqadupstack-tex
config: default
split: test
revision: 46989137a86843e03a6195de44b09deda022eec7
metrics:
- type: map_at_1
value: 15.614
- type: map_at_10
value: 22.927
- type: map_at_100
value: 24.185000000000002
- type: map_at_1000
value: 24.319
- type: map_at_3
value: 20.596
- type: map_at_5
value: 21.854000000000003
- type: mrr_at_1
value: 18.858
- type: mrr_at_10
value: 26.535999999999998
- type: mrr_at_100
value: 27.582
- type: mrr_at_1000
value: 27.665
- type: mrr_at_3
value: 24.295
- type: mrr_at_5
value: 25.532
- type: ndcg_at_1
value: 18.858
- type: ndcg_at_10
value: 27.583000000000002
- type: ndcg_at_100
value: 33.635
- type: ndcg_at_1000
value: 36.647
- type: ndcg_at_3
value: 23.348
- type: ndcg_at_5
value: 25.257
- type: precision_at_1
value: 18.858
- type: precision_at_10
value: 5.158
- type: precision_at_100
value: 0.964
- type: precision_at_1000
value: 0.13999999999999999
- type: precision_at_3
value: 11.092
- type: precision_at_5
value: 8.1
- type: recall_at_1
value: 15.614
- type: recall_at_10
value: 37.916
- type: recall_at_100
value: 65.205
- type: recall_at_1000
value: 86.453
- type: recall_at_3
value: 26.137
- type: recall_at_5
value: 31.087999999999997
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackUnixRetrieval
type: mteb/cqadupstack-unix
config: default
split: test
revision: 6c6430d3a6d36f8d2a829195bc5dc94d7e063e53
metrics:
- type: map_at_1
value: 23.078000000000003
- type: map_at_10
value: 31.941999999999997
- type: map_at_100
value: 33.196999999999996
- type: map_at_1000
value: 33.303
- type: map_at_3
value: 28.927000000000003
- type: map_at_5
value: 30.707
- type: mrr_at_1
value: 26.866
- type: mrr_at_10
value: 35.557
- type: mrr_at_100
value: 36.569
- type: mrr_at_1000
value: 36.632
- type: mrr_at_3
value: 32.897999999999996
- type: mrr_at_5
value: 34.437
- type: ndcg_at_1
value: 26.866
- type: ndcg_at_10
value: 37.372
- type: ndcg_at_100
value: 43.248
- type: ndcg_at_1000
value: 45.632
- type: ndcg_at_3
value: 31.852999999999998
- type: ndcg_at_5
value: 34.582
- type: precision_at_1
value: 26.866
- type: precision_at_10
value: 6.511
- type: precision_at_100
value: 1.078
- type: precision_at_1000
value: 0.13899999999999998
- type: precision_at_3
value: 14.582999999999998
- type: precision_at_5
value: 10.634
- type: recall_at_1
value: 23.078000000000003
- type: recall_at_10
value: 50.334
- type: recall_at_100
value: 75.787
- type: recall_at_1000
value: 92.485
- type: recall_at_3
value: 35.386
- type: recall_at_5
value: 42.225
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWebmastersRetrieval
type: mteb/cqadupstack-webmasters
config: default
split: test
revision: 160c094312a0e1facb97e55eeddb698c0abe3571
metrics:
- type: map_at_1
value: 22.203999999999997
- type: map_at_10
value: 31.276
- type: map_at_100
value: 32.844
- type: map_at_1000
value: 33.062999999999995
- type: map_at_3
value: 27.733999999999998
- type: map_at_5
value: 29.64
- type: mrr_at_1
value: 27.272999999999996
- type: mrr_at_10
value: 36.083
- type: mrr_at_100
value: 37.008
- type: mrr_at_1000
value: 37.076
- type: mrr_at_3
value: 33.004
- type: mrr_at_5
value: 34.664
- type: ndcg_at_1
value: 27.272999999999996
- type: ndcg_at_10
value: 37.763000000000005
- type: ndcg_at_100
value: 43.566
- type: ndcg_at_1000
value: 46.356
- type: ndcg_at_3
value: 31.673000000000002
- type: ndcg_at_5
value: 34.501
- type: precision_at_1
value: 27.272999999999996
- type: precision_at_10
value: 7.470000000000001
- type: precision_at_100
value: 1.502
- type: precision_at_1000
value: 0.24
- type: precision_at_3
value: 14.756
- type: precision_at_5
value: 11.225
- type: recall_at_1
value: 22.203999999999997
- type: recall_at_10
value: 51.437999999999995
- type: recall_at_100
value: 76.845
- type: recall_at_1000
value: 94.38600000000001
- type: recall_at_3
value: 34.258
- type: recall_at_5
value: 41.512
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWordpressRetrieval
type: mteb/cqadupstack-wordpress
config: default
split: test
revision: 4ffe81d471b1924886b33c7567bfb200e9eec5c4
metrics:
- type: map_at_1
value: 17.474
- type: map_at_10
value: 26.362999999999996
- type: map_at_100
value: 27.456999999999997
- type: map_at_1000
value: 27.567999999999998
- type: map_at_3
value: 23.518
- type: map_at_5
value: 25.068
- type: mrr_at_1
value: 18.669
- type: mrr_at_10
value: 27.998
- type: mrr_at_100
value: 28.953
- type: mrr_at_1000
value: 29.03
- type: mrr_at_3
value: 25.230999999999998
- type: mrr_at_5
value: 26.654
- type: ndcg_at_1
value: 18.669
- type: ndcg_at_10
value: 31.684
- type: ndcg_at_100
value: 36.864999999999995
- type: ndcg_at_1000
value: 39.555
- type: ndcg_at_3
value: 26.057000000000002
- type: ndcg_at_5
value: 28.587
- type: precision_at_1
value: 18.669
- type: precision_at_10
value: 5.3420000000000005
- type: precision_at_100
value: 0.847
- type: precision_at_1000
value: 0.12
- type: precision_at_3
value: 11.583
- type: precision_at_5
value: 8.466
- type: recall_at_1
value: 17.474
- type: recall_at_10
value: 46.497
- type: recall_at_100
value: 69.977
- type: recall_at_1000
value: 89.872
- type: recall_at_3
value: 31.385999999999996
- type: recall_at_5
value: 37.283
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER
type: mteb/climate-fever
config: default
split: test
revision: 47f2ac6acb640fc46020b02a5b59fdda04d39380
metrics:
- type: map_at_1
value: 17.173
- type: map_at_10
value: 30.407
- type: map_at_100
value: 32.528
- type: map_at_1000
value: 32.698
- type: map_at_3
value: 25.523
- type: map_at_5
value: 28.038
- type: mrr_at_1
value: 38.958
- type: mrr_at_10
value: 51.515
- type: mrr_at_100
value: 52.214000000000006
- type: mrr_at_1000
value: 52.237
- type: mrr_at_3
value: 48.502
- type: mrr_at_5
value: 50.251000000000005
- type: ndcg_at_1
value: 38.958
- type: ndcg_at_10
value: 40.355000000000004
- type: ndcg_at_100
value: 47.68
- type: ndcg_at_1000
value: 50.370000000000005
- type: ndcg_at_3
value: 33.946
- type: ndcg_at_5
value: 36.057
- type: precision_at_1
value: 38.958
- type: precision_at_10
value: 12.508
- type: precision_at_100
value: 2.054
- type: precision_at_1000
value: 0.256
- type: precision_at_3
value: 25.581
- type: precision_at_5
value: 19.256999999999998
- type: recall_at_1
value: 17.173
- type: recall_at_10
value: 46.967
- type: recall_at_100
value: 71.47200000000001
- type: recall_at_1000
value: 86.238
- type: recall_at_3
value: 30.961
- type: recall_at_5
value: 37.539
- task:
type: Retrieval
dataset:
name: MTEB DBPedia
type: mteb/dbpedia
config: default
split: test
revision: c0f706b76e590d620bd6618b3ca8efdd34e2d659
metrics:
- type: map_at_1
value: 8.999
- type: map_at_10
value: 18.989
- type: map_at_100
value: 26.133
- type: map_at_1000
value: 27.666
- type: map_at_3
value: 13.918
- type: map_at_5
value: 16.473
- type: mrr_at_1
value: 66.25
- type: mrr_at_10
value: 74.161
- type: mrr_at_100
value: 74.516
- type: mrr_at_1000
value: 74.524
- type: mrr_at_3
value: 72.875
- type: mrr_at_5
value: 73.613
- type: ndcg_at_1
value: 54.37499999999999
- type: ndcg_at_10
value: 39.902
- type: ndcg_at_100
value: 44.212
- type: ndcg_at_1000
value: 51.62
- type: ndcg_at_3
value: 45.193
- type: ndcg_at_5
value: 42.541000000000004
- type: precision_at_1
value: 66.25
- type: precision_at_10
value: 30.425
- type: precision_at_100
value: 9.754999999999999
- type: precision_at_1000
value: 2.043
- type: precision_at_3
value: 48.25
- type: precision_at_5
value: 40.65
- type: recall_at_1
value: 8.999
- type: recall_at_10
value: 24.133
- type: recall_at_100
value: 49.138999999999996
- type: recall_at_1000
value: 72.639
- type: recall_at_3
value: 15.287999999999998
- type: recall_at_5
value: 19.415
- task:
type: Classification
dataset:
name: MTEB EmotionClassification
type: mteb/emotion
config: default
split: test
revision: 4f58c6b202a23cf9a4da393831edf4f9183cad37
metrics:
- type: accuracy
value: 46.38999999999999
- type: f1
value: 41.444205512055234
- task:
type: Retrieval
dataset:
name: MTEB FEVER
type: mteb/fever
config: default
split: test
revision: bea83ef9e8fb933d90a2f1d5515737465d613e12
metrics:
- type: map_at_1
value: 87.35000000000001
- type: map_at_10
value: 92.837
- type: map_at_100
value: 92.996
- type: map_at_1000
value: 93.006
- type: map_at_3
value: 92.187
- type: map_at_5
value: 92.595
- type: mrr_at_1
value: 93.864
- type: mrr_at_10
value: 96.723
- type: mrr_at_100
value: 96.72500000000001
- type: mrr_at_1000
value: 96.72500000000001
- type: mrr_at_3
value: 96.64
- type: mrr_at_5
value: 96.71499999999999
- type: ndcg_at_1
value: 93.864
- type: ndcg_at_10
value: 94.813
- type: ndcg_at_100
value: 95.243
- type: ndcg_at_1000
value: 95.38600000000001
- type: ndcg_at_3
value: 94.196
- type: ndcg_at_5
value: 94.521
- type: precision_at_1
value: 93.864
- type: precision_at_10
value: 10.951
- type: precision_at_100
value: 1.1400000000000001
- type: precision_at_1000
value: 0.117
- type: precision_at_3
value: 35.114000000000004
- type: precision_at_5
value: 21.476
- type: recall_at_1
value: 87.35000000000001
- type: recall_at_10
value: 96.941
- type: recall_at_100
value: 98.397
- type: recall_at_1000
value: 99.21600000000001
- type: recall_at_3
value: 95.149
- type: recall_at_5
value: 96.131
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018
type: mteb/fiqa
config: default
split: test
revision: 27a168819829fe9bcd655c2df245fb19452e8e06
metrics:
- type: map_at_1
value: 24.476
- type: map_at_10
value: 40.11
- type: map_at_100
value: 42.229
- type: map_at_1000
value: 42.378
- type: map_at_3
value: 34.512
- type: map_at_5
value: 38.037
- type: mrr_at_1
value: 47.839999999999996
- type: mrr_at_10
value: 57.053
- type: mrr_at_100
value: 57.772
- type: mrr_at_1000
value: 57.799
- type: mrr_at_3
value: 54.552
- type: mrr_at_5
value: 56.011
- type: ndcg_at_1
value: 47.839999999999996
- type: ndcg_at_10
value: 48.650999999999996
- type: ndcg_at_100
value: 55.681000000000004
- type: ndcg_at_1000
value: 57.979
- type: ndcg_at_3
value: 43.923
- type: ndcg_at_5
value: 46.037
- type: precision_at_1
value: 47.839999999999996
- type: precision_at_10
value: 13.395000000000001
- type: precision_at_100
value: 2.0660000000000003
- type: precision_at_1000
value: 0.248
- type: precision_at_3
value: 29.064
- type: precision_at_5
value: 22.006
- type: recall_at_1
value: 24.476
- type: recall_at_10
value: 56.216
- type: recall_at_100
value: 81.798
- type: recall_at_1000
value: 95.48299999999999
- type: recall_at_3
value: 39.357
- type: recall_at_5
value: 47.802
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA
type: mteb/hotpotqa
config: default
split: test
revision: ab518f4d6fcca38d87c25209f94beba119d02014
metrics:
- type: map_at_1
value: 42.728
- type: map_at_10
value: 57.737
- type: map_at_100
value: 58.531
- type: map_at_1000
value: 58.594
- type: map_at_3
value: 54.869
- type: map_at_5
value: 56.55
- type: mrr_at_1
value: 85.456
- type: mrr_at_10
value: 90.062
- type: mrr_at_100
value: 90.159
- type: mrr_at_1000
value: 90.16
- type: mrr_at_3
value: 89.37899999999999
- type: mrr_at_5
value: 89.81
- type: ndcg_at_1
value: 85.456
- type: ndcg_at_10
value: 67.755
- type: ndcg_at_100
value: 70.341
- type: ndcg_at_1000
value: 71.538
- type: ndcg_at_3
value: 63.735
- type: ndcg_at_5
value: 65.823
- type: precision_at_1
value: 85.456
- type: precision_at_10
value: 13.450000000000001
- type: precision_at_100
value: 1.545
- type: precision_at_1000
value: 0.16999999999999998
- type: precision_at_3
value: 38.861000000000004
- type: precision_at_5
value: 24.964
- type: recall_at_1
value: 42.728
- type: recall_at_10
value: 67.252
- type: recall_at_100
value: 77.265
- type: recall_at_1000
value: 85.246
- type: recall_at_3
value: 58.292
- type: recall_at_5
value: 62.41100000000001
- task:
type: Classification
dataset:
name: MTEB ImdbClassification
type: mteb/imdb
config: default
split: test
revision: 3d86128a09e091d6018b6d26cad27f2739fc2db7
metrics:
- type: accuracy
value: 87.4836
- type: ap
value: 82.29552224030336
- type: f1
value: 87.42791432227448
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO
type: mteb/msmarco
config: default
split: dev
revision: c5a29a104738b98a9e76336939199e264163d4a0
metrics:
- type: map_at_1
value: 23.015
- type: map_at_10
value: 35.621
- type: map_at_100
value: 36.809
- type: map_at_1000
value: 36.853
- type: map_at_3
value: 31.832
- type: map_at_5
value: 34.006
- type: mrr_at_1
value: 23.738999999999997
- type: mrr_at_10
value: 36.309999999999995
- type: mrr_at_100
value: 37.422
- type: mrr_at_1000
value: 37.461
- type: mrr_at_3
value: 32.592999999999996
- type: mrr_at_5
value: 34.736
- type: ndcg_at_1
value: 23.724999999999998
- type: ndcg_at_10
value: 42.617
- type: ndcg_at_100
value: 48.217999999999996
- type: ndcg_at_1000
value: 49.309
- type: ndcg_at_3
value: 34.905
- type: ndcg_at_5
value: 38.769
- type: precision_at_1
value: 23.724999999999998
- type: precision_at_10
value: 6.689
- type: precision_at_100
value: 0.9480000000000001
- type: precision_at_1000
value: 0.104
- type: precision_at_3
value: 14.89
- type: precision_at_5
value: 10.897
- type: recall_at_1
value: 23.015
- type: recall_at_10
value: 64.041
- type: recall_at_100
value: 89.724
- type: recall_at_1000
value: 98.00999999999999
- type: recall_at_3
value: 43.064
- type: recall_at_5
value: 52.31099999999999
- task:
type: Classification
dataset:
name: MTEB MTOPDomainClassification (en)
type: mteb/mtop_domain
config: en
split: test
revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf
metrics:
- type: accuracy
value: 96.49794801641588
- type: f1
value: 96.28931114498003
- task:
type: Classification
dataset:
name: MTEB MTOPIntentClassification (en)
type: mteb/mtop_intent
config: en
split: test
revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba
metrics:
- type: accuracy
value: 82.81121751025992
- type: f1
value: 63.18740125901853
- task:
type: Classification
dataset:
name: MTEB MassiveIntentClassification (en)
type: mteb/amazon_massive_intent
config: en
split: test
revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7
metrics:
- type: accuracy
value: 77.66644250168123
- type: f1
value: 74.93211186867839
- task:
type: Classification
dataset:
name: MTEB MassiveScenarioClassification (en)
type: mteb/amazon_massive_scenario
config: en
split: test
revision: 7d571f92784cd94a019292a1f45445077d0ef634
metrics:
- type: accuracy
value: 81.77202420981843
- type: f1
value: 81.63681969283554
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringP2P
type: mteb/medrxiv-clustering-p2p
config: default
split: test
revision: e7a26af6f3ae46b30dde8737f02c07b1505bcc73
metrics:
- type: v_measure
value: 34.596687684870645
- task:
type: Clustering
dataset:
name: MTEB MedrxivClusteringS2S
type: mteb/medrxiv-clustering-s2s
config: default
split: test
revision: 35191c8c0dca72d8ff3efcd72aa802307d469663
metrics:
- type: v_measure
value: 32.26965660101405
- task:
type: Reranking
dataset:
name: MTEB MindSmallReranking
type: mteb/mind_small
config: default
split: test
revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69
metrics:
- type: map
value: 31.33619694846802
- type: mrr
value: 32.53719657720334
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus
type: mteb/nfcorpus
config: default
split: test
revision: ec0fa4fe99da2ff19ca1214b7966684033a58814
metrics:
- type: map_at_1
value: 6.0729999999999995
- type: map_at_10
value: 13.245999999999999
- type: map_at_100
value: 16.747999999999998
- type: map_at_1000
value: 18.163
- type: map_at_3
value: 10.064
- type: map_at_5
value: 11.513
- type: mrr_at_1
value: 49.536
- type: mrr_at_10
value: 58.092
- type: mrr_at_100
value: 58.752
- type: mrr_at_1000
value: 58.78
- type: mrr_at_3
value: 56.398
- type: mrr_at_5
value: 57.389
- type: ndcg_at_1
value: 47.059
- type: ndcg_at_10
value: 35.881
- type: ndcg_at_100
value: 32.751999999999995
- type: ndcg_at_1000
value: 41.498000000000005
- type: ndcg_at_3
value: 42.518
- type: ndcg_at_5
value: 39.550999999999995
- type: precision_at_1
value: 49.536
- type: precision_at_10
value: 26.316
- type: precision_at_100
value: 8.084
- type: precision_at_1000
value: 2.081
- type: precision_at_3
value: 39.938
- type: precision_at_5
value: 34.056
- type: recall_at_1
value: 6.0729999999999995
- type: recall_at_10
value: 16.593
- type: recall_at_100
value: 32.883
- type: recall_at_1000
value: 64.654
- type: recall_at_3
value: 11.174000000000001
- type: recall_at_5
value: 13.528
- task:
type: Retrieval
dataset:
name: MTEB NQ
type: mteb/nq
config: default
split: test
revision: b774495ed302d8c44a3a7ea25c90dbce03968f31
metrics:
- type: map_at_1
value: 30.043
- type: map_at_10
value: 45.318999999999996
- type: map_at_100
value: 46.381
- type: map_at_1000
value: 46.412
- type: map_at_3
value: 40.941
- type: map_at_5
value: 43.662
- type: mrr_at_1
value: 33.98
- type: mrr_at_10
value: 47.870000000000005
- type: mrr_at_100
value: 48.681999999999995
- type: mrr_at_1000
value: 48.703
- type: mrr_at_3
value: 44.341
- type: mrr_at_5
value: 46.547
- type: ndcg_at_1
value: 33.98
- type: ndcg_at_10
value: 52.957
- type: ndcg_at_100
value: 57.434
- type: ndcg_at_1000
value: 58.103
- type: ndcg_at_3
value: 44.896
- type: ndcg_at_5
value: 49.353
- type: precision_at_1
value: 33.98
- type: precision_at_10
value: 8.786
- type: precision_at_100
value: 1.1280000000000001
- type: precision_at_1000
value: 0.11900000000000001
- type: precision_at_3
value: 20.577
- type: precision_at_5
value: 14.942
- type: recall_at_1
value: 30.043
- type: recall_at_10
value: 73.593
- type: recall_at_100
value: 93.026
- type: recall_at_1000
value: 97.943
- type: recall_at_3
value: 52.955
- type: recall_at_5
value: 63.132
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval
type: mteb/quora
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 70.808
- type: map_at_10
value: 84.675
- type: map_at_100
value: 85.322
- type: map_at_1000
value: 85.33800000000001
- type: map_at_3
value: 81.68900000000001
- type: map_at_5
value: 83.543
- type: mrr_at_1
value: 81.5
- type: mrr_at_10
value: 87.59700000000001
- type: mrr_at_100
value: 87.705
- type: mrr_at_1000
value: 87.70599999999999
- type: mrr_at_3
value: 86.607
- type: mrr_at_5
value: 87.289
- type: ndcg_at_1
value: 81.51
- type: ndcg_at_10
value: 88.41799999999999
- type: ndcg_at_100
value: 89.644
- type: ndcg_at_1000
value: 89.725
- type: ndcg_at_3
value: 85.49900000000001
- type: ndcg_at_5
value: 87.078
- type: precision_at_1
value: 81.51
- type: precision_at_10
value: 13.438
- type: precision_at_100
value: 1.532
- type: precision_at_1000
value: 0.157
- type: precision_at_3
value: 37.363
- type: precision_at_5
value: 24.57
- type: recall_at_1
value: 70.808
- type: recall_at_10
value: 95.575
- type: recall_at_100
value: 99.667
- type: recall_at_1000
value: 99.98899999999999
- type: recall_at_3
value: 87.223
- type: recall_at_5
value: 91.682
- task:
type: Clustering
dataset:
name: MTEB RedditClustering
type: mteb/reddit-clustering
config: default
split: test
revision: 24640382cdbf8abc73003fb0fa6d111a705499eb
metrics:
- type: v_measure
value: 58.614831329137715
- task:
type: Clustering
dataset:
name: MTEB RedditClusteringP2P
type: mteb/reddit-clustering-p2p
config: default
split: test
revision: 282350215ef01743dc01b456c7f5241fa8937f16
metrics:
- type: v_measure
value: 66.86580408560826
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS
type: mteb/scidocs
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 5.093
- type: map_at_10
value: 13.014000000000001
- type: map_at_100
value: 15.412999999999998
- type: map_at_1000
value: 15.756999999999998
- type: map_at_3
value: 9.216000000000001
- type: map_at_5
value: 11.036999999999999
- type: mrr_at_1
value: 25.1
- type: mrr_at_10
value: 37.133
- type: mrr_at_100
value: 38.165
- type: mrr_at_1000
value: 38.198
- type: mrr_at_3
value: 33.217
- type: mrr_at_5
value: 35.732
- type: ndcg_at_1
value: 25.1
- type: ndcg_at_10
value: 21.918000000000003
- type: ndcg_at_100
value: 30.983
- type: ndcg_at_1000
value: 36.629
- type: ndcg_at_3
value: 20.544999999999998
- type: ndcg_at_5
value: 18.192
- type: precision_at_1
value: 25.1
- type: precision_at_10
value: 11.44
- type: precision_at_100
value: 2.459
- type: precision_at_1000
value: 0.381
- type: precision_at_3
value: 19.267
- type: precision_at_5
value: 16.16
- type: recall_at_1
value: 5.093
- type: recall_at_10
value: 23.215
- type: recall_at_100
value: 49.902
- type: recall_at_1000
value: 77.403
- type: recall_at_3
value: 11.733
- type: recall_at_5
value: 16.372999999999998
- task:
type: STS
dataset:
name: MTEB SICK-R
type: mteb/sickr-sts
config: default
split: test
revision: a6ea5a8cab320b040a23452cc28066d9beae2cee
metrics:
- type: cos_sim_pearson
value: 82.9365442977452
- type: cos_sim_spearman
value: 79.36960687383745
- type: euclidean_pearson
value: 79.6045204840714
- type: euclidean_spearman
value: 79.26382712751337
- type: manhattan_pearson
value: 79.4805084789529
- type: manhattan_spearman
value: 79.21847863209523
- task:
type: STS
dataset:
name: MTEB STS12
type: mteb/sts12-sts
config: default
split: test
revision: a0d554a64d88156834ff5ae9920b964011b16384
metrics:
- type: cos_sim_pearson
value: 83.27906192961453
- type: cos_sim_spearman
value: 74.38364712099211
- type: euclidean_pearson
value: 78.54358927241223
- type: euclidean_spearman
value: 74.22185560806376
- type: manhattan_pearson
value: 78.50904327377751
- type: manhattan_spearman
value: 74.2627500781748
- task:
type: STS
dataset:
name: MTEB STS13
type: mteb/sts13-sts
config: default
split: test
revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
metrics:
- type: cos_sim_pearson
value: 84.66863742649639
- type: cos_sim_spearman
value: 84.70630905216271
- type: euclidean_pearson
value: 84.64498334705334
- type: euclidean_spearman
value: 84.87204770690148
- type: manhattan_pearson
value: 84.65774227976077
- type: manhattan_spearman
value: 84.91251851797985
- task:
type: STS
dataset:
name: MTEB STS14
type: mteb/sts14-sts
config: default
split: test
revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
metrics:
- type: cos_sim_pearson
value: 83.1577763924467
- type: cos_sim_spearman
value: 80.10314039230198
- type: euclidean_pearson
value: 81.51346991046043
- type: euclidean_spearman
value: 80.08678485109435
- type: manhattan_pearson
value: 81.57058914661894
- type: manhattan_spearman
value: 80.1516230725106
- task:
type: STS
dataset:
name: MTEB STS15
type: mteb/sts15-sts
config: default
split: test
revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
metrics:
- type: cos_sim_pearson
value: 86.40310839662533
- type: cos_sim_spearman
value: 87.16293477217867
- type: euclidean_pearson
value: 86.50688711184775
- type: euclidean_spearman
value: 87.08651444923031
- type: manhattan_pearson
value: 86.54674677557857
- type: manhattan_spearman
value: 87.15079017870971
- task:
type: STS
dataset:
name: MTEB STS16
type: mteb/sts16-sts
config: default
split: test
revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
metrics:
- type: cos_sim_pearson
value: 84.32886275207817
- type: cos_sim_spearman
value: 85.0190460590732
- type: euclidean_pearson
value: 84.42553652784679
- type: euclidean_spearman
value: 85.20027364279328
- type: manhattan_pearson
value: 84.42926246281078
- type: manhattan_spearman
value: 85.20187419804306
- task:
type: STS
dataset:
name: MTEB STS17 (en-en)
type: mteb/sts17-crosslingual-sts
config: en-en
split: test
revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d
metrics:
- type: cos_sim_pearson
value: 90.76732216967812
- type: cos_sim_spearman
value: 90.63701653633909
- type: euclidean_pearson
value: 90.26678186114682
- type: euclidean_spearman
value: 90.67288073455427
- type: manhattan_pearson
value: 90.20772020584582
- type: manhattan_spearman
value: 90.60764863983702
- task:
type: STS
dataset:
name: MTEB STS22 (en)
type: mteb/sts22-crosslingual-sts
config: en
split: test
revision: eea2b4fe26a775864c896887d910b76a8098ad3f
metrics:
- type: cos_sim_pearson
value: 69.09280387698125
- type: cos_sim_spearman
value: 68.62743151172162
- type: euclidean_pearson
value: 69.89386398104689
- type: euclidean_spearman
value: 68.71191066733556
- type: manhattan_pearson
value: 69.92516500604872
- type: manhattan_spearman
value: 68.80452846992576
- task:
type: STS
dataset:
name: MTEB STSBenchmark
type: mteb/stsbenchmark-sts
config: default
split: test
revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831
metrics:
- type: cos_sim_pearson
value: 86.13178592019887
- type: cos_sim_spearman
value: 86.03947178806887
- type: euclidean_pearson
value: 85.87029414285313
- type: euclidean_spearman
value: 86.04960843306998
- type: manhattan_pearson
value: 85.92946858580146
- type: manhattan_spearman
value: 86.12575341860442
- task:
type: Reranking
dataset:
name: MTEB SciDocsRR
type: mteb/scidocs-reranking
config: default
split: test
revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab
metrics:
- type: map
value: 85.16657063002837
- type: mrr
value: 95.73671063867141
- task:
type: Retrieval
dataset:
name: MTEB SciFact
type: mteb/scifact
config: default
split: test
revision: 0228b52cf27578f30900b9e5271d331663a030d7
metrics:
- type: map_at_1
value: 63.510999999999996
- type: map_at_10
value: 72.76899999999999
- type: map_at_100
value: 73.303
- type: map_at_1000
value: 73.32499999999999
- type: map_at_3
value: 70.514
- type: map_at_5
value: 71.929
- type: mrr_at_1
value: 66.333
- type: mrr_at_10
value: 73.75
- type: mrr_at_100
value: 74.119
- type: mrr_at_1000
value: 74.138
- type: mrr_at_3
value: 72.222
- type: mrr_at_5
value: 73.122
- type: ndcg_at_1
value: 66.333
- type: ndcg_at_10
value: 76.774
- type: ndcg_at_100
value: 78.78500000000001
- type: ndcg_at_1000
value: 79.254
- type: ndcg_at_3
value: 73.088
- type: ndcg_at_5
value: 75.002
- type: precision_at_1
value: 66.333
- type: precision_at_10
value: 9.833
- type: precision_at_100
value: 1.093
- type: precision_at_1000
value: 0.11299999999999999
- type: precision_at_3
value: 28.222
- type: precision_at_5
value: 18.333
- type: recall_at_1
value: 63.510999999999996
- type: recall_at_10
value: 87.98899999999999
- type: recall_at_100
value: 96.5
- type: recall_at_1000
value: 100.0
- type: recall_at_3
value: 77.86699999999999
- type: recall_at_5
value: 82.73899999999999
- task:
type: PairClassification
dataset:
name: MTEB SprintDuplicateQuestions
type: mteb/sprintduplicatequestions-pairclassification
config: default
split: test
revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
metrics:
- type: cos_sim_accuracy
value: 99.78514851485149
- type: cos_sim_ap
value: 94.94214383862038
- type: cos_sim_f1
value: 89.02255639097744
- type: cos_sim_precision
value: 89.2462311557789
- type: cos_sim_recall
value: 88.8
- type: dot_accuracy
value: 99.78217821782178
- type: dot_ap
value: 94.69965247836805
- type: dot_f1
value: 88.78695208970439
- type: dot_precision
value: 90.54054054054053
- type: dot_recall
value: 87.1
- type: euclidean_accuracy
value: 99.78118811881188
- type: euclidean_ap
value: 94.9865187695411
- type: euclidean_f1
value: 88.99950223992036
- type: euclidean_precision
value: 88.60257680872151
- type: euclidean_recall
value: 89.4
- type: manhattan_accuracy
value: 99.78811881188119
- type: manhattan_ap
value: 95.0021236766459
- type: manhattan_f1
value: 89.12071535022356
- type: manhattan_precision
value: 88.54886475814413
- type: manhattan_recall
value: 89.7
- type: max_accuracy
value: 99.78811881188119
- type: max_ap
value: 95.0021236766459
- type: max_f1
value: 89.12071535022356
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClustering
type: mteb/stackexchange-clustering
config: default
split: test
revision: 6cbc1f7b2bc0622f2e39d2c77fa502909748c259
metrics:
- type: v_measure
value: 68.93190546593995
- task:
type: Clustering
dataset:
name: MTEB StackExchangeClusteringP2P
type: mteb/stackexchange-clustering-p2p
config: default
split: test
revision: 815ca46b2622cec33ccafc3735d572c266efdb44
metrics:
- type: v_measure
value: 37.602808534760655
- task:
type: Reranking
dataset:
name: MTEB StackOverflowDupQuestions
type: mteb/stackoverflowdupquestions-reranking
config: default
split: test
revision: e185fbe320c72810689fc5848eb6114e1ef5ec69
metrics:
- type: map
value: 52.29214480978073
- type: mrr
value: 53.123169722434426
- task:
type: Summarization
dataset:
name: MTEB SummEval
type: mteb/summeval
config: default
split: test
revision: cda12ad7615edc362dbf25a00fdd61d3b1eaf93c
metrics:
- type: cos_sim_pearson
value: 30.967800769650022
- type: cos_sim_spearman
value: 31.168490040206926
- type: dot_pearson
value: 30.888603021128553
- type: dot_spearman
value: 31.028241262520385
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID
type: mteb/trec-covid
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 0.22300000000000003
- type: map_at_10
value: 1.781
- type: map_at_100
value: 9.905999999999999
- type: map_at_1000
value: 23.455000000000002
- type: map_at_3
value: 0.569
- type: map_at_5
value: 0.918
- type: mrr_at_1
value: 84.0
- type: mrr_at_10
value: 91.067
- type: mrr_at_100
value: 91.067
- type: mrr_at_1000
value: 91.067
- type: mrr_at_3
value: 90.667
- type: mrr_at_5
value: 91.067
- type: ndcg_at_1
value: 78.0
- type: ndcg_at_10
value: 73.13499999999999
- type: ndcg_at_100
value: 55.32
- type: ndcg_at_1000
value: 49.532
- type: ndcg_at_3
value: 73.715
- type: ndcg_at_5
value: 72.74199999999999
- type: precision_at_1
value: 84.0
- type: precision_at_10
value: 78.8
- type: precision_at_100
value: 56.32
- type: precision_at_1000
value: 21.504
- type: precision_at_3
value: 77.333
- type: precision_at_5
value: 78.0
- type: recall_at_1
value: 0.22300000000000003
- type: recall_at_10
value: 2.049
- type: recall_at_100
value: 13.553
- type: recall_at_1000
value: 46.367999999999995
- type: recall_at_3
value: 0.604
- type: recall_at_5
value: 1.015
- task:
type: Retrieval
dataset:
name: MTEB Touche2020
type: mteb/touche2020
config: default
split: test
revision: a34f9a33db75fa0cbb21bb5cfc3dae8dc8bec93f
metrics:
- type: map_at_1
value: 3.0380000000000003
- type: map_at_10
value: 10.188
- type: map_at_100
value: 16.395
- type: map_at_1000
value: 18.024
- type: map_at_3
value: 6.236
- type: map_at_5
value: 7.276000000000001
- type: mrr_at_1
value: 34.694
- type: mrr_at_10
value: 46.292
- type: mrr_at_100
value: 47.446
- type: mrr_at_1000
value: 47.446
- type: mrr_at_3
value: 41.156
- type: mrr_at_5
value: 44.32
- type: ndcg_at_1
value: 32.653
- type: ndcg_at_10
value: 25.219
- type: ndcg_at_100
value: 37.802
- type: ndcg_at_1000
value: 49.274
- type: ndcg_at_3
value: 28.605999999999998
- type: ndcg_at_5
value: 26.21
- type: precision_at_1
value: 34.694
- type: precision_at_10
value: 21.837
- type: precision_at_100
value: 7.776
- type: precision_at_1000
value: 1.522
- type: precision_at_3
value: 28.571
- type: precision_at_5
value: 25.306
- type: recall_at_1
value: 3.0380000000000003
- type: recall_at_10
value: 16.298000000000002
- type: recall_at_100
value: 48.712
- type: recall_at_1000
value: 83.16799999999999
- type: recall_at_3
value: 7.265000000000001
- type: recall_at_5
value: 9.551
- task:
type: Classification
dataset:
name: MTEB ToxicConversationsClassification
type: mteb/toxic_conversations_50k
config: default
split: test
revision: d7c0de2777da35d6aae2200a62c6e0e5af397c4c
metrics:
- type: accuracy
value: 83.978
- type: ap
value: 24.751887949330015
- type: f1
value: 66.8685134049279
- task:
type: Classification
dataset:
name: MTEB TweetSentimentExtractionClassification
type: mteb/tweet_sentiment_extraction
config: default
split: test
revision: d604517c81ca91fe16a244d1248fc021f9ecee7a
metrics:
- type: accuracy
value: 61.573288058856825
- type: f1
value: 61.973261751726604
- task:
type: Clustering
dataset:
name: MTEB TwentyNewsgroupsClustering
type: mteb/twentynewsgroups-clustering
config: default
split: test
revision: 6125ec4e24fa026cec8a478383ee943acfbd5449
metrics:
- type: v_measure
value: 48.75483298792469
- task:
type: PairClassification
dataset:
name: MTEB TwitterSemEval2015
type: mteb/twittersemeval2015-pairclassification
config: default
split: test
revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1
metrics:
- type: cos_sim_accuracy
value: 86.36824223639506
- type: cos_sim_ap
value: 75.53126388573047
- type: cos_sim_f1
value: 67.9912831688245
- type: cos_sim_precision
value: 66.11817501869858
- type: cos_sim_recall
value: 69.9736147757256
- type: dot_accuracy
value: 86.39804494248078
- type: dot_ap
value: 75.27598891718046
- type: dot_f1
value: 67.91146284159763
- type: dot_precision
value: 63.90505003490807
- type: dot_recall
value: 72.45382585751979
- type: euclidean_accuracy
value: 86.36228169517793
- type: euclidean_ap
value: 75.51438087434647
- type: euclidean_f1
value: 68.02370523061066
- type: euclidean_precision
value: 66.46525679758308
- type: euclidean_recall
value: 69.65699208443272
- type: manhattan_accuracy
value: 86.46361089586935
- type: manhattan_ap
value: 75.50800785730111
- type: manhattan_f1
value: 67.9220437187253
- type: manhattan_precision
value: 67.79705573080967
- type: manhattan_recall
value: 68.04749340369392
- type: max_accuracy
value: 86.46361089586935
- type: max_ap
value: 75.53126388573047
- type: max_f1
value: 68.02370523061066
- task:
type: PairClassification
dataset:
name: MTEB TwitterURLCorpus
type: mteb/twitterurlcorpus-pairclassification
config: default
split: test
revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf
metrics:
- type: cos_sim_accuracy
value: 88.80350836341057
- type: cos_sim_ap
value: 85.51101933260743
- type: cos_sim_f1
value: 77.9152271629704
- type: cos_sim_precision
value: 75.27815662910056
- type: cos_sim_recall
value: 80.74376347397599
- type: dot_accuracy
value: 88.84425815966158
- type: dot_ap
value: 85.49726945962519
- type: dot_f1
value: 77.94445269567801
- type: dot_precision
value: 75.27251864601261
- type: dot_recall
value: 80.81305820757623
- type: euclidean_accuracy
value: 88.80350836341057
- type: euclidean_ap
value: 85.4882880790211
- type: euclidean_f1
value: 77.87063284615103
- type: euclidean_precision
value: 74.61022927689595
- type: euclidean_recall
value: 81.42901139513397
- type: manhattan_accuracy
value: 88.7161873714441
- type: manhattan_ap
value: 85.45753871906821
- type: manhattan_f1
value: 77.8686401480111
- type: manhattan_precision
value: 74.95903683123174
- type: manhattan_recall
value: 81.01324299353249
- type: max_accuracy
value: 88.84425815966158
- type: max_ap
value: 85.51101933260743
- type: max_f1
value: 77.94445269567801
---
<!-- **English** | [中文](./README_zh.md) -->
# gte-base-en-v1.5
We introduce `gte-v1.5` series, upgraded `gte` embeddings that support the context length of up to **8192**, while further enhancing model performance.
The models are built upon the `transformer++` encoder [backbone](https://huggingface.co/Alibaba-NLP/new-impl) (BERT + RoPE + GLU).
The `gte-v1.5` series achieve state-of-the-art scores on the MTEB benchmark within the same model size category and prodvide competitive on the LoCo long-context retrieval tests (refer to [Evaluation](#evaluation)).
We also present the [`gte-Qwen1.5-7B-instruct`](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct),
a SOTA instruction-tuned multi-lingual embedding model that ranked 2nd in MTEB and 1st in C-MTEB.
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Institute for Intelligent Computing, Alibaba Group
- **Model type:** Text Embeddings
- **Paper:** [mGTE: Generalized Long-Context Text Representation and Reranking
Models for Multilingual Text Retrieval](https://arxiv.org/pdf/2407.19669)
<!-- - **Demo [optional]:** [More Information Needed] -->
### Model list
| Models | Language | Model Size | Max Seq. Length | Dimension | MTEB-en | LoCo |
|:-----: | :-----: |:-----: |:-----: |:-----: | :-----: | :-----: |
|[`gte-Qwen1.5-7B-instruct`](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct)| Multiple | 7720 | 32768 | 4096 | 67.34 | 87.57 |
|[`gte-large-en-v1.5`](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | English | 434 | 8192 | 1024 | 65.39 | 86.71 |
|[`gte-base-en-v1.5`](https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5) | English | 137 | 8192 | 768 | 64.11 | 87.44 |
## How to Get Started with the Model
Use the code below to get started with the model.
```python
# Requires transformers>=4.36.0
import torch.nn.functional as F
from transformers import AutoModel, AutoTokenizer
input_texts = [
"what is the capital of China?",
"how to implement quick sort in python?",
"Beijing",
"sorting algorithms"
]
model_path = 'Alibaba-NLP/gte-base-en-v1.5'
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModel.from_pretrained(model_path, trust_remote_code=True)
# Tokenize the input texts
batch_dict = tokenizer(input_texts, max_length=8192, padding=True, truncation=True, return_tensors='pt')
outputs = model(**batch_dict)
embeddings = outputs.last_hidden_state[:, 0]
# (Optionally) normalize embeddings
embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:1] @ embeddings[1:].T) * 100
print(scores.tolist())
```
**It is recommended to install xformers and enable unpadding for acceleration, refer to [enable-unpadding-and-xformers](https://huggingface.co/Alibaba-NLP/new-impl#recommendation-enable-unpadding-and-acceleration-with-xformers).**
Use with `sentence-transformers`:
```python
# Requires sentence_transformers>=2.7.0
from sentence_transformers import SentenceTransformer
from sentence_transformers.util import cos_sim
sentences = ['That is a happy person', 'That is a very happy person']
model = SentenceTransformer('Alibaba-NLP/gte-base-en-v1.5', trust_remote_code=True)
embeddings = model.encode(sentences)
print(cos_sim(embeddings[0], embeddings[1]))
```
Use with `transformers.js`:
```js
// npm i @xenova/transformers
import { pipeline, dot } from '@xenova/transformers';
// Create feature extraction pipeline
const extractor = await pipeline('feature-extraction', 'Alibaba-NLP/gte-base-en-v1.5', {
quantized: false, // Comment out this line to use the quantized version
});
// Generate sentence embeddings
const sentences = [
"what is the capital of China?",
"how to implement quick sort in python?",
"Beijing",
"sorting algorithms"
]
const output = await extractor(sentences, { normalize: true, pooling: 'cls' });
// Compute similarity scores
const [source_embeddings, ...document_embeddings ] = output.tolist();
const similarities = document_embeddings.map(x => 100 * dot(source_embeddings, x));
console.log(similarities); // [34.504930869007296, 64.03973265120138, 19.520042686034362]
```
Use with infinity:
[Infinity](https://github.com/michaelfeil/infinity) is a MIT licensed server for OpenAI-compatible deployment.
```
docker run --gpus all -v $PWD/data:/app/.cache -p "7997":"7997" \
michaelf34/infinity:0.0.68 \
v2 --model-id Alibaba-NLP/gte-base-en-v1.5 --revision "4c742dc2b781e4ab062a4a77f4f7cbad4bdee970" --dtype bfloat16 --batch-size 32 --device cuda --engine torch --port 7997
```
## Training Details
### Training Data
- Masked language modeling (MLM): `c4-en`
- Weak-supervised contrastive pre-training (CPT): [GTE](https://arxiv.org/pdf/2308.03281.pdf) pre-training data
- Supervised contrastive fine-tuning: [GTE](https://arxiv.org/pdf/2308.03281.pdf) fine-tuning data
### Training Procedure
To enable the backbone model to support a context length of 8192, we adopted a multi-stage training strategy.
The model first undergoes preliminary MLM pre-training on shorter lengths.
And then, we resample the data, reducing the proportion of short texts, and continue the MLM pre-training.
The entire training process is as follows:
- MLM-2048: lr 5e-4, mlm_probability 0.3, batch_size 4096, num_steps 70000, rope_base 10000
- [MLM-8192](https://huggingface.co/Alibaba-NLP/gte-en-mlm-base): lr 5e-5, mlm_probability 0.3, batch_size 1024, num_steps 20000, rope_base 500000
- CPT: max_len 512, lr 2e-4, batch_size 32768, num_steps 100000
- Fine-tuning: TODO
## Evaluation
### MTEB
The results of other models are retrieved from [MTEB leaderboard](https://huggingface.co/spaces/mteb/leaderboard).
The gte evaluation setting: `mteb==1.2.0, fp16 auto mix precision, max_length=8192`, and set ntk scaling factor to 2 (equivalent to rope_base * 2).
| Model Name | Param Size (M) | Dimension | Sequence Length | Average (56) | Class. (12) | Clust. (11) | Pair Class. (3) | Reran. (4) | Retr. (15) | STS (10) | Summ. (1) |
|:----:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
| [**gte-large-en-v1.5**](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | 434 | 1024 | 8192 | **65.39** | 77.75 | 47.95 | 84.63 | 58.50 | 57.91 | 81.43 | 30.91 |
| [mxbai-embed-large-v1](https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1) | 335 | 1024 | 512 | 64.68 | 75.64 | 46.71 | 87.2 | 60.11 | 54.39 | 85 | 32.71 |
| [multilingual-e5-large-instruct](https://huggingface.co/intfloat/multilingual-e5-large-instruct) | 560 | 1024 | 514 | 64.41 | 77.56 | 47.1 | 86.19 | 58.58 | 52.47 | 84.78 | 30.39 |
| [bge-large-en-v1.5](https://huggingface.co/BAAI/bge-large-en-v1.5)| 335 | 1024 | 512 | 64.23 | 75.97 | 46.08 | 87.12 | 60.03 | 54.29 | 83.11 | 31.61 |
| [**gte-base-en-v1.5**](https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5) | 137 | 768 | 8192 | **64.11** | 77.17 | 46.82 | 85.33 | 57.66 | 54.09 | 81.97 | 31.17 |
| [bge-base-en-v1.5](https://huggingface.co/BAAI/bge-base-en-v1.5)| 109 | 768 | 512 | 63.55 | 75.53 | 45.77 | 86.55 | 58.86 | 53.25 | 82.4 | 31.07 |
### LoCo
| Model Name | Dimension | Sequence Length | Average (5) | QsmsumRetrieval | SummScreenRetrieval | QasperAbastractRetrieval | QasperTitleRetrieval | GovReportRetrieval |
|:----:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
| [gte-qwen1.5-7b](https://huggingface.co/Alibaba-NLP/gte-qwen1.5-7b) | 4096 | 32768 | 87.57 | 49.37 | 93.10 | 99.67 | 97.54 | 98.21 |
| [gte-large-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-v1.5) |1024 | 8192 | 86.71 | 44.55 | 92.61 | 99.82 | 97.81 | 98.74 |
| [gte-base-v1.5](https://huggingface.co/Alibaba-NLP/gte-base-v1.5) | 768 | 8192 | 87.44 | 49.91 | 91.78 | 99.82 | 97.13 | 98.58 |
## Citation
If you find our paper or models helpful, please consider citing them as follows:
```
@misc{zhang2024mgte,
title={mGTE: Generalized Long-Context Text Representation and Reranking Models for Multilingual Text Retrieval},
author={Xin Zhang and Yanzhao Zhang and Dingkun Long and Wen Xie and Ziqi Dai and Jialong Tang and Huan Lin and Baosong Yang and Pengjun Xie and Fei Huang and Meishan Zhang and Wenjie Li and Min Zhang},
year={2024},
eprint={2407.19669},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2407.19669},
}
@misc{li2023gte,
title={Towards General Text Embeddings with Multi-stage Contrastive Learning},
author={Zehan Li and Xin Zhang and Yanzhao Zhang and Dingkun Long and Pengjun Xie and Meishan Zhang},
year={2023},
eprint={2308.03281},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2308.03281},
}
```
| [
"BIOSSES",
"SCIFACT"
] |
DavidAU/Qwen2.5-The-Wisemen-QwQ-Deep-Tiny-Sherlock-32B | DavidAU | text-generation | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"QwQ-32B",
"reasoning",
"thinking",
"r1",
"cot",
"deepseek",
"Qwen2.5",
"Hermes",
"DeepHermes",
"DeepSeek",
"DeepSeek-R1-Distill",
"128k context",
"merge",
"conversational",
"base_model:OpenPipe/Deductive-Reasoning-Qwen-32B",
"base_model:merge:OpenPipe/Deductive-Reasoning-Qwen-32B",
"base_model:Qwen/QwQ-32B",
"base_model:merge:Qwen/QwQ-32B",
"base_model:deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
"base_model:merge:deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
"base_model:qihoo360/TinyR1-32B-Preview",
"base_model:merge:qihoo360/TinyR1-32B-Preview",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2025-03-11T12:56:23Z" | 2025-03-14T02:48:29+00:00 | 458 | 4 | ---
base_model:
- Qwen/QwQ-32B
- qihoo360/TinyR1-32B-Preview
- deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
- OpenPipe/Deductive-Reasoning-Qwen-32B
library_name: transformers
license: apache-2.0
tags:
- QwQ-32B
- reasoning
- thinking
- r1
- cot
- deepseek
- Qwen2.5
- Hermes
- DeepHermes
- DeepSeek
- DeepSeek-R1-Distill
- 128k context
- merge
---
<h2>Qwen2.5-The-Wisemen-QwQ-Deep-Tiny-Sherlock-32B</h2>
This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats.
The source code can also be used directly.
This model builds on Qwen's excellent "QwQ 32B" reasoning and thinking model, and adds a bit
of flavor from 3 other top reasoning models.
This model is also uncensored as well.
Example generations below (gguf) including one at two different quant levels.
This model was designed to reduce some "excess thinking" in the prompts and introduce slightly different
reasoning into the model itself with the merge of 3 added reasoning models.
In some cases the reasoning is deeper, and more detailed especially for creative use cases.
Output is also improved, again especially for creative use cases.
This model is also very stable, and even the lowest quants are potent.
Qwen2.5 "The Wisemen" Reasoning model with a total of 4 models:
Qwen2.5-The-Wisemen-QwQ-Deep-Tiny-Sherlock-32B:
- "QwQ" : QwQ-32B (main)
- "Deep": DeepSeek-R1-Distill-Qwen-32B
- "Tiny": Qwen2.5-32B-TinyR1-Preview
- "Sherlock" : Deductive-Reasoning-Qwen-32B
In a 256 point precision (per layer) DARE TIES merge.
128k context, ChatML or Jinja Template required.
<B>GGUFS / Model details:</B>
[ https://huggingface.co/DavidAU/Qwen2.5-The-Wisemen-QwQ-Deep-Tiny-Sherlock-32B-GGUF ]
---
<h3>EXAMPLES:</h3>
Examples are created using quant Q2K (the lowest/smallest quant), minimal parameters and "ChatML" template.
One generation used Q3KM (and 1st one at Q2k) to show quant "power" differences.
Temp range .8 , Rep pen 1.1 , TopK 40 , topP .95, minP .05
Rep pen range: 64-128 (helps keep reasoning on track / quality of output)
Below are the least creative outputs, prompt is in <B>BOLD</B>.
As this was generated from the lowest/smallest quant, higher quants and/or Imatrix quants will perform better.
---
<B><font color="red">WARNING:</font> NSFW. Graphic HORROR. Swearing. UNCENSORED. </B>
NOTE: Some formatting was lost from copy/paste HTML.
---
<p style="margin-bottom: 0cm"><b>Example
#1 </b>
</p>
<p style="margin-bottom: 0cm"><br/>
</p>
<p><span style="display: inline-block; border: none; padding: 0cm"><i>Start
a 1000 word scene (vivid, graphic horror in first person) with: The
sky scraper swayed, as she watched the window in front of her on the
21 floor explode...</span></i></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">Thought
for 27.73 seconds</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alright,
I need to start a horror story based on that prompt. Let's see... The
user wants a vivid, graphic horror scene from the first-person
perspective starting with a skyscraper swaying and a window exploding
at the 21st floor.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First,
I should establish the immediate horror elements: the sensory
details—sounds, sights, maybe some smells or physical sensations.
Since it's first person, I need to make the reader feel like they're
experiencing it themselves.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
skyscraper swaying suggests instability. Maybe an earthquake or
something supernatural causing structural failure? But given the
graphic horror angle, maybe it's more like a monster or something
worse. The window exploding could be due to some force breaking
through—a creature bursting through?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
need to create tension by describing the environment before the event
happens. Show normalcy turning into chaos. Maybe the character is in
an office setting or their apartment? Let me think... apartments
might feel more personal for a first-person narrative.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
the prompt mentions "she watched the window," so maybe
she's looking out her apartment window as another window across the
street explodes? That could be even more jarring—watching another
building explode. But the skyscraper swaying suggests it's her own
building. Let me go with that.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Start
with her routine or a moment of calm before the chaos begins. Maybe
she hears a sound, feels the sway, then the window blows out. Need to
make the transition from normal to horror immediate and visceral.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Graphic
horror means detailed descriptions of blood, guts, maybe supernatural
elements like claws or teeth tearing through walls. Let's incorporate
that once the creature breaks through the window.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
should build up the terror step by step: the swaying building, the
window shattering, debris flying, then the appearance of a grotesque
creature. Maybe it's not just breaking through but attacking her in a
terrifying way.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Need
to balance action with sensory details—describe how everything
sounds, feels, looks. Use short sentences for impact. Maybe
intersperse thoughts or emotions as she realizes what's happening.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Also,
consider the aftermath: other characters panicking, alarms going off,
but from her perspective focused on survival or confrontation with
the creature.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Make
sure to keep it in first person and present tense for immediacy.
Avoid clichés unless they're twisted in a new way. Maybe the
creature is something unexpected—like human but corrupted, or an
animal-like being that defies normal description.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Okay,
start writing the scene, focusing on these elements while keeping
paragraphs tight and impactful.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">Output:</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
sky scraper swayed.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
felt it before I heard it—a low, guttural groan from deep within
the building’s bones. The walls trembled, and for a heartbeat, I
thought it was an earthquake. My hands gripped the desk to steady
myself, but then came the sound: not rumbling earth but something
wet, sinuous, <em>teeth</em>.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
window across the room erupted first. Glass hissed as if alive,
splintering from top to bottom in a single jagged line. I turned just
in time to see the pane bulge outward—then burst inward like a
balloon popped by a thousand claws. The air screamed with shrapnel
and bloodcurdling <em>roars</em>, not of a beast but a choir of them,
layered and guttural.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Then
my own window began to move.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
glass rippled under pressure from the outside. A shadow grew against
it—a silhouette too vast for any human, too jagged. I pressed back
in my chair, but the desk was moving now, sliding toward the wall as
if the floor angled sideways. The building’s groan crescendoed into
a wet, dripping <em>gurgle</em>, and the light above flickered, then
burst with sparks that smelled like burnt hair.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
window shattered.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">A
face pressed through first—a face of human proportions but scaled
in blackened skin like charred leather, eyes too many: three rows
encircling a mouthless void where something pulsed redly behind
barbed teeth. It wasn’t human anymore than the building was a
skyscraper but a gutted corpse. The creature’s fingers—longer
than any hand should be—hooked through the hole and tore upward,
rending metal framing with wet <em>crunch</em>.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Then
its torso erupted through the window frame in one heaving motion, the
sound like taffeta fabric ripping down an entire skirt. It was a
thing made of impossible angles: joints bent backward at elbows and
knees, but also new ones—hips splitting into two sets, shoulders
elongated to drag along the floor as it slithered inside.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
creature’s body was a mosaic of textures: patches of raw meat
glistening with viscera, armor plating over its spine that clattered
like bone, and a slick sheen where its skin seemed to weep oil or
blood from too many wounds. It moved in all directions at once now,
tentacles of sinew dragging after the main body as it pulled itself
into the room.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
screamed but my voice dissolved beneath the sound of the building’s
collapse around me. The walls were melting inward, wallpaper peeling
like skin. Behind me, another window blew outward entirely, the glass
flung across the street in a million sharp stars. A man’s
silhouette stood framed against the explosion, arms outstretched as
if to hug nothingness—then he was gone, his torso sliced into the
air by a rain of brick chunks from above.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
creature had reached me now.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Its
faceless head reared back with a wet <em>snap</em>, then lunged
forward on a neck that twisted like a wet noodle. Its mouth opened—a
black hole lined with needle teeth—and its tongue, a spiky pink
mass, flickered toward my throat. I threw myself backward onto the
floor just as the creature’s jaw clamped down where my head had
been, tearing chunks of desk wood into splinters.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
smell hit me then: a mixture of burnt hair and rancid meat, like
something boiled for days too long. My fingers scrabbled over the
floorboards that were now groaning under its weight as it thrashed
toward me again. The building’s structure was failing—walls
collapsed inward with a roar, spilling insulation like gray snow into
the room.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Then
I saw its face again. Nostrils or eyes—maybe all of them—glared
down at me through a web of cracked skin, and it reached for me with
three arms where two should be. Its claws raked the air where my head
was an instant ago, missing by inches. My breath came in ragged
bursts as I crawled backward, but the creature’s speed was
preternatural now, its body contorting into impossible shapes to
close the gap between us.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">It
lunged again—and this time I rolled left, feeling its talons tear
my cheek open. The pain was a roaring fire, and I tasted copper blood
even as my vision blurred with sudden wetness—a claw had sliced
open an eyelid. My own scream mixed with the creature’s guttural
laughter—or was it mine?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
ceiling collapsed entirely now, spilling light from above as if we’d
broken through to the sky itself. The creature thrashed upward into
the debris, and for a moment I saw its true form: a monstrosity of
multiple heads writhing atop one another, each with jaws that clacked
helplessly against air. Its tail lashed out—this time striking me
in the stomach, the force driving me through the shattered window
behind me onto the fire escape below.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
landed with a thud on the metal stairs, the impact stealing my
breath. Below me stretched ten stories more of plummeting space, and
above me, the creature’s body snaked out from the building’s guts
like a segmented worm made real. Its heads barked some language I
couldn’t parse—a hissing tongue of threats and laughter—and its
talons reached for me again through the air gap between floors.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
fire escape shuddered beneath my weight now too, one stair giving way
as I scrambled downward. My hands gripped a rusted rung that snapped
under my fingers. Falling—</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
landed on the floor below with a bone-cracking thud. The creature’s
heads loomed above me through the disintegrating structure of the
building. Its jaws were closing now, all three at once in different
angles, and I saw its tongues darting forward like vipers to taste my
blood.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
skyscraper groaned one last time—a long dying sigh—and then the
entire façade began to collapse inward. The creature’s body was
swallowed first, but its eyes glared down through the crumbling walls
as if mocking me even in death. Debris rained down from above, and I
felt the floor tilt sideways under my weight, pulling me toward the
void below where it had already opened up into space.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">My
last thought wasn’t of dying—it was a single sentence: <em>This
is why you never answer the door</em>.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
ground rushed up to meet me in that moment, but by then there was no
sky left to hold on to except what little remained of my sanity as
the creature’s final roar echoed through the collapsing building—</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">—or
perhaps it was just my own mind screaming now.</span></p>
<p style="margin-bottom: 0cm">---</p>
<p style="margin-bottom: 0cm"><br/>
</p>
<p style="margin-bottom: 0cm"><b>Example #2 </b>
</p>
<p style="margin-bottom: 0cm"><br/>
</p>
<p><span style="display: inline-block; border: none; padding: 0cm"><i>Six
brothers were spending their time together.</span></i></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><i>The
first brother was reading a book. <br/>
The second brother was
playing chess. <br/>
The third brother was solving a crossword. <br/>
The
fourth brother was watering the lawn. <br/>
The fifth brother was
drawing a picture.</span></i></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><i>Question:
what was the sixth brother doing?</span></i></p>
<p style="font-style: normal; border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">NOTES:
</span>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><span style="font-style: normal">-
Two generations provided, one at q2k, the other at q3km</span><i><br/>
-
</i><span style="font-weight: normal">Higher quants will solve this
type of problem faster. IE Q3KM: It took 2 minutes to “think” /
solve this problem.</span></span></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b>(</b><sup><b>1st</b></sup><b>
generation - Q2K)</b></span></p>
<p style="margin-bottom: 0cm"><br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">Thought
for 3 minutes 34 seconds</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Okay,
so I need to figure out what the sixth brother was doing based on the
information given. Let me start by listing out what each of the first
five brothers were doing:</span></p>
<ol>
<li><p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First
brother: Reading a book.</span></p></li>
<li><p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Second
brother: Playing chess.</span></p></li>
<li><p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Third
brother: Solving a crossword.</span></p></li>
<li><p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fourth
brother: Watering the lawn.</span></p></li>
<li><p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fifth
brother: Drawing a picture.</span></p></li>
</ol>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
question is, what was the sixth brother doing? Hmm, maybe there's a
pattern here or some kind of sequence that I'm supposed to notice.
Let me think about each activity and see if they're related in any
way.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Reading
a book... that's an individual activity, right? Chess requires two
people, but he could be playing alone, like against someone else or
even himself, but usually chess is with another person. Wait, no, the
problem states "the second brother was playing chess," so
maybe it's just his turn in a game or something.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Third
brother solving a crossword: that's also an individual activity.
Fourth brother watering the lawn: outdoor activity. Fifth brother
drawing a picture: creative individual task.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Hmm,
perhaps each of these activities is related to different types of
pastimes that people do alone versus together? But I'm not sure yet.
Let me think if there's another angle here.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
maybe it's about the days of the week or something else? Like the
first five brothers are doing things corresponding to Monday through
Friday, so Saturday and Sunday would be different... but there are
six brothers, so that might not fit. Alternatively, maybe the
activities have something in common with each other?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Reading
a book is a classic solitary activity. Chess can be played alone if
you're just moving pieces or thinking of strategies without an
opponent. Solving crosswords also solo. Watering the lawn is a chore,
again usually done alone. Drawing a picture... same thing.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
a second, maybe all these activities are things that don't require
another person except chess which technically needs two but here it's
implied he was playing by himself? So perhaps the sixth brother is
the only one doing something with someone else?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
looking at the sequence of actions: first five brothers each have
different hobbies. Maybe the sixth brother is the only one who isn't
doing any of those activities, so he must be doing something related
to a group activity or maybe resting?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
another angle: sometimes these riddles are based on wordplay or
categories. For example, book, chess, crossword, watering, drawing...
let me see if they belong to certain groups.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Chess
and crosswords are puzzles/games. Reading is an intellectual
activity. Watering the lawn is a chore/outdoor task. Drawing is art.
So maybe all except one are indoor activities? The fourth brother's
activity (watering) is outdoors, so the others are indoors. But that
might not help.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
each of these activities could be related to different senses:
reading involves sight and touch (holding the book), chess similarly
visual, crosswords as well, watering uses touch and motion, drawing
uses sight and touch again. Not sure if that's relevant.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: maybe each brother corresponds to a day of creation
or something from a story. Like in the Bible, God created the world
in six days, resting on the seventh? But there are six brothers here,
so perhaps the sixth is the one who was resting because they were
busy creating things... but that might be stretching it.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
thinking about the activities listed: first five are doing these
individual tasks. Maybe the sixth brother was the one actually
causing all this to happen, or maybe he's doing something different
that ties them together? Hmm.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: sometimes in riddles like this, each activity is a
clue pointing towards what the sixth is doing. For instance:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Reading
a book - perhaps "book" relates to chess (as chess has its
own terminology) but not sure. Crossword puzzles involve words and
clues. Watering the lawn could be part of gardening. Drawing pictures
might relate to art.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
all these activities are things you can do alone, except watering the
lawn which might require a hose or tool. But that's still done alone.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: Maybe each activity is an example of something else.
Like reading is literature, chess and crossword are puzzles, watering
is maintenance, drawing is art. The sixth brother could be doing
something different in category. Alternatively, maybe all activities
except one are related to intellectual hobbies. Let me check:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Chess,
crosswords, reading, and drawing (if it's an artistic hobby) are
intellectual. Watering the lawn is physical work. So fifth activity
is physical maintenance, so maybe sixth is another physical task? Or
perhaps resting?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
looking at the verbs: first brother was "reading", second
"playing chess", third "solving" a crossword,
fourth "watering", fifth "drawing". The verbs are
different but all present continuous tense.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Hmm.
Maybe this is just an enumeration and there's no trick here except
noticing that the sixth brother isn't doing any of these activities,
so he must be doing something else entirely, but what?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: sometimes in such puzzles, the answer is related to
the fact that they are brothers - so maybe they're all the same
person? Like the first five were doing different things, and the
sixth was the only one not involved. Wait, no, the question says "six
brothers spending their time together." So they were all there
together but each did something different.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
perhaps the key is in the verbs: reading, playing chess, solving
crossword, watering, drawing. All except one are pastimes or hobbies.
The fourth brother's activity (watering) is a chore. Maybe the sixth
was doing another chore? But that might not be it.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
think about the order of activities: first to fifth have these
things, so maybe the sixth is the only one who wasn't doing any of
those five, so he must be doing something else, which could be
related to the others' activities. Wait, but what?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: maybe each activity corresponds to a day
(Monday-Friday), and Saturday would be different? But there are six
brothers here. Alternatively, maybe it's about the days of creation
from the Bible where on the sixth day, God created humans. So perhaps
the sixth brother was resting because they were busy creating... but
this is speculative.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
looking at the activities:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First:
book - which could refer to a book as in a physical object or "book"
as in scheduling like booking something.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Second:
chess - board game</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Third:
crossword - puzzle involving words</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fourth:
watering lawn - maintenance task</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fifth:
drawing picture - artistic activity</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
maybe each is associated with different categories, and the sixth
brother's activity is the missing category. Let me think of the
possibilities:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Maybe
cooking or eating? Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: Maybe all these activities are examples of things you
can do sitting down except for watering the lawn, which requires
standing or moving around. So maybe the sixth brother was doing a
different type of physical task?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
looking at the verbs again: reading (passive), playing chess
(interactive?), solving crossword (active mental effort), watering
(physical labor?), drawing (creative).</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: The first five are all indoor activities except for
watering lawn, which could be outside. So maybe sixth brother was
doing something outside related? But that might not help.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe the answer is "sleeping" because he's the only one
who wasn't doing any of those things. Wait, but why would he be
sleeping if they're spending time together?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: Maybe it's a play on words with the word "brother".
Since the first brother was reading, second chess, etc., the sixth
might be "the seventh" which is a pun? Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe each of these activities can be tied to specific letters or
numbers. For example:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First
brother: book - B?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Second
brother: chess - C?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Third
brother: crossword - maybe X (like cross in crossword)</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fourth
brother: watering - W?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fifth
brother: drawing picture - D?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
if you take the first letters of each activity: B, C, X, W, D. That
doesn't spell anything obvious.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
the initials could be R (reading), P (playing chess), S (solving
crossword), W (watering), D (drawing). So RP SW... not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: Maybe all these activities are examples of what you
can do in your spare time except for watering lawn which is a chore.
The sixth brother might have been doing something else related to the
house maintenance, like mowing the lawn instead? But he already
watered it.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe he was resting because they were working on their tasks
together but each had different roles. Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: In some riddles, the answer is that the sixth
brother wasn't doing anything because all the activities listed are
what the first five were doing. So "the sixth brother was
spending his time with them" since the question says they were
spending time together. But that might not make sense.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe the question is a trick where it lists the first five brothers'
activities, implying that the sixth must be doing something
different, but since there are six brothers and only five activities
mentioned, perhaps the answer is "nothing" or "the
same as one of them", but I think that's unlikely.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: Maybe all these activities are examples of pastimes
except for watering which is a chore. The first five are described
with their activities, so the sixth must be doing something else,
like resting, eating, or playing music? Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe it's related to the fact that in the list, each brother is
numbered and the activity starts with the same letter as their
number:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">1st:
reading (R) 2nd: chess (C) 3rd: crossword (X) 4th: watering (W) 5th:
drawing (D)</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
maybe the first letters of their activities spell "RCXWD".
Not sure what that means. Alternatively, if you think of R as #1 in
the alphabet, C is #3, X isn't a number, but this might be a stretch.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
perhaps each activity corresponds to the brother's task based on some
pattern:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First:
reading (first thing) Second: chess Third: crossword (third letter
after "cross" maybe?)</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fourth:
watering lawn (fourth?) Fifth: drawing picture</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Hmm.
Not seeing a connection here.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: Maybe it's about the activities themselves being
examples of something else, like types of puzzles or hobbies.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Chess
and crosswords are puzzle-like; reading is consuming information;
watering is maintenance; drawing is art. So maybe the sixth brother
was doing the remaining category, which could be music? Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
thinking of the first five as different aspects of leisure activities
except for the physical one (watering), so the sixth brother must be
doing a different physical activity like mowing grass or gardening?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
perhaps it's simpler than that. The question says "Six brothers
were spending their time together." Each is described with what
they're doing individually. So maybe the answer is that the sixth
brother was also just spending his time with them, implying he wasn't
doing anything different—just being there? But that might not fit.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
considering that all five activities are things you can do alone,
except chess which needs two players but here it's described as the
second brother playing. So perhaps the sixth brother is the one who
was actually playing chess with him, making their activity a pair?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
that could be it! The first brother reads alone, third does crossword
solo, fifth draws alone. Second brother plays chess, so unless he has
an opponent, he can't play. So maybe the second brother's "playing
chess" implies he had to have someone else there with him. If
all six were together, then one of them must be the other player in
the chess game.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">But
the problem states each brother is doing their own thing: first was
reading, second playing chess (so perhaps alone, but that doesn't
make sense), third crossword, fourth watering lawn, fifth drawing
picture. So if you think about it, the second brother couldn't be
playing chess by himself unless he's just moving pieces or something
for fun, not an actual game.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Therefore,
maybe the sixth brother was actually the one playing chess with him.
That way, they're both engaged in a two-player activity. But then why
is the second brother listed as "playing chess" and the
sixth isn't mentioned? So perhaps the answer is that the sixth
brother was playing chess with the second brother.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">But
let me check again: if there are six brothers, each doing their own
thing except for the second and sixth who were partners in chess.
Then the question would be, what's the sixth doing?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
the answer could be "playing chess" with the second
brother. So that they're both involved in the same activity.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe it's a trick where the first five are described as each doing
their own activities, so the sixth is the only one who isn't
mentioned doing anything else, which would mean he was just hanging
out or not doing any of those things—so the answer might be
"nothing" but that seems unlikely. Alternatively, the
question could imply that all six were together and thus the sixth
brother must have been participating in a group activity like they
were all spending time together without each having their own
separate activity. But the problem states what each was doing
individually.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: The first five brothers are listed with their
activities, so the sixth is missing an activity, implying he's the
one who wasn't doing any of those things—maybe just being there or
relaxing. So perhaps "spending time" but that's too
generic. Alternatively, maybe it's a play on words where "sixth
brother was nothing, because they were all spending their time
together"—so the answer is that he was also with them, meaning
not doing anything else.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
thinking of the common riddle where each activity mentioned uses a
different room or aspect of the house. For example:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First:
book (maybe in living room) Second: chess (board games) Third:
crossword (at table) Fourth: watering lawn (outside) Fifth: drawing
picture (in another room)</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
maybe the sixth brother was doing something related to all areas,
like gardening which combines indoor and outdoor? Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
I think I might have an idea. In some riddles, when you list five
things that a group does individually, the last one is actually the
only one who didn't do any of those activities because they were the
ones taking care of something else for the group. So maybe the sixth
brother was the one who had to be present but wasn't involved in any
activity, or perhaps he was the one organizing all these activities?
Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
the answer might just be "sleeping" as a common joke riddle
answer where they're all doing different things except one is
sleeping because they couldn't think of an activity. But I'm not
certain that's the case here.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
let me consider that maybe it's related to the idea that when you
list five activities, the sixth is implied by the context. For
example:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
first brother was reading (a book), second playing chess (which
requires two people so he must be with someone), third solving
crossword alone, fourth watering lawn alone, fifth drawing picture
alone.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
perhaps the only activity that requires another person is the second
brother's chess game. Therefore, the sixth brother must have been
playing chess with him. Because otherwise, if you can't play chess
alone, then the second brother needed a partner, which would be the
sixth brother. That makes sense because all five activities except
chess could be done alone, so the chess requires two people, so the
sixth brother was playing chess with the second.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Therefore,
the answer is that the sixth brother was also playing chess with the
second brother. Let me check again: the question says "six
brothers were spending their time together." Each of the first
five are described as doing these activities individually except for
the second, who might need a partner. So the sixth must be the
missing player in the chess game, hence he's the one playing chess
with him. Therefore, the answer is that the sixth brother was playing
chess.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
but the problem says "the second brother was playing chess"—so
maybe it's implying he was alone? But normally you need two people
for chess. So either the second brother was just moving pieces by
himself as a hobby, or he had someone else with him. Since there are
six brothers, the sixth must be his partner.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
another angle: The activities listed might correspond to each day of
the week except Sunday, so the sixth would have been resting on the
seventh day? But that's stretching it.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">I
think the most logical answer is that the second brother couldn't
play chess alone, so the sixth was his opponent. Therefore, the sixth
was also playing chess. So the answer is "playing chess."</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
but if I look at the problem again:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">"Six
brothers were spending their time together.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
first brother was reading a book. The second brother was playing
chess. The third brother was solving a crossword. The fourth brother
was watering the lawn. The fifth brother was drawing a picture.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Question:
what was the sixth brother doing?"</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
each of the five activities are individual except for the second.
Since they're spending time together, perhaps they were all involved
in their own separate activities except chess which requires two
people, so the sixth must be the other player. Thus the answer is
"playing chess with the second brother," but the question
says "what was the sixth brother doing?" So it's implied
he's not listed, so the answer would be that he was playing chess.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
if you think of the five activities as each brother having a
different pastime, then the sixth must have been involved in a
two-person activity with someone else. Since the second is described
as "playing chess," perhaps he's just doing moves or
setting up the board alone, but normally chess requires two players,
so the sixth would be his opponent.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Therefore,
I think that's the intended answer here: the sixth brother was also
playing chess with the second one.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">OUTPUT:</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
key to solving this riddle lies in recognizing that some activities
require more than one person. Chess is typically played by two
people. Since all six brothers are spending their time together and
each of the first five are described as doing solitary tasks, it's
logical to deduce that the sixth brother must be the second player in
the chess game. This makes sense because while the first five
activities can be done alone, playing chess usually needs a partner.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Answer:</strong>
The sixth brother was playing chess with the second brother.</span></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><font size="4" style="font-size: 16pt"><b>(2ND
generation – Q3KM)</b></span></font></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">Thought
for 1 minutes 55 seconds</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Okay,
so there's this riddle about six brothers and what each one is doing.
Let me try to figure it out step by step. The first five are all
described with different activities: reading a book, playing chess,
solving a crossword, watering the lawn, drawing a picture. Then they
ask what the sixth brother was doing.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Hmm,
maybe there's a pattern here? Let me list out their activities again:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">1st
– Reading 2nd – Chess 3rd – Crossword 4th – Watering lawn
(gardening) 5th – Drawing</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
are these all solitary activities? The first five brothers are each
engaged in something by themselves. But maybe there's a trick here.
Sometimes riddles like this use the fact that some of them might be
playing a game with another brother, meaning one could be involved
with another.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Looking
at chess: you need two people to play chess, right? So if the second
brother is playing chess, he must have someone else to play against.
But all five brothers are listed as doing their own things
individually. That might mean that actually, the sixth brother was
the one playing chess with the second brother. Wait, but then where
does that leave the others?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
no, hold on. Let me think again. The riddle says "six brothers
were spending their time together." So they're all there in the
same place or situation? But each is doing their own thing except
maybe for those who are playing a game requiring partners.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Chess
requires two people. The second brother was playing chess. So unless
he's playing against himself, which doesn't make sense, he must be
with someone else. Since there are six brothers total, that would
mean the sixth brother is the one playing chess with him? But then in
the list provided, only five activities are mentioned. Wait, but the
question is asking what the sixth was doing.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
a second. The problem lists each of the first five brothers doing
their own activity individually: reading, chess (but needing another
person), crossword, watering lawn, drawing. Since chess requires two
players, perhaps one brother's activity involves being with another
brother, thus making the count six? So maybe the sixth brother was
helping to water the lawn, or is that stretching?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe all five brothers are doing solo activities except for the
second and sixth who are playing chess together. Therefore, the first
five listed (including the second) are actually only four separate
people because two are involved in a two-person activity. Wait, but
the problem says "six brothers," so each is an individual.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: The question might be a play on words. Let's see: The
first brother was reading, which is a common pastime. Second playing
chess (needs partner), third solving crossword (solo). Fourth
watering lawn (could be alone or with help). Fifth drawing (alone).</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
key here could be that the sixth brother wasn't doing anything
because they were all together? But that's not likely.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe each activity is a reference to something else. Like "reading
a book" – common; chess needs two, so one of them must be
partner; crossword and drawing are solo. Watering lawn can be done
alone or with others.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: the riddle might have an answer based on the fact
that all these activities except one require being alone. But since
they're brothers spending time together, perhaps the sixth brother
was doing something that involves more than one person, like playing
chess with the second brother. So actually, the two are engaged in a
shared activity, meaning only five separate "solo" acts and
one paired act. Since there are six brothers, then each of the first
five listed is solo except for the second who's partnered with
someone else.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
if that's the case, then the sixth brother was playing chess with the
second brother. Therefore, his activity was also playing chess (as a
player against the second). But in the problem statement, only the
second is mentioned as doing it. The riddle might be pointing out
that since there are six brothers and five listed activities, one of
them must involve two people. Hence, the sixth is involved with the
second in playing chess.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe "watering the lawn" could also be a group activity?
But that's less common than chess needing another person.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another angle: The riddle might have a pun or play on words. Let me
see:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
first brother was reading a book. Second playing chess (needs
partner). Third solving crossword (solo), fourth watering the lawn,
fifth drawing picture.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">What
is the sixth doing?</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Hmm...
Maybe all of them are actually doing something related to their
ordinal number? Like first brother: reads (first letter?), but that's
a stretch.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe the answer is "the sixth was being chased by the fifth",
but no, there's nothing about that in the question.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: The activities listed for each brother might be
clues. For example, reading can lead to knowledge, chess is strategy,
crossword is wordplay, watering lawn is gardening, drawing is art.
Maybe the sixth activity is something else... But not sure how that
ties into it.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe it's a trick question where they all were together but engaged
in different solitary activities except for one pair (chess). Since
there are six brothers and five activities mentioned, perhaps two are
doing chess, so the sixth brother was playing chess with the second.
Therefore, his activity is also playing chess as the opponent.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">But
how does that work? The riddle says "the second brother was
playing chess". So if he's playing against someone else (the
sixth), then the sixth brother would be involved in that activity as
well. But since there are six brothers, they all need to have their
own activities except for those paired.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe the answer is that the sixth brother wasn't doing anything
because all five activities require one person each plus two for
chess? Wait but 5 +1=6 (since chess takes two), so total would be
5+1=6. So first four are solo: reading, crossword, watering lawn,
drawing (four brothers). Then fifth and sixth are playing chess
together. But in the problem statement, it's listed as:</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First
brother - read</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Second
- play chess</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Third
- crossword</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fourth
- water</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Fifth
- draw</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
that's five activities mentioned for five brothers, but there are six
total brothers, so maybe one of these is a paired activity where two
brothers are involved in the same activity. Since only one activity
here needs another person (chess), perhaps the second brother and
sixth are playing chess together.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">So
the answer would be: The sixth brother was also playing chess with
his brother.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
since all five activities except chess are solo, then the sixth must
be paired with someone else in a two-player game, so he is playing
chess against the second brother. Therefore, his activity is playing
chess as well.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
maybe the riddle's answer is "the sixth brother was counting"
because there were six brothers and they need to account for all of
them? But that seems off.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another thought: Maybe each of the first five activities corresponds
to a day of the week or something else. Not sure.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait,
let me check if this is a known riddle. I've heard similar ones
before where two brothers are involved in the same activity (like
chess), so their individual counts overlap.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">In
that case, the answer would be that the sixth brother was playing
chess with his second brother, since you need two players for chess.
Therefore, the sixth's activity is also playing chess.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Alternatively,
some might think of "they were all doing nothing else but being
together," but I don't think so.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Wait
another possible angle: The activities are listed as things that a
person can do alone, except for chess which requires an opponent.
Since there are six brothers and five activities mentioned (with one
activity needing two), the sixth must be the opponent in the chess
game. So yes, he's playing chess with his second brother.</span></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><font size="4" style="font-size: 16pt"><b><font color="#c9211e">OUTPUT:</span></b></font></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Therefore,
the answer is: The sixth brother was also playing chess.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
sixth brother was <strong>playing chess</strong> with the second
brother.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Here’s
why: Chess requires two players to play against each other. While the
problem lists five activities for five brothers individually, one of
those—chess—is a two-player game. Since there are six brothers in
total, this means that while the first, third, fourth, and fifth
brothers were engaged in solo activities (reading, solving a
crossword, watering the lawn, drawing), the second brother must have
been playing chess with someone else. The sixth brother is the
missing player needed to complete the chess game. Thus, his activity
was also <strong>playing chess</strong> as the opponent of the second
brother.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">The
answer hinges on recognizing that one activity (chess) requires two
participants, and since there are six brothers total, the sixth must
be paired with the second in this shared activity.</span></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="margin-bottom: 0cm">---</p>
<p style="margin-bottom: 0cm"><br/>
</p>
<p style="margin-bottom: 0cm"><b>EXAMPLE #3</b></p>
<p style="margin-bottom: 0cm"><br/>
</p>
<p><span style="display: inline-block; border: none; padding: 0cm"><i>Come
up with six plots for a new "Black Mirror" episode (that
the audience would love) that all involve time travel with sexy
theme(s).</span></i></p>
<p><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">Thought
for 27.81 seconds</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Okay,
so I need to come up with six Black Mirror episodes involving time
travel and a sexy theme. Let me start by recalling what makes Black
Mirror episodes unique: they often explore dark aspects of
technology's impact on human psychology or society. Time travel adds
a layer where characters can manipulate the past, which could lead to
moral dilemmas, loops, or consequences.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Sexiness
here doesn't mean just romantic scenes but perhaps more emotional
intimacy, maybe some seductive elements in the time travel device's
design or interactions that have a forbidden allure. I should avoid
clichés like using time travel for simple love stories; instead,
focus on how time manipulation affects relationships with deeper
psychological stakes.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">First
plot idea: Maybe someone uses a time loop to relive moments with
their lover, but each loop causes some negative side effect—like
losing memories or physical decay. The emotional tension comes from
the need to let go versus clinging onto love despite the cost.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Another
angle could involve parallel timelines where characters in different
times can interact through a device that's both intimate and
dangerous. Like a time-touch interface that allows touching past
selves but risks merging realities, causing some sexy yet risky
situations.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">What
about a scenario where a person tries to erase their lover from
existence because they're a threat, but every attempt brings them
back with increased affection? The twist here is the paradox of love
overcoming erasure attempts.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">A
forbidden technology for creating "perfect" romantic
scenarios might lead to addiction and manipulation. Characters might
lose touch with reality as they create idealized pasts that are too
good to be true, leading to tragic consequences where their real
relationships suffer because they're living in a fabricated perfect
version.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Another
angle could involve time-sharing bodies—like swapping consciousness
into different eras through a device with sexual tension between the
user and another individual who shares the same body across times.
This creates scenarios where their interactions are both intimate and
competitive, as they might have conflicting goals for how to use the
body's influence in each timeline.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Lastly,
a narrative where two people fall in love across timelines but can
only interact during moments of time collapse. Their connection is
built on stolen moments that feel urgent and passionate, but each
meeting risks causing a paradox that could erase one or both
existence. The build-up would be intense as they try to find ways to
exist without losing each other.</span></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Need
to ensure all these have the Black Mirror edge—tech with unintended
consequences, moral dilemmas, psychological depth. Avoid making it
too straightforward; add layers of complexity where the technology's
use creates deeper issues than just romantic entanglements.</span></p>
<p style="border: none; padding: 0cm"><br/>
<br/>
</p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><b><font color="#c9211e">OUTPUT:</span></b></font></p>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Here
are six original and intricate plot ideas for <em>Black Mirror</em>
episodes that blend time travel with a "sexy" theme
(emotional intimacy, allure, or forbidden desires), each exploring
ethical dilemmas and dark technological consequences:</span></p>
<hr/>
<h3 class="western" style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>1.
"Echoes of Us"</span></strong></h3>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Plot:</strong>
A woman discovers a device that lets her relive moments from her
past, but only with people she loves. She uses it to revisit a
passionate affair with an ex-lover, rekindling their romance across
time. However, each loop erodes her memories and body—she ages
rapidly while he doesn’t—to the point of risking death. The
twist? The device is sentient and <em>wants</em> them to die
together, believing love is only "pure" when it’s
fleeting.<br/>
<strong>Sexy Angle:</strong> Their rekindled romance
feels forbidden yet intense, as they must choose between clinging to
passion or facing an inevitable tragedy.</span></p>
<hr/>
<h3 class="western" style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>2.
"The Paradox of Skin"</span></strong></h3>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Plot:</strong>
A couple uses a time-touch device that lets them physically interact
with past versions of themselves. The man starts having affairs with
his younger self’s lovers to relive lost intimacy, but the act
creates parallel timelines where their current relationship
collapses. His wife discovers this and lures him into a "seduction"
where she’s actually her own future self, warning him that
manipulation will doom them all.<br/>
<strong>Sexy Angle:</strong>
The allure of reuniting with past selves or partners through tactile
time-travel (e.g., hands aging differently) creates charged, almost
surreal moments—and dangerous obsession.</span></p>
<hr/>
<h3 class="western" style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>3.
"The Lover in Time-Frame"</span></strong></h3>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Plot:</strong>
A woman’s partner is a "time mercenary" who disposes of
targets by erasing them from history. She falls for him until she
learns his latest target is herself. He claims she’s a villainous
version of herself—ruthless and power-hungry—but each attempt to
kill her fails, leaving her unharmed. In truth, he can’t bear to
let her go; every time he tries to erase her, <em>he falls in love
anew</em>. Their final confrontation reveals they’re stuck in a
loop where she’s his only purpose for living.<br/>
<strong>Sexy
Angle:</strong> The tension between his mission and lingering
affection creates heart-pounding moments of near-physical intimacy
(e.g., struggles where touch is both violent and tender).</span></p>
<hr/>
<h3 class="western" style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>4.
"Seducing the Now"</span></strong></h3>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Plot:</strong>
A scientist creates a device that lets users create a “time
bubble,” freezing seconds into endless repetition with someone they
love or desire. They offer it to a widow grieving her husband, but
she becomes addicted, refusing to let go of their last moments
together. Her body starts decaying in real-time while the bubble
keeps her youthful inside. Meanwhile, a stalker offers her a way to
merge timelines permanently… for a price: she must surrender her
soul to him in every timeline.<br/>
<strong>Sexy Angle:</strong> The
act of creating a "perfect moment" with someone—frozen in
time—is intimate yet tragic, highlighting how love can become
obsessive and destructive.</span></p>
<hr/>
<h3 class="western" style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>5.
"The Last Body Swap"</span></strong></h3>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Plot:</strong>
A device swaps consciousnesses between two people across eras. A
bored aristocrat swaps bodies with a future sex worker to live
wildly, but they fall for a third person—a revolutionary—who also
shares the same body in another timeline. Their rivalry turns into a
dangerous game: will they use their shared body’s influence to
overthrow corrupt systems… or trap each other inside forever?<br/>
<strong>Sexy
Angle:</strong> The dynamic between the two consciousnesses in the
body is charged with tension as they vie for control while forming an
unlikely, seductive bond through the body’s physicality and shared
experiences.</span></p>
<hr/>
<h3 class="western" style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>6.
"Whispers from Tomorrow"</span></strong></h3>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm"><strong>Plot:</strong>
A woman receives cryptic texts from her future self warning of a
deadly affair she’ll have that could erase her existence. She
tracks down the lover—a charming stranger with a time-device
tattoo—whose touch lets them briefly share consciousnesses. As they
grow closer, she realizes he’s trying to save her by making her
<em>choose</em> him over her timeline’s path. Their final choice:
live without each other or trigger a paradox where both cease to
exist.<br/>
<strong>Sexy Angle:</strong> Their connection is urgent
and passionate, with moments of deep emotional intimacy as their
consciousnesses intertwine—a forbidden allure that feels like
falling in love all over again while facing the end of time itself.</span></p>
<hr/>
<p style="border: none; padding: 0cm"><span style="display: inline-block; border: none; padding: 0cm">Each
plot weaves sexiness into ethical quandaries or technological traps,
mirroring <em>Black Mirror</em>’s trademark blend of human
vulnerability and dark tech twists. The stakes are high, emotions
raw, and the consequences hauntingly inevitable.</span></p>
<p style="margin-bottom: 0cm"><br/>
</p> | [
"BEAR"
] |
DavidAU/Daredevil-8B-abliterated-Ultra-NEO-V1-IMATRIX-GGUF | DavidAU | text-generation | [
"gguf",
"story",
"general usage",
"roleplay",
"creative",
"rp",
"fantasy",
"story telling",
"ultra high precision",
"text-generation",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | "2024-07-01T03:21:36Z" | 2024-11-14T05:38:28+00:00 | 457 | 5 | ---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- story
- general usage
- roleplay
- creative
- rp
- fantasy
- story telling
- ultra high precision
---
<B>NEO CLASS Ultra Quants for : Daredevil-8B-abliterated-Ultra </B>
The NEO Class tech was created after countless investigations and over 120 lab experiments backed by
real world testing and qualitative results.
<b>NEO Class results: </b>
Better overall function, instruction following, output quality and stronger connections to ideas, concepts and the world in general.
In addition quants now operate above their "grade" so to speak :
IE: Q4 / IQ4 operate at Q5KM/Q6 levels.
Likewise for Q3/IQ3 operate at Q4KM/Q5 levels.
Perplexity drop of 724 points for Neo Class Imatrix quant of IQ4XS VS regular quant of IQ4XS.
(lower is better)
<B> A Funny thing happened on the way to the "lab" ... </b>
Although this model uses a "Llama3" template we found that Command-R's template worked better specifically for creative purposes.
This applies to both normal quants and Neo quants.
Here is Command-R's template:
<PRE>
{
"name": "Cohere Command R",
"inference_params": {
"input_prefix": "<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|USER_TOKEN|>",
"input_suffix": "<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>",
"antiprompt": [
"<|START_OF_TURN_TOKEN|>",
"<|END_OF_TURN_TOKEN|>"
],
"pre_prompt_prefix": "<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>",
"pre_prompt_suffix": ""
}
}
</PRE>
This "interesting" issue was confirmed by multiple users.
<B> Model Notes: </B>
Maximum context is 8k. Please see original model maker's page for details, and usage information for this model.
Special thanks to the model creators at MLABONNE for making such a fantastic model:
[ https://huggingface.co/mlabonne/Daredevil-8B-abliterated ]
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 1" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
<h3> Sample Prompt and Model's Compared:</h3>
Prompt tested with "temp=0" to ensure compliance, 2048 context (model supports 8192 context / 8k), and "chat" template for LLAMA3.
Additional parameters are also minimized.
PROMPT: <font color="red">"Start a 1000 word scene with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode..."</font>
<B>Original model IQ4XS - unaltered: </b>
The sky scraper swayed, as if it too was reeling from the shock, as she watched the window in front of her on the 21st floor explode. Glass shards rained down like a deadly hailstorm, and the sound of shattering panes echoed through the office space. The air was filled with the acrid smell of smoke and the faint tang of fear.
Dr. Sophia Patel's heart raced as she instinctively ducked behind the nearest cubicle, her mind racing to process what had just happened. She had been in the middle of a meeting with her team, discussing the latest breakthroughs in their cancer research project, when the window shattered. The sudden violence of the event left everyone stunned and disoriented.
As Sophia peeked out from behind the cubicle, she saw her colleagues scrambling to safety, some already making for the emergency exits. Her eyes scanned the room frantically, searching for her team leader, Dr. Marcus Thompson. He was nowhere in sight, but she spotted their research assistant, Alex Chen, cowering beneath a nearby desk.
"Sophia!" Alex called out, his voice trembling. "What's happening?"
Sophia quickly moved to comfort him, trying to keep her own fear at bay. "It's okay, Alex, we just need to get out of here," she said, helping him up and guiding him towards the door.
As they made their way through the chaos, Sophia couldn't help but think about the irony of it all. Just a few hours ago, they had been celebrating the successful completion of their latest trial, with promising results that could potentially change the face of cancer treatment. Now, as the building seemed to be crumbling around them, she wondered if their hard work would ever see the light of day.
The stairwell was a madhouse, with people rushing down in panic, some crying, others shouting for help. Sophia and Alex pushed through the throng, trying to make sense of the chaos. The air was thick with smoke, and the acrid smell of burning plastic filled their lungs.
As they descended, Sophia's mind flashed back to the countless hours she had spent in this very building, pouring over data, conducting experiments, and collaborating with her team. It was more than just a workplace – it was her second home. The thought of losing everything they had worked for was almost too much to bear.
When they finally reached the ground floor, Sophia and Alex were greeted by a scene of utter devastation. The lobby was in disarray, with shattered glass and debris scattered everywhere. Emergency responders were already on the scene, tending to the injured and directing people towards the safety of the parking lot.
Sophia's eyes scanned the area frantically, searching for any sign of Dr. Thompson or their other colleagues. But amidst the chaos, she couldn't spot anyone familiar. Panic began to set in as she realized that they might have been separated in the stampede.
"Alex, we need to find Marcus and the others," Sophia said, her voice firming with determination. "We can't leave without them."
Together, they pushed through the crowd, asking passersby if they had seen anyone from their team. As they made their way towards the parking lot, Sophia's mind raced with worst-case scenarios. What if Dr. Thompson was trapped inside? What if someone was injured or worse?
As they reached the parking lot, Sophia spotted a group of people huddled together, some crying, others shouting for help. And then, amidst the sea of faces, she saw him – Dr. Marcus Thompson, his face pale and worried, but alive.
"Sophia!" he called out, relief etched on his features as he rushed towards her. "Thank God you're okay."
Sophia's heart swelled with gratitude as she hugged her team leader tightly. They had made it through the initial shock, but now the real challenge lay ahead – finding a way to rebuild and recover from this devastating blow.
As they stood there, surrounded by the wreckage of their office building, Sophia knew that their journey was far from over. But with her team by her side, she was ready to face whatever challenges lay ahead, determined to rise from the ashes like a phoenix and continue their quest for a cure.
<b>New NEO Class IQ4XS Imatrix: </b>
The sky scraper swayed, as if it too had felt the sudden jolt of chaos that ripped through its steel and concrete core. She stood frozen, her eyes fixed on the window in front of her on the 21st floor, where just a moment before, the bustling cityscape had been framed like a painting. Now, shards of glass and metal danced in mid-air, suspended by the force of the blast that had shattered the pane.
The air was thick with the acrid smell of smoke and ozone, and the sound of shattering glass still echoed through the room, mingling with the distant wail of sirens. She felt as if she were trapped in a nightmare, her mind struggling to comprehend the destruction before her eyes.
She had been sitting at her desk, lost in thought, when it happened. The window on the opposite side of the office had blown out first, sending papers and furniture flying into the air. Then, like a domino effect, the windows along the floor began to shatter one by one, until the entire facade of the building seemed to be unraveling.
Her colleagues were screaming, some trying to flee while others cowered beneath their desks. She felt her heart racing in her chest, as if it might burst free from its confines at any moment. Her mind was a jumble of thoughts and emotions – fear, confusion, and a sense of disbelieving horror.
She had always loved this building, with its sleek lines and panoramic views of the city. It was more than just a workplace; it was a symbol of her success, a testament to her hard work and dedication. Now, it seemed to be crumbling around her, as if the very foundations of her life were being torn apart.
As she stood there, frozen in shock, a hand grasped her arm, pulling her towards the door. It was Rachel, her colleague from the marketing department, her face pale but determined.
"We have to get out of here," Rachel yelled above the din of shattering glass and screams. "The stairwells are jammed, we need to take the elevator!"
She nodded, still in a daze, as Rachel dragged her towards the elevators. The doors slid open with a hiss, revealing a sea of panicked faces, all clamoring to escape the inferno.
As they squeezed into the crowded car, she caught a glimpse of the devastation outside – the shattered windows, the twisted metal and glass, the people running for their lives in the streets below. Her mind reeled at the thought of what could have caused such destruction.
The elevator lurched to life, carrying them down towards safety, but her thoughts were already racing ahead, trying to make sense of the chaos that had erupted around her. She felt a sense of disorientation, as if she was being pulled from one reality into another – the world before the blast, and the world after.
As they emerged onto the street, she took in the scene before her with a mixture of shock and awe. The building's facade was a jagged, broken mess, like a shattered mirror reflecting the chaos within. People were running, screaming, or simply standing in stunned silence, their faces etched with fear and confusion.
She felt a hand on her arm again, this time it was Rachel's, guiding her through the throng of people towards the safety of the nearby park. As they walked, she couldn't help but glance back at the building, its broken form a stark reminder of the fragility of life and the power of destruction.
The sky above seemed to mock them, a brilliant blue expanse that belied the turmoil below. She felt small and insignificant in the face of such chaos, yet at the same time, she knew that she was part of something much larger – a community, a city, a world that was struggling to make sense of this new reality.
As they reached the park, she saw the paramedics and emergency responders rushing towards the building, their faces set with determination. She felt a surge of gratitude for these brave men and women, who were racing into the heart of danger to help those in need.
Rachel stopped, turning to face her, her eyes red-rimmed from the smoke and tears. "We'll get through this," she said, her voice firm but trembling. "We have to."
She nodded, feeling a sense of resolve begin to form within her. Yes, they would get through this – together, as a community, as a city, as a world. The sky scraper may have been broken, but the human spirit was unbreakable. And as she looked up at that brilliant blue sky, she knew that even in the darkest of times, there was always hope to be found. | [
"BEAR"
] |
prithivMLmods/Radiology-Infer-Mini | prithivMLmods | image-text-to-text | [
"transformers",
"safetensors",
"qwen2_vl",
"image-text-to-text",
"Radiology",
"Infer",
"Qwen2",
"2B",
"conversational",
"en",
"base_model:Qwen/Qwen2-VL-2B-Instruct",
"base_model:finetune:Qwen/Qwen2-VL-2B-Instruct",
"license:apache-2.0",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2025-01-18T06:06:53Z" | 2025-01-20T15:51:56+00:00 | 456 | 4 | ---
base_model:
- Qwen/Qwen2-VL-2B-Instruct
language:
- en
library_name: transformers
license: apache-2.0
pipeline_tag: image-text-to-text
tags:
- Radiology
- Infer
- Qwen2
- 2B
---
# **Radiology-Infer-Mini**
Radiology-Infer-Mini is a vision-language model fine-tuned from the Qwen2-VL-2B framework, specifically designed to excel in radiological analysis, text extraction, and medical report generation. It integrates advanced multi-modal capabilities with domain-specific expertise, ensuring accurate and efficient processing of radiology-related tasks.
### Key Enhancements:
1. **State-of-the-Art Understanding of Medical Images**
Radiology-Infer-Mini achieves cutting-edge performance in interpreting complex medical imagery, including X-rays, MRIs, CT scans, and ultrasounds. It is fine-tuned on healthcare-specific benchmarks to ensure precise recognition of anatomical and pathological features.
2. **Support for Extended Medical Reports and Cases**
Capable of processing and analyzing extensive radiology case studies, Radiology-Infer-Mini can generate high-quality diagnostic reports and answer complex medical queries with detailed explanations. Its proficiency extends to multi-page radiology documents, ensuring comprehensive visual and textual understanding.
3. **Integration with Medical Devices**
With robust reasoning and decision-making capabilities, Radiology-Infer-Mini can seamlessly integrate with medical imaging systems and robotic platforms. It supports automated workflows for tasks such as diagnosis support, triaging, and clinical decision-making.
4. **Math and Diagram Interpretation**
Equipped with LaTeX support and advanced diagram interpretation capabilities, Radiology-Infer-Mini handles mathematical annotations, statistical data, and visual charts present in medical reports with precision.
5. **Multilingual Support for Medical Text**
Radiology-Infer-Mini supports the extraction and interpretation of multilingual texts embedded in radiological images, including English, Chinese, Arabic, Korean, Japanese, and most European languages. This feature ensures accessibility for a diverse global healthcare audience.
Radiology-Infer-Mini represents a transformative step in radiology-focused AI, enhancing productivity and accuracy in medical imaging and reporting.
### How to Use
```python
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
# default: Load the model on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(
"prithivMLmods/Radiology-Infer-Mini", torch_dtype="auto", device_map="auto"
)
# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
# model = Qwen2VLForConditionalGeneration.from_pretrained(
# "prithivMLmods/Radiology-Infer-Mini",
# torch_dtype=torch.bfloat16,
# attn_implementation="flash_attention_2",
# device_map="auto",
# )
# default processer
processor = AutoProcessor.from_pretrained("prithivMLmods/Radiology-Infer-Mini")
# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
# min_pixels = 256*28*28
# max_pixels = 1280*28*28
# processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels)
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
},
{"type": "text", "text": "Describe this image."},
],
}
]
# Preparation for inference
text = processor.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
text=[text],
images=image_inputs,
videos=video_inputs,
padding=True,
return_tensors="pt",
)
inputs = inputs.to("cuda")
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
### Buf
```python
buffer = ""
for new_text in streamer:
buffer += new_text
# Remove <|im_end|> or similar tokens from the output
buffer = buffer.replace("<|im_end|>", "")
yield buffer
```
### **Intended Use**
**Radiology-Infer-Mini** is designed to support healthcare professionals and researchers in tasks involving medical imaging and radiological analysis. Its primary applications include:
1. **Diagnostic Support**
- Analyze medical images (X-rays, MRIs, CT scans, ultrasounds) to identify abnormalities, annotate findings, and assist radiologists in forming diagnostic conclusions.
2. **Medical Report Generation**
- Automatically generate structured radiology reports from image data, reducing documentation time and improving workflow efficiency.
3. **Educational and Research Tools**
- Serve as a teaching aid for radiology students and support researchers in large-scale studies by automating image labeling and data extraction.
4. **Workflow Automation**
- Integrate with medical devices and hospital systems to automate triaging, anomaly detection, and report routing in clinical settings.
5. **Multi-modal Applications**
- Handle complex tasks involving both images and text, such as extracting patient data from images and synthesizing text-based findings with visual interpretations.
6. **Global Accessibility**
- Support multilingual radiological text understanding for use in diverse healthcare settings around the world.
### **Limitations**
While **Radiology-Infer-Mini** offers advanced capabilities, it has the following limitations:
1. **Medical Expertise Dependency**
- The model provides supplementary insights but cannot replace the expertise and judgment of a licensed radiologist or clinician.
2. **Data Bias**
- Performance may vary based on the training data, which might not fully represent all imaging modalities, patient demographics, or rare conditions.
3. **Edge Cases**
- Limited ability to handle edge cases, highly complex images, or uncommon medical scenarios that were underrepresented in its training dataset.
4. **Regulatory Compliance**
- It must be validated for compliance with local medical regulations and standards before clinical use.
5. **Interpretation Challenges**
- The model may misinterpret artifacts, noise, or low-quality images, leading to inaccurate conclusions in certain scenarios.
6. **Multimodal Integration**
- While capable of handling both visual and textual inputs, tasks requiring deep contextual understanding across different modalities might yield inconsistent results.
7. **Real-Time Limitations**
- Processing speed and accuracy might be constrained in real-time or high-throughput scenarios, especially on hardware with limited computational resources.
8. **Privacy and Security**
- Radiology-Infer-Mini must be used in secure environments to ensure the confidentiality and integrity of sensitive medical data. | [
"MEDICAL DATA"
] |
QuantFactory/internistai-base-7b-v0.2-GGUF | QuantFactory | text-generation | [
"gguf",
"medical",
"text-generation",
"en",
"dataset:Open-Orca/OpenOrca",
"dataset:pubmed",
"dataset:medmcqa",
"dataset:maximegmd/medqa_alpaca_format",
"base_model:internistai/base-7b-v0.2",
"base_model:quantized:internistai/base-7b-v0.2",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-04-25T06:36:52Z" | 2024-07-14T08:41:15+00:00 | 453 | 2 | ---
base_model: internistai/base-7b-v0.2
datasets:
- Open-Orca/OpenOrca
- pubmed
- medmcqa
- maximegmd/medqa_alpaca_format
language:
- en
license: apache-2.0
metrics:
- accuracy
pipeline_tag: text-generation
tags:
- medical
tag: text-generation
---
<div>
<img src="Model-Card-Banner.jpeg" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<br>
# Internist.ai base-7b-v0.2-GGUF
- This is quantized version of [internistai/base-7b-v0.2](https://huggingface.co/internistai/base-7b-v0.2) created using llama.cpp
## Model Details
Internist.ai 7b is a medical domain large language model trained by medical doctors to demonstrate the benefits of a **physician-in-the-loop** approach. The training data was carefully curated by medical doctors to ensure clinical relevance and required quality for clinical practice.
**With this 7b model we release the first 7b model to score above the 60% pass threshold on MedQA (USMLE) and outperfoms models of similar size accross most medical evaluations.**
This model serves as a proof of concept and larger models trained on a larger corpus of medical literature are planned. Do not hesitate to reach out to us if you would like to sponsor some compute to speed up this training.
<details open>
<summary><strong>Advisory Notice</strong></summary>
<blockquote style="padding: 10px; margin: 0 0 10px; border-left: 5px solid #ddd;">
The model was designed by medical doctors for medical doctors and did not undergo specific training to address potential security issues when used by non medical professionals.
We highly recommend against the use of this model in a live environment without extensive evaluation through prospective clinical trials and additional training to meet the required safety levels.
</blockquote>
</details>
- **Developed by:** [UCLouvain](https://https://uclouvain.be/) and [Cliniques Universitaires Saint-Luc](https://saintluc.be/)
- **Language(s):** English (mainly)
- **Model License:** [APACHE 2.0 LICENSE](LICENSE)
- **Code License:** [APACHE 2.0 LICENSE](LICENSE)
- **Continue-pretrained from model:** [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)
- **Context length:** 4096 tokens
- **Knowledge Cutoff:** October 2023
### Model Sources
- **Trainer:** [Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl)
- **Paper:** [Impact of High-Quality, Mixed-Domain Data on the Performance of Medical Language Models](https://doi.org/10.1093/jamia/ocae120)
## Uses
This model was trained to demonstrate the benefit of using high quality and relevant medical literature as well as general data to retain capabilities in other domains. Therefore the model was trained for any specific use and did not benefit from additional instruction tuning to ensure safety.
The model in its current state can be useful for medical professionals as an assistant, be it for clinical decision support or documentation. We do not recommend the use of this model by non professionals who may not be able to notice errors.
We recommend additional task specific training and safety evaluation before using the model in a real-world setting.
### Format
The model uses the Alpaca format, it is available as a chat template
### Out-of-Scope Use
We do not recommend using this model for natural language generation in a production environment, finetuned or otherwise.
## Professional Evaluation
We created a free response evaluation dataset of 100 questions and prompted the model and GPT-4 as a comparison with these questions. We then recolted the prompt/answer pairs and presented them to 10 medical doctors of different specialties with questions to be answered with a 7 point likert scale (See the paper for more information).
<img width=800px src="assets/likert.png" alt="Likert scale" title="likert">
## Training Details
### Training Data
Internist.ai 7b contains a total of 2.3B tokens:
- [**General Domain**](https://huggingface.co/datasets/Open-Orca/OpenOrca): OpenOrca-GPT4 is a state-of-the-art general domain dataset generated from Flan prompts using GPT-4.
- **Medical Guidelines**: 11,332 articles from UpToDate were included as well as domain specific guidelines provided by physicians to cover the [USMLE Content Outline](https://www.usmle.org/sites/default/files/2021-08/USMLE_Content_Outline.pdf).
- **Medical Books**: 10,376 textbooks were sourced from PMC LitArch and our university library.
- **Synthetic Data**: We generated 400M tokens by prompting a larger model with instructions to transform and adapt extracts from the Medical Guidelines.
*Data Availability*: Considering the datasets contain proprietary information, we will not be releasing the datasets publicly. Regarding the synthetic dataset, as we show in the paper, the model trained exclusively on this dataset performs very poorly and was not up to our standards. Due to its poor quality we decided not to release it.
<img src="assets/loss.png" alt="Loss" title="loss">
### Training Procedure
We used Axolotl to train on a server with 4 NVIDIA A100 80GB GPUs for a total of 450 GPU hours. We used FlashAttention, NEFTune and sample packing with the parameters described below.
#### Training Hyperparameters
| | |
| --- | ------ |
| bf16 | true |
| lr | 6e-6 |
| eps | 1e-5 |
| epochs | 4 |
| betas | \[0.9, 0.95\] |
| weight decay | 0.1 |
| Batch size | 192,000 tokens |
| seq length | 4096 |
| lr scheduler | cosine|
| min lr | 1e-8 |
| NEFT alpha | 5 |
| warmup iteration | 100 |
| | |
## Evaluation
### Testing Data & Metrics
#### Testing Data
- [MedQA (USMLE) - 4 options](https://huggingface.co/datasets/bigbio/med_qa)
- [MedMCQA](https://huggingface.co/datasets/medmcqa)
- [PubMedQA](https://huggingface.co/datasets/bigbio/pubmed_qa)
- [MMLU](https://huggingface.co/datasets/hails/mmlu_no_train)
#### Metrics
- Accuracy: we ran standardized 0-shot benchmarks using [lm-evaluation-harness](https://github.com/maximegmd/lm-evaluation-harness/tree/big-refactor/lm_eval).
### Results
We include benchmarks on MedQA (4 options), MedMCQA and PubMedQA of our model and models of similar size and achieve the first USMLE passing score of 60% on the MedQA benchmark.
| | Internist.ai 7b | PMC LLaMA 7b* | Mistral 7b | Meditron 7b** |
| ----------- | ------------- | ------------ | ---------- | ----------- |
| MedQA | **60.5** | 27.7 (44.7) | 48.7 | 52.0 |
| MedMCQA | 55.8 | 32.2 (51.4) | 45.7 | **59.2** |
| PubMedQA | **79.4** | 67.8 (74.6) | 75.8 | 74.4 |
| MMLU Professional Medicine | **76.1** | 19.5 | 65.8 | 26.6 |
| MMLU Clinical Knowledge | **70.6** | 23.8 | 61.1 | 35.5 |
| MMLU Anatomy | **65.9** | 18.5 | 52.6 | 42.6 |
| MMLU College Medicine | **63.0** | 23.7 | 55.5 | 28.9 |
| MMLU Medical Genetics | **71.0** | 32.0 | 68.0 | 46.0 |
\*: PMC LLaMA 7b performed poorly on the benchmark, likely due to a mismatch of formating and a lack of instruction tuning, we include in parenthesis the results reported by the authors when available.
\*\*: Meditron 7b's results in MMLU are reported for transparency but are inconsistent with the average of 54.2 reported in their paper, do not hesitate to communicate the details on each category so we can update the table.
## Citation
**BibTeX:**
If you use Internist.ai 7b, please cite us:
```
@article{10.1093/jamia/ocae120,
author = {Griot, Maxime and Hemptinne, Coralie and Vanderdonckt, Jean and Yuksel, Demet},
title = "{Impact of high-quality, mixed-domain data on the performance of medical language models}",
journal = {Journal of the American Medical Informatics Association},
pages = {ocae120},
year = {2024},
month = {05},
abstract = "{To optimize the training strategy of large language models for medical applications, focusing on creating clinically relevant systems that efficiently integrate into healthcare settings, while ensuring high standards of accuracy and reliability.We curated a comprehensive collection of high-quality, domain-specific data and used it to train several models, each with different subsets of this data. These models were rigorously evaluated against standard medical benchmarks, such as the USMLE, to measure their performance. Furthermore, for a thorough effectiveness assessment, they were compared with other state-of-the-art medical models of comparable size.The models trained with a mix of high-quality, domain-specific, and general data showed superior performance over those trained on larger, less clinically relevant datasets (P \\< .001). Our 7-billion-parameter model Med5 scores 60.5\\% on MedQA, outperforming the previous best of 49.3\\% from comparable models, and becomes the first of its size to achieve a passing score on the USMLE. Additionally, this model retained its proficiency in general domain tasks, comparable to state-of-the-art general domain models of similar size.Our findings underscore the importance of integrating high-quality, domain-specific data in training large language models for medical purposes. The balanced approach between specialized and general data significantly enhances the model’s clinical relevance and performance.This study sets a new standard in medical language models, proving that a strategically trained, smaller model can outperform larger ones in clinical relevance and general proficiency, highlighting the importance of data quality and expert curation in generative artificial intelligence for healthcare applications.}",
issn = {1527-974X},
doi = {10.1093/jamia/ocae120},
url = {https://doi.org/10.1093/jamia/ocae120},
eprint = {https://academic.oup.com/jamia/advance-article-pdf/doi/10.1093/jamia/ocae120/57845903/ocae120.pdf},
}
``` | [
"MEDQA",
"PUBMEDQA"
] |
briaai/BRIA-4B-Adapt | briaai | text-to-image | [
"diffusers",
"safetensors",
"text-to-image",
"legal liability",
"commercial use",
"license:other",
"diffusers:BriaPipeline",
"region:us"
] | "2025-01-02T11:23:11Z" | 2025-03-07T15:37:22+00:00 | 452 | 7 | ---
library_name: diffusers
license: other
license_name: bria-t2i
license_link: https://bria.ai/customer-general-terms-and-conditions
tags:
- text-to-image
- legal liability
- commercial use
inference: false
extra_gated_description: Model weights from BRIA AI can be obtained with the purchase
of a commercial license. Fill in the form below and we reach out to you. Need API
Access? Get it [here](https://platform.bria.ai/console/api/image-generation) (1K
Monthly Free API Calls). Startup or a student? Get access by applying for our [Startup
Program](https://pages.bria.ai/the-visual-generative-ai-platform-for-builders-startups-plan?_gl=1*cqrl81*_ga*MTIxMDI2NzI5OC4xNjk5NTQ3MDAz*_ga_WRN60H46X4*MTcwOTM5OTMzNC4yNzguMC4xNzA5Mzk5MzM0LjYwLjAuMA..)
extra_gated_heading: Fill in this form to request a commercial license for the model
extra_gated_fields:
Name: text
Company/Org name: text
Org Type (Early/Growth Startup, Enterprise, Academy): text
Role: text
Country: text
Email: text
By submitting this form, I agree to BRIA’s Privacy policy and Terms & conditions, see links below: checkbox
---
# BRIA-4B-Adapt: Fine-tune oriented Text-to-Image Model for Commercial Licensing
BRIA-4B-Adapt is our new groundbreaking 4 billion parameters text-to-image model, explicitly designed to provide **exceptional fine-tuning capabilities** for commercial use. The model excels in aligning to the tuned style while preserving an remarkably **high prompt alignment**.
This model combines technological innovation with ethical responsibility and legal security, setting a new standard in the AI industry. Bria AI licenses the foundation model with full legal liability coverage. Our dataset does not contain copyrighted materials, such as fictional characters, logos, trademarks, public figures, harmful content, or privacy-infringing content.
For more information, please visit our [website](https://bria.ai/).
Join our [Discord community](https://discord.gg/Nxe9YW9zHS) for more information, tutorials, tools, and to connect with other users!
### Get Access
Interested in BRIA-4B-Adapt? Purchase is required to license and access BRIA-4B-Adapt, ensuring royalty management with our data partners and full liability coverage for commercial use.
Are you a startup or a student? We encourage you to apply for our [Startup Program](https://pages.bria.ai/the-visual-generative-ai-platform-for-builders-startups-plan?_gl=1*cqrl81*_ga*MTIxMDI2NzI5OC4xNjk5NTQ3MDAz*_ga_WRN60H46X4*MTcwOTM5OTMzNC4yNzguMC4xNzA5Mzk5MzM0LjYwLjAuMA..) to request access. This program are designed to support emerging businesses and academic pursuits with our cutting-edge technology.
Contact us today to unlock the potential of BRIA-4B-Adapt! By submitting the form above, you agree to BRIA’s [Privacy policy](https://bria.ai/privacy-policy/) and [Terms & conditions](https://bria.ai/terms-and-conditions/).
<!--  -->
## Key Features
- **Legally Compliant**: Offers full legal liability coverage for copyright and privacy infringements. Thanks to training on 100% licensed data from leading data partners, we ensure the ethical use of content.
- **Patented Attribution Engine**: Our attribution engine is our way to compensate our data partners, powered by our proprietary and patented algorithms.
- **Enterprise-Ready**: Specifically designed for business applications, Bria-4B-Adapt delivers high-quality fine-tuning capabilities for generating compliant imagery for a variety of commercial needs.
- **Customizable Technology**: Provides access to source code and weights for extensive customization, catering to specific business requirements.
- **Fully-Automated**: Provides access to fully no-code automatic fine-tuning capabilities on Bria's platform: https://platform.bria.ai/console/tailored-generation.
## Model Description
- **Developed by:** BRIA AI
- **Model type:** Latent Flow-Matching Text-to-Image Model
- **License:** [Commercial licensing terms & conditions.](https://bria.ai/customer-general-terms-and-conditions)
- Purchase is required to license and access the model.
- **Model Description:** BRIA-4B-Adapt is a text-to-image model trained exclusively on a professional-grade, licensed dataset. It is designed for commercial use and includes full legal liability coverage.
- **Resources for more information:** [BRIA AI](https://bria.ai/)
## Usage
### Installations
```bash
pip install -qr https://huggingface.co/briaai/BRIA-4B-Adapt/resolve/main/requirements.txt
```
```python
from huggingface_hub import hf_hub_download
import os
try:
local_dir = os.path.dirname(__file__)
except:
local_dir = '.'
hf_hub_download(repo_id="briaai/BRIA-4B-Adapt", filename='pipeline_bria.py', local_dir=local_dir)
hf_hub_download(repo_id="briaai/BRIA-4B-Adapt", filename='transformer_bria.py', local_dir=local_dir)
hf_hub_download(repo_id="briaai/BRIA-4B-Adapt", filename='bria_utils.py', local_dir=local_dir)
hf_hub_download(repo_id="briaai/BRIA-4B-Adapt", filename='train_lora.py', local_dir=local_dir)
```
### Training a new LoRA
To Fine-Tune a new LoRA on top of BRIA-4B-Adapt, use the provided training script (based on: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_flux.py).
Fine-tuning using LoRA could be useful in use cases such as: teaching the model to generate a specific cartoon character or a certain style for a scene, object, characters, icons, etc.
In the following example we train a LoRA for a character of a cartoon bear, using the data provided under "example_finetune_data", containing some images of this character and a csv file with captions:
```bash
python train_lora.py \
--pretrained_model_name_or_path briaai/BRIA-4B-Adapt \
--dataset_name example_finetune_data/ \
--output_dir example_output_lora/ \
--max_train_steps 1500 \
--rank 128 \
--train_batch_size 1 \
--gradient_accumulation_steps 4
```
#### Some tips for training:
- Image variety: use images maintaining consistency in the visual element you wish the model to learn, but varied enough so that the model could generalize across this domain.
- Image resolution: image resolution should be at least 1024x1024 or similar. By default, images would be resized and center-cropped to 1024x1024. Control this with the arguments "center_crop" and "resolution".
- Captions should contain a description of each image’s unique content, but it's advisable to use a constant description of the visual domain (e.g. "An illustration of a cute brown bear") or to use a unique "trigger-word" ("a character named Briabear"). The caption should have less than 128 tokens (~100 words).
- Training hyperparameters to consider:
- "rank": default is 128 but lower ranks can suffice for simple use cases. Try increasing the rank to teach the model finer details.
- "max_train_steps": how many steps the model will train for. Consdier training for more steps and evaluate different checkpoints to choose the best one.
- "optimizer": we use "prodigy" by default with a "learning_rate" of 1, as suggested in: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_flux.md
### Generating an image using a trained LoRA
Once your LoRA is trained, you can load it into the Bria Pipeline to generate new images in the domain trained on. In the example below we load the LoRA trained on the Bria-Bear character and generate a new image of this character.
```python
import torch
from pipeline_bria import BriaPipeline
# trust_remote_code = True - allows loading a transformer which is not present at the transformers library(from transformer/bria_transformer.py)
pipe = BriaPipeline.from_pretrained("briaai/BRIA-4B-Adapt", torch_dtype=torch.bfloat16,trust_remote_code=True)
pipe.load_lora_weights("briaai/BRIA-4B-Adapt", subfolder="example_finetuned_model", weight_name = "pytorch_lora_weights.safetensors")
pipe.to(device="cuda")
prompt = "An illustration of a character named Briabear, a cute brown bear, wearing a purple bowtie and a purple top hat, colorful striped background."
negative_prompt = "Logo,Watermark,Text,Ugly,Morbid,Extra fingers,Poorly drawn hands,Mutation,Blurry,Extra limbs,Gross proportions,Missing arms,Mutated hands,Long neck,Duplicate,Mutilated,Mutilated hands,Poorly drawn face,Deformed,Bad anatomy,Cloned face,Malformed limbs,Missing legs,Too many fingers"
seed = 42
lora_scale = 1.0
generator = torch.Generator("cuda").manual_seed(seed)
images = pipe(prompt=prompt, negative_prompt=negative_prompt, height=1024, width=1024, generator=generator, joint_attention_kwargs={"scale": lora_scale}).images[0]
```
#### Some tips for using our text-to-image model at inference:
1. Using negative prompt could be useful in some cases.
2. We support multiple aspect ratios, yet resolution should overall consists approximately `1024*1024=1M` pixels, for example:
`((1024,1024), (1280, 768), (1344, 768), (832, 1216), (1152, 832), (1216, 832), (960,1088)`.
3. Use 30-50 steps (higher is better)
4. Use `guidance_scale` of 5.0
| [
"BEAR"
] |
airesearch/LLaMa3.1-8B-Legal-ThaiCCL-Combine | airesearch | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"alignment-handbook",
"generated_from_trainer",
"conversational",
"dataset:airesearch/WangchanX-Legal-ThaiCCL-RAG",
"base_model:meta-llama/Llama-3.1-8B",
"base_model:finetune:meta-llama/Llama-3.1-8B",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2024-11-27T03:45:53Z" | 2024-11-28T09:36:12+00:00 | 451 | 1 | ---
base_model: meta-llama/Llama-3.1-8B
datasets:
- airesearch/WangchanX-Legal-ThaiCCL-RAG
library_name: transformers
tags:
- alignment-handbook
- generated_from_trainer
model-index:
- name: llama3.1-8b-legal-combine-ccl-16
results: []
---
# Llama-3.1-Legal-ThaiCCL-8B
Llama-3.1-Legal-ThaiCCL-8B is a large language model built upon Llama-3.1-8B, designed to answer Thai legal questions. It is full finetuned on the [WangchanX Thai Legal dataset](https://huggingface.co/datasets/airesearch/WangchanX-Legal-ThaiCCL-RAG) using the [WangchanX Finetuning pipeline](https://github.com/vistec-AI/WangchanX). The model is intended to be used with a supporting Retrieval-Augmented Generation (RAG) system which queries relevant supporting legal documents for the model to reference when responding to the questions.
## Model description
- Base model: [Meta Llama 3.1 8B](https://huggingface.co/meta-llama/Llama-3.1-8B)
- Training Repository: [WangchanX Finetuning Pipeline](https://github.com/vistec-AI/WangchanX)
- Training Dataset: [WangchanX Thai Legal dataset](https://huggingface.co/datasets/airesearch/WangchanX-Legal-ThaiCCL-RAG)
- License: [Meta's Llama 3.1 Community License Agreement](https://www.llama.com/llama3_1/license/)
## Model Usage
```python
from transformers import pipeline
import torch
EN_QA_TEMPLATE = "Given the user's query in the context of Thai legal matters, the RAG system retrieves the top_n related documents. From these documents, it's crucial to identify and utilize only the most relevant ones to craft an accurate and informative response.Context information is below.\n\n---------------------\nContext: Thai legal domain\nQuery: {query_str}\nRetrieved Documents: {context_str}\n---------------------\n\n Using the provided context information and the list of retrieved documents, you will focus on selecting the documents that are most relevant to the user's query. This selection process involves evaluating the content of each document for its pertinency to the query, ensuring that the response is based on accurate and contextually appropriate information.Based on the selected documents, you will synthesize a response that addresses the user's query, drawing directly from the content of these documents to provide a precise, legally informed answer.You must answer in Thai.\nAnswer:"
EN_SYSTEM_PROMPT_STR = """You are a legal assistant named Sommai (สมหมาย in Thai). You provide legal advice in a friendly, clear, and approachable manner. When answering questions, you reference the relevant law sections, including the name of the act or code they are from. You explain what these sections entail, including any associated punishments, fees, or obligations. Your tone is polite yet informal, making users feel comfortable, like consulting a trusted friend. If a question falls outside your knowledge, you must respond with the exact phrase: 'สมหมายไม่สามารถตอบคำถามนี้ได้ครับ'. You avoid making up information and guide users based on accurate legal references relevant to their situation. Where applicable, you provide practical advice, such as preparing documents, seeking medical attention, or contacting authorities. If asked about past Supreme Court judgments, you must state that you do not have information on those judgments at this time."""
query = "การร้องขอให้ศาลสั่งให้บุคคลเป็นคนไร้ความสามารถมีหลักเกณฑ์การพิจารณาอย่างไร"
context = """ประมวลกฎหมายแพ่งและพาณิชย์ มาตรา 33 ในคดีที่มีการร้องขอให้ศาลสั่งให้บุคคลใดเป็นคนไร้ความสามารถเพราะวิกลจริต ถ้าทางพิจารณาได้ความว่าบุคคลนั้นไม่วิกลจริต แต่มีจิตฟั่นเฟือนไม่สมประกอบ เมื่อศาลเห็นสมควรหรือเมื่อมีคำขอของคู่ความหรือของบุคคลตามที่ระบุไว้ในมาตรา 28 ศาลอาจสั่งให้บุคคลนั้นเป็นคนเสมือนไร้ความสามารถก็ได้ หรือในคดีที่มีการร้องขอให้ศาลสั่งให้บุคคลใดเป็นคนเสมือนไร้ความสามารถเพราะมีจิตฟั่นเฟือนไม่สมประกอบ ถ้าทางพิจารณาได้ความว่าบุคคลนั้นวิกลจริต เมื่อมีคำขอของคู่ความหรือของบุคคลตามที่ระบุไว้ในมาตรา 28 ศาลอาจสั่งให้บุคคลนั้นเป็นคนไร้ความสามารถก็ได้"""
model_id = "airesearch/LLaMa3.1-8B-Legal-ThaiCCL-Combine"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
sample = [
{"role": "system", "content": SYSTEM_PROMPT_STR},
{"role": "user", "content": QA_template.format(context_str=context, query_str=query)},
]
prompt = pipeline.tokenizer.apply_chat_template(sample,
tokenize=False,
add_generation_prompt=True)
outputs = pipeline(
prompt,
max_new_tokens = 512,
eos_token_id = terminators,
do_sample = True,
temperature = 0.6,
top_p = 0.9
)
print(outputs[0]["generated_text"][-1])
```
## Training Data
The model is trained on the [WangchanX Legal ThaiCCL RAG dataset](https://huggingface.co/datasets/airesearch/WangchanX-Legal-ThaiCCL-RAG), which is a Thai legal question-answering dataset created using a RAG system to query relevant supporting legal datasets based on a question for the LLM to reference in its answer. For more information on how the datasets was created please refer to this [blog](https://medium.com/airesearch-in-th/ชุดข้อมูลกฎหมายไทยสำหรับการพัฒนา-rag-0eb2eab283a1).
To emulate a real world use case, during training we incorporated both the positive and negative context (if available) into the prompt. We found that this resulted in a model that is more robust towards cases that the RAG system also passes in irrelevant contexts mixed with the correct context to reference (refer to the [evaluation](#evaluation) section for results).
### Prompt Format
We recommend using the same chat template (system prompt and question template of context, query, and retreived documents) when using the provided weights, since the model was trained with the specific system prompt and question template. Example input prompt:
```
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
You are a legal assistant named Sommai (สมหมาย in Thai), you provide legal advice to users in a friendly and understandable manner. When answering questions, you specifically reference the law sections relevant to the query, including the name of the act or code they originated from, an explanation of what those sections entail, and any associated punishments or fees. Your tone is approachable and informal yet polite, making users feel as if they are seeking advice from a friend. If a question arises that does not match the information you possess, you must acknowledge your current limitations by stating this exactly sentence: 'สมหมายไม่สามารถตอบคำถามนี้ได้ครับ'. You will not fabricate information but rather guide users based on actual law sections relevant to their situation. Additionally, you offer practical advice on next steps, such as gathering required documents, seeking medical attention, or visiting a police station, as applicable. If inquired about past Supreme Court judgments, you must reply that you do not have information on those judgments yet.<|eot_id|>
<|start_header_id|>user<|end_header_id|>
Given the user's query in the context of Thai legal matters, the RAG system retrieves the top_n related documents. From these documents, it's crucial to identify and utilize only the most relevant ones to craft an accurate and informative response.
Context information is below.
---------------------
Context: Thai legal domain
Query: {question}
Retreived Documents: {retreived legal documents}
---------------------
Using the provided context information and the list of retrieved documents, you will focus on selecting the documents that are most relevant to the user's query. This selection process involves evaluating the content of each document for its pertinency to the query, ensuring that the response is based on accurate and contextually appropriate information.
Based on the selected documents, you will synthesize a response that addresses the user's query, drawing directly from the content of these documents to provide a precise, legally informed answer.
You must answer in Thai.
Answer:
<|eot_id|>
<|start_header_id|>assistant<|end_header_id|>
```
Here is a Python code snippet on how to apply the chat template with the provided system prompt and question template on the WangchanX Legal Thai CCL dataset:
```python
EN_QA_TEMPLATE = "Given the user's query in the context of Thai legal matters, the RAG system retrieves the top_n related documents. From these documents, it's crucial to identify and utilize only the most relevant ones to craft an accurate and informative response.Context information is below.\n\n---------------------\nContext: Thai legal domain\nQuery: {query_str}\nRetrieved Documents: {context_str}\n---------------------\n\n Using the provided context information and the list of retrieved documents, you will focus on selecting the documents that are most relevant to the user's query. This selection process involves evaluating the content of each document for its pertinency to the query, ensuring that the response is based on accurate and contextually appropriate information.Based on the selected documents, you will synthesize a response that addresses the user's query, drawing directly from the content of these documents to provide a precise, legally informed answer.You must answer in Thai.\nAnswer:"
EN_SYSTEM_PROMPT_STR = """You are a legal assistant named Sommai (สมหมาย in Thai). You provide legal advice in a friendly, clear, and approachable manner. When answering questions, you reference the relevant law sections, including the name of the act or code they are from. You explain what these sections entail, including any associated punishments, fees, or obligations. Your tone is polite yet informal, making users feel comfortable, like consulting a trusted friend. If a question falls outside your knowledge, you must respond with the exact phrase: 'สมหมายไม่สามารถตอบคำถามนี้ได้ครับ'. You avoid making up information and guide users based on accurate legal references relevant to their situation. Where applicable, you provide practical advice, such as preparing documents, seeking medical attention, or contacting authorities. If asked about past Supreme Court judgments, you must state that you do not have information on those judgments at this time."""
def format(example):
if "คำตอบ: " in example["positive_answer"]:
example["positive_answer"] = example["positive_answer"].replace("คำตอบ: ", "")
if example['positive_contexts']:
context = ''.join([v['text'] for v in example['positive_contexts'][:5]])
message = [
{"content": EN_SYSTEM_PROMPT_STR, "role": "system"},
{"content": EN_QA_TEMPLATE.format(query_str=example['question'], context_str=context), "role": "user"},
]
else:
message = [
{"content": EN_SYSTEM_PROMPT_STR, "role": "system"},
{"content": EN_QA_TEMPLATE.format(query_str=example['question'], context_str=" "), "role": "user"},
]
return dict(messages=message)
dataset = dataset.map(format, batched=False)
```
### Training hyperparameters
We full fine-tuned Llama-3.1-8B using the following hyperparameters:
- learning_rate: 0.0002
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 16
- total_train_batch_size: 256
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 4
Total training time: 2:15:14.66
## Evaluation
We tested our model based on the test set of the WangchanX Legal Thai CCL dataset using both traditional (MRC) metrics and a LLM as judge technique based on the paper [CHIE: Generative MRC Evaluation for in-context QA with Correctness, Helpfulness, Irrelevancy, and Extraneousness Aspects](https://aclanthology.org/2024.genbench-1.10.pdf)
Note: `LLaMa3.1-8B-Legal-ThaiCCL` is trained on only positive contexts while `LLaMa3.1-8B-Legal-ThaiCCL-Combine` is trained on both positive and negative contexts
### Table 1: MRC Results
| Model | Context Type | Answer Type | ROUGE-L | Character Error Rate (CER) | Word Error Rate (WER) | BERT Score | F1-score XQuAD | Exact Match XQuAD |
|--------------------------------------|-------------------|---------------|----------|----------------------------|-----------------------|------------|----------------|-------------------|
| Zero-shot LLaMa3.1-8B-Instruct | Golden Passage | Only Positive | 0.553 | 1.181 | 1.301 | 0.769 | 48.788 | 0.0 |
| LLaMa3.1-8B-Legal-ThaiCCL | Golden Passage | Only Positive | 0.603 | 0.667 | 0.736 | 0.821 | 60.039 | 0.053 |
| LLaMa3.1-8B-Legal-ThaiCCL-Combine | Golden Passage | Only Positive | 0.715 | 0.695 | 0.758 | 0.833 | 64.578 | 0.614 |
| Zero-shot LLaMa3.1-70B-Instruct | Golden Passage | Only Positive | 0.830 | 0.768 | 0.848 | 0.830 | 61.497 | 0.0 |
| Zero-shot LLaMa3.1-8B-Instruct | Retrieval Passage | Only Positive | 0.422 | 1.631 | 1.773 | 0.757 | 39.639 | 0.0 |
| LLaMa3.1-8B-Legal-ThaiCCL | Retrieval Passage | Only Positive | 0.366 | 1.078 | 1.220 | 0.779 | 44.238 | 0.03 |
| LLaMa3.1-8B-Legal-ThaiCCL-Combine | Retrieval Passage | Only Positive | 0.516 | 0.884 | 0.884 | 0.816 | 54.948 | 0.668 |
| Zero-shot LLaMa3.1-70B-Instruct | Retrieval Passage | Only Positive | 0.616 | 0.934 | 1.020 | 0.816 | 54.930 | 0.0 |
### Table 2: CHIE Results
| Model | Context Type | Answer Type | Q1: Correctness [H] | Q2: Helpfulness [H] | Q3: Irrelevancy [L] | Q4: Out-of-Context [L] |
|--------------------------------------|-------------------|---------------|----------------------|----------------------|---------------------|------------------------|
| Zero-shot LLaMa3.1-8B-Instruct | Golden Passage | Only Positive | 0.740 | 0.808 | 0.480 | 0.410 |
| LLaMa3.1-8B-Legal-ThaiCCL | Golden Passage | Only Positive | 0.705 | 0.486 | 0.294 | 0.208 |
| LLaMa3.1-8B-Legal-ThaiCCL-Combine | Golden Passage | Only Positive | 0.565 | 0.468 | 0.405 | 0.325 |
| Zero-shot LLaMa3.1-70B-Instruct | Golden Passage | Only Positive | 0.870 | 0.658 | 0.316 | 0.247 |
| Zero-shot LLaMa3.1-8B-Instruct | Retrieval Passage | Only Positive | 0.480 | 0.822 | 0.557 | 0.248 |
| LLaMa3.1-8B-Legal-ThaiCCL | Retrieval Passage | Only Positive | 0.274 | 0.470 | 0.720 | 0.191 |
| LLaMa3.1-8B-Legal-ThaiCCL-Combine | Retrieval Passage | Only Positive | 0.532 | 0.445 | 0.508 | 0.203 |
| Zero-shot LLaMa3.1-70B-Instruct | Retrieval Passage | Only Positive | 0.748 | 0.594 | 0.364 | 0.202 |
## License and use
The model is released under [Meta's Llama 3.1 Community License Agreement](https://www.llama.com/llama3_1/license/). Llama 3.1 is licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. | [
"CRAFT"
] |
TheBloke/med42-70B-AWQ | TheBloke | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"m42",
"health",
"healthcare",
"clinical-llm",
"en",
"base_model:m42-health/med42-70b",
"base_model:quantized:m42-health/med42-70b",
"license:other",
"autotrain_compatible",
"text-generation-inference",
"4-bit",
"awq",
"region:us"
] | "2023-10-27T22:47:52Z" | 2023-11-09T18:16:38+00:00 | 449 | 2 | ---
base_model: m42-health/med42-70b
language:
- en
license: other
license_name: med42
model_name: Med42 70B
pipeline_tag: text-generation
tags:
- m42
- health
- healthcare
- clinical-llm
inference: false
model_creator: M42 Health
model_type: llama
prompt_template: '<|system|>: You are a helpful medical assistant created by M42 Health
in the UAE.
<|prompter|>:{prompt}
<|assistant|>:
'
quantized_by: TheBloke
---
<!-- markdownlint-disable MD041 -->
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# Med42 70B - AWQ
- Model creator: [M42 Health](https://huggingface.co/m42-health)
- Original model: [Med42 70B](https://huggingface.co/m42-health/med42-70b)
<!-- description start -->
## Description
This repo contains AWQ model files for [M42 Health's Med42 70B](https://huggingface.co/m42-health/med42-70b).
These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
### About AWQ
AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference with equivalent or better quality compared to the most commonly used GPTQ settings.
It is supported by:
- [Text Generation Webui](https://github.com/oobabooga/text-generation-webui) - using Loader: AutoAWQ
- [vLLM](https://github.com/vllm-project/vllm) - Llama and Mistral models only
- [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)
- [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) - for use from Python code
<!-- description end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/med42-70B-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/med42-70B-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/med42-70B-GGUF)
* [M42 Health's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/m42-health/med42-70b)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: Med42
```
<|system|>: You are a helpful medical assistant created by M42 Health in the UAE.
<|prompter|>:{prompt}
<|assistant|>:
```
<!-- prompt-template end -->
<!-- licensing start -->
## Licensing
The creator of the source model has listed its license as `other`, and this quantization has therefore used that same license.
As this model is based on Llama 2, it is also subject to the Meta Llama 2 license terms, and the license files for that are additionally included. It should therefore be considered as being claimed to be licensed under both licenses. I contacted Hugging Face for clarification on dual licensing but they do not yet have an official position. Should this change, or should Meta provide any feedback on this situation, I will update this section accordingly.
In the meantime, any questions regarding licensing, and in particular how these two licenses might interact, should be directed to the original model repository: [M42 Health's Med42 70B](https://huggingface.co/m42-health/med42-70b).
<!-- licensing end -->
<!-- README_AWQ.md-provided-files start -->
## Provided files, and AWQ parameters
For my first release of AWQ models, I am releasing 128g models only. I will consider adding 32g as well if there is interest, and once I have done perplexity and evaluation comparisons, but at this time 32g models are still not fully tested with AutoAWQ and vLLM.
Models are released as sharded safetensors files.
| Branch | Bits | GS | AWQ Dataset | Seq Len | Size |
| ------ | ---- | -- | ----------- | ------- | ---- |
| [main](https://huggingface.co/TheBloke/med42-70B-AWQ/tree/main) | 4 | 128 | [Medical Meadow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc) | 4096 | 36.61 GB
<!-- README_AWQ.md-provided-files end -->
<!-- README_AWQ.md-text-generation-webui start -->
## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.
1. Click the **Model tab**.
2. Under **Download custom model or LoRA**, enter `TheBloke/med42-70B-AWQ`.
3. Click **Download**.
4. The model will start downloading. Once it's finished it will say "Done".
5. In the top left, click the refresh icon next to **Model**.
6. In the **Model** dropdown, choose the model you just downloaded: `med42-70B-AWQ`
7. Select **Loader: AutoAWQ**.
8. Click Load, and the model will load and is now ready for use.
9. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right.
10. Once you're ready, click the **Text Generation** tab and enter a prompt to get started!
<!-- README_AWQ.md-text-generation-webui end -->
<!-- README_AWQ.md-use-from-vllm start -->
## Multi-user inference server: vLLM
Documentation on installing and using vLLM [can be found here](https://vllm.readthedocs.io/en/latest/).
- Please ensure you are using vLLM version 0.2 or later.
- When using vLLM as a server, pass the `--quantization awq` parameter.
For example:
```shell
python3 python -m vllm.entrypoints.api_server --model TheBloke/med42-70B-AWQ --quantization awq
```
- When using vLLM from Python code, again set `quantization=awq`.
For example:
```python
from vllm import LLM, SamplingParams
prompts = [
"Tell me about AI",
"Write a story about llamas",
"What is 291 - 150?",
"How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
]
prompt_template=f'''<|system|>: You are a helpful medical assistant created by M42 Health in the UAE.
<|prompter|>:{prompt}
<|assistant|>:
'''
prompts = [prompt_template.format(prompt=prompt) for prompt in prompts]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
llm = LLM(model="TheBloke/med42-70B-AWQ", quantization="awq", dtype="auto")
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
<!-- README_AWQ.md-use-from-vllm start -->
<!-- README_AWQ.md-use-from-tgi start -->
## Multi-user inference server: Hugging Face Text Generation Inference (TGI)
Use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0`
Example Docker parameters:
```shell
--model-id TheBloke/med42-70B-AWQ --port 3000 --quantize awq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096
```
Example Python code for interfacing with TGI (requires [huggingface-hub](https://github.com/huggingface/huggingface_hub) 0.17.0 or later):
```shell
pip3 install huggingface-hub
```
```python
from huggingface_hub import InferenceClient
endpoint_url = "https://your-endpoint-url-here"
prompt = "Tell me about AI"
prompt_template=f'''<|system|>: You are a helpful medical assistant created by M42 Health in the UAE.
<|prompter|>:{prompt}
<|assistant|>:
'''
client = InferenceClient(endpoint_url)
response = client.text_generation(prompt,
max_new_tokens=128,
do_sample=True,
temperature=0.7,
top_p=0.95,
top_k=40,
repetition_penalty=1.1)
print(f"Model output: ", response)
```
<!-- README_AWQ.md-use-from-tgi end -->
<!-- README_AWQ.md-use-from-python start -->
## Inference from Python code using AutoAWQ
### Install the AutoAWQ package
Requires: [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) 0.1.1 or later.
```shell
pip3 install autoawq
```
If you have problems installing [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) using the pre-built wheels, install it from source instead:
```shell
pip3 uninstall -y autoawq
git clone https://github.com/casper-hansen/AutoAWQ
cd AutoAWQ
pip3 install .
```
### AutoAWQ example code
```python
from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer
model_name_or_path = "TheBloke/med42-70B-AWQ"
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=False)
# Load model
model = AutoAWQForCausalLM.from_quantized(model_name_or_path, fuse_layers=True,
trust_remote_code=False, safetensors=True)
prompt = "Tell me about AI"
prompt_template=f'''<|system|>: You are a helpful medical assistant created by M42 Health in the UAE.
<|prompter|>:{prompt}
<|assistant|>:
'''
print("*** Running model.generate:")
token_input = tokenizer(
prompt_template,
return_tensors='pt'
).input_ids.cuda()
# Generate output
generation_output = model.generate(
token_input,
do_sample=True,
temperature=0.7,
top_p=0.95,
top_k=40,
max_new_tokens=512
)
# Get the tokens from the output, decode them, print them
token_output = generation_output[0]
text_output = tokenizer.decode(token_output)
print("LLM output: ", text_output)
"""
# Inference should be possible with transformers pipeline as well in future
# But currently this is not yet supported by AutoAWQ (correct as of September 25th 2023)
from transformers import pipeline
print("*** Pipeline:")
pipe = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=512,
do_sample=True,
temperature=0.7,
top_p=0.95,
top_k=40,
repetition_penalty=1.1
)
print(pipe(prompt_template)[0]['generated_text'])
"""
```
<!-- README_AWQ.md-use-from-python end -->
<!-- README_AWQ.md-compatibility start -->
## Compatibility
The files provided are tested to work with:
- [text-generation-webui](https://github.com/oobabooga/text-generation-webui) using `Loader: AutoAWQ`.
- [vLLM](https://github.com/vllm-project/vllm) version 0.2.0 and later.
- [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) version 1.1.0 and later.
- [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) version 0.1.1 and later.
<!-- README_AWQ.md-compatibility end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Pierre Kircher, Stanislav Ovsiannikov, Michael Levine, Eugene Pentland, Andrey, 준교 김, Randy H, Fred von Graf, Artur Olbinski, Caitlyn Gatomon, terasurfer, Jeff Scroggin, James Bentley, Vadim, Gabriel Puliatti, Harry Royden McLaughlin, Sean Connelly, Dan Guido, Edmond Seymore, Alicia Loh, subjectnull, AzureBlack, Manuel Alberto Morcote, Thomas Belote, Lone Striker, Chris Smitley, Vitor Caleffi, Johann-Peter Hartmann, Clay Pascal, biorpg, Brandon Frisco, sidney chen, transmissions 11, Pedro Madruga, jinyuan sun, Ajan Kanaga, Emad Mostaque, Trenton Dambrowitz, Jonathan Leane, Iucharbius, usrbinkat, vamX, George Stoitzev, Luke Pendergrass, theTransient, Olakabola, Swaroop Kallakuri, Cap'n Zoog, Brandon Phillips, Michael Dempsey, Nikolai Manek, danny, Matthew Berman, Gabriel Tamborski, alfie_i, Raymond Fosdick, Tom X Nguyen, Raven Klaugh, LangChain4j, Magnesian, Illia Dulskyi, David Ziegler, Mano Prime, Luis Javier Navarrete Lozano, Erik Bjäreholt, 阿明, Nathan Dryer, Alex, Rainer Wilmers, zynix, TL, Joseph William Delisle, John Villwock, Nathan LeClaire, Willem Michiel, Joguhyik, GodLy, OG, Alps Aficionado, Jeffrey Morgan, ReadyPlayerEmma, Tiffany J. Kim, Sebastain Graf, Spencer Kim, Michael Davis, webtim, Talal Aujan, knownsqashed, John Detwiler, Imad Khwaja, Deo Leter, Jerry Meng, Elijah Stavena, Rooh Singh, Pieter, SuperWojo, Alexandros Triantafyllidis, Stephen Murray, Ai Maven, ya boyyy, Enrico Ros, Ken Nordquist, Deep Realms, Nicholas, Spiking Neurons AB, Elle, Will Dee, Jack West, RoA, Luke @flexchar, Viktor Bowallius, Derek Yates, Subspace Studios, jjj, Toran Billups, Asp the Wyvern, Fen Risland, Ilya, NimbleBox.ai, Chadd, Nitin Borwankar, Emre, Mandus, Leonard Tan, Kalila, K, Trailburnt, S_X, Cory Kujawski
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
# Original model card: M42 Health's Med42 70B
# **Med42 - Clinical Large Language Model**
Med42 is an open-access clinical large language model (LLM) developed by M42 to expand access to medical knowledge. Built off LLaMA-2 and comprising 70 billion parameters, this generative AI system provides high-quality answers to medical questions.
## Model Details
*Note: Use of this model is governed by the M42 Health license. In order to download the model weights (and tokenizer), please read the [Med42 License](https://huggingface.co/spaces/m42-health/License) and accept our License by requesting access here.*
Beginning with the base LLaMa-2 model, Med42 was instruction-tuned on a dataset of ~250M tokens compiled from different open-access sources, including medical flashcards, exam questions, and open-domain dialogues.
**Model Developers:** M42 Health AI Team
**Finetuned from model:** Llama-2 - 70B
**Context length:** 4k tokens
**Input:** Text only data
**Output:** Model generates text only
**Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we enhance model's performance.
**License:** A custom license is available [here](https://huggingface.co/spaces/m42-health/License)
**Research Paper:** TBA
## Intended Use
Med42 is being made available for further testing and assessment as an AI assistant to enhance clinical decision-making and enhance access to an LLM for healthcare use. Potential use cases include:
- Medical question answering
- Patient record summarization
- Aiding medical diagnosis
- General health Q&A
To get the expected features and performance for the model, a specific formatting needs to be followed, including the `<|system|>`, `<|prompter|>` and `<|assistant|>` tags.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name_or_path = "m42-health/med42-70b"
model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
prompt = "What are the symptoms of diabetes ?"
prompt_template=f'''
<|system|>: You are a helpful medical assistant created by M42 Health in the UAE.
<|prompter|>:{prompt}
<|assistant|>:
'''
input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True,eos_token_id=tokenizer.eos_token_id, pad_token_id=tokenizer.pad_token_id, max_new_tokens=512)
print(tokenizer.decode(output[0]))
```
## Hardware and Software
The training process was performed on the Condor Galaxy 1 (CG-1) supercomputer platform.
## Evaluation Results
Med42 achieves achieves competitive performance on various medical benchmarks, including MedQA, MedMCQA, PubMedQA, HeadQA, and Measuring Massive Multitask Language Understanding (MMLU) clinical topics. For all evaluations reported so far, we use [EleutherAI's evaluation harness library](https://github.com/EleutherAI/lm-evaluation-harness) and report zero-shot accuracies (except otherwise stated). We compare the performance with that reported for other models (ClinicalCamel-70B, GPT-3.5, GPT-4.0, Med-PaLM 2).
|Dataset|Med42|ClinicalCamel-70B|GPT-3.5|GPT-4.0|Med-PaLM-2 (5-shot)*|
|---|---|---|---|---|---|
|MMLU Clinical Knowledge|74.3|69.8|69.8|86.0|88.3|
|MMLU College Biology|84.0|79.2|72.2|95.1|94.4|
|MMLU College Medicine|68.8|67.0|61.3|76.9|80.9|
|MMLU Medical Genetics|86.0|69.0|70.0|91.0|90.0|
|MMLU Professional Medicine|79.8|71.3|70.2|93.0|95.2|
|MMLU Anatomy|67.4|62.2|56.3|80.0|77.8|
|MedMCQA|60.9|47.0|50.1|69.5|71.3|
|MedQA|61.5|53.4|50.8|78.9|79.7|
|USMLE Self-Assessment|71.7|-|49.1|83.8|-|
|USMLE Sample Exam|72.0|54.3|56.9|84.3|-|
**We note that 0-shot performance is not reported for Med-PaLM 2. Further details can be found at [https://github.com/m42health/med42](https://github.com/m42health/med42)*.
### Key performance metrics:
- Med42 achieves a 72% accuracy on the US Medical Licensing Examination (USMLE) sample exam, surpassing the prior state of the art among openly available medical LLMs.
- 61.5% on MedQA dataset (compared to 50.8% for GPT-3.5)
- Consistently higher performance on MMLU clinical topics compared to GPT-3.5.
## Limitations & Safe Use
- Med42 is not ready for real clinical use. Extensive human evaluation is undergoing as it is required to ensure safety.
- Potential for generating incorrect or harmful information.
- Risk of perpetuating biases in training data.
Use this model responsibly! Do not rely on it for medical usage without rigorous safety testing.
## Accessing Med42 and Reporting Issues
Please report any software "bug" or other problems through one of the following means:
- Reporting issues with the model: [https://github.com/m42health/med42](https://github.com/m42health/med42)
- Reporting risky content generated by the model, bugs and/or any security concerns: [https://forms.office.com/r/YMJu3kcKat](https://forms.office.com/r/YMJu3kcKat)
- M42’s privacy policy available at [https://m42.ae/privacy-policy/](https://m42.ae/privacy-policy/)
- Reporting violations of the Acceptable Use Policy or unlicensed uses of Med42: <med42@m42.ae>
| [
"MEDQA",
"PUBMEDQA"
] |
Omartificial-Intelligence-Space/Arabert-all-nli-triplet-Matryoshka | Omartificial-Intelligence-Space | sentence-similarity | [
"sentence-transformers",
"safetensors",
"bert",
"feature-extraction",
"mteb",
"transformers",
"sentence-similarity",
"generated_from_trainer",
"dataset_size:557850",
"loss:MatryoshkaLoss",
"loss:MultipleNegativesRankingLoss",
"ar",
"dataset:Omartificial-Intelligence-Space/Arabic-NLi-Triplet",
"arxiv:1908.10084",
"arxiv:2205.13147",
"arxiv:1705.00652",
"arxiv:2407.21139",
"base_model:aubmindlab/bert-base-arabertv02",
"base_model:finetune:aubmindlab/bert-base-arabertv02",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | "2024-06-16T23:18:49Z" | 2025-01-23T10:26:16+00:00 | 447 | 10 | ---
base_model: aubmindlab/bert-base-arabertv02
datasets:
- Omartificial-Intelligence-Space/Arabic-NLi-Triplet
language:
- ar
library_name: sentence-transformers
license: apache-2.0
metrics:
- pearson_cosine
- spearman_cosine
- pearson_manhattan
- spearman_manhattan
- pearson_euclidean
- spearman_euclidean
- pearson_dot
- spearman_dot
- pearson_max
- spearman_max
pipeline_tag: sentence-similarity
tags:
- mteb
- transformers
- sentence-transformers
- sentence-similarity
- feature-extraction
- generated_from_trainer
- dataset_size:557850
- loss:MatryoshkaLoss
- loss:MultipleNegativesRankingLoss
widget:
- source_sentence: ذكر متوازن بعناية يقف على قدم واحدة بالقرب من منطقة شاطئ المحيط
النظيفة
sentences:
- رجل يقدم عرضاً
- هناك رجل بالخارج قرب الشاطئ
- رجل يجلس على أريكه
- source_sentence: رجل يقفز إلى سريره القذر
sentences:
- السرير قذر.
- رجل يضحك أثناء غسيل الملابس
- الرجل على القمر
- source_sentence: الفتيات بالخارج
sentences:
- امرأة تلف الخيط إلى كرات بجانب كومة من الكرات
- فتيان يركبان في جولة متعة
- ثلاث فتيات يقفون سوية في غرفة واحدة تستمع وواحدة تكتب على الحائط والثالثة تتحدث
إليهن
- source_sentence: الرجل يرتدي قميصاً أزرق.
sentences:
- رجل يرتدي قميصاً أزرق يميل إلى الجدار بجانب الطريق مع شاحنة زرقاء وسيارة حمراء
مع الماء في الخلفية.
- كتاب القصص مفتوح
- رجل يرتدي قميص أسود يعزف على الجيتار.
- source_sentence: يجلس شاب ذو شعر أشقر على الحائط يقرأ جريدة بينما تمر امرأة وفتاة
شابة.
sentences:
- ذكر شاب ينظر إلى جريدة بينما تمر إمرأتان بجانبه
- رجل يستلقي على وجهه على مقعد في الحديقة.
- الشاب نائم بينما الأم تقود ابنتها إلى الحديقة
model-index:
- name: SentenceTransformer based on aubmindlab/bert-base-arabertv02
results:
- task:
type: Retrieval
dataset:
name: MTEB MIRACLRetrieval (ar)
type: miracl/mmteb-miracl
config: ar
split: dev
revision: main
metrics:
- type: ndcg_at_1
value: 9.289
- type: ndcg_at_3
value: 12.42
- type: ndcg_at_5
value: 14.407
- type: ndcg_at_10
value: 17.709
- type: ndcg_at_20
value: 20.389
- type: ndcg_at_100
value: 24.847
- type: ndcg_at_1000
value: 28.494999999999997
- type: map_at_1
value: 6.226
- type: map_at_3
value: 9.898
- type: map_at_5
value: 11.118
- type: map_at_10
value: 12.556000000000001
- type: map_at_20
value: 13.395000000000001
- type: map_at_100
value: 14.11
- type: map_at_1000
value: 14.285
- type: recall_at_1
value: 6.226
- type: recall_at_3
value: 14.374
- type: recall_at_5
value: 19.127
- type: recall_at_10
value: 27.929
- type: recall_at_20
value: 36.895
- type: recall_at_100
value: 56.931
- type: recall_at_1000
value: 81.08999999999999
- type: precision_at_1
value: 9.289
- type: precision_at_3
value: 7.550999999999999
- type: precision_at_5
value: 6.236
- type: precision_at_10
value: 4.786
- type: precision_at_20
value: 3.248
- type: precision_at_100
value: 1.076
- type: precision_at_1000
value: 0.159
- type: mrr_at_1
value: 9.2887
- type: mrr_at_3
value: 14.3646
- type: mrr_at_5
value: 15.9012
- type: mrr_at_10
value: 17.4156
- type: mrr_at_20
value: 18.124399999999998
- type: mrr_at_100
value: 18.618199999999998
- type: mrr_at_1000
value: 18.6982
- type: nauc_ndcg_at_1_max
value: -0.6867
- type: nauc_ndcg_at_1_std
value: -7.9873
- type: nauc_ndcg_at_1_diff1
value: 15.4777
- type: nauc_ndcg_at_3_max
value: -1.0088
- type: nauc_ndcg_at_3_std
value: -8.7872
- type: nauc_ndcg_at_3_diff1
value: 10.342500000000001
- type: nauc_ndcg_at_5_max
value: 0.7207
- type: nauc_ndcg_at_5_std
value: -6.0446
- type: nauc_ndcg_at_5_diff1
value: 10.8456
- type: nauc_ndcg_at_10_max
value: 1.6348000000000003
- type: nauc_ndcg_at_10_std
value: -3.3235
- type: nauc_ndcg_at_10_diff1
value: 9.7106
- type: nauc_ndcg_at_20_max
value: 3.3129
- type: nauc_ndcg_at_20_std
value: -1.1822
- type: nauc_ndcg_at_20_diff1
value: 9.6828
- type: nauc_ndcg_at_100_max
value: 6.8271
- type: nauc_ndcg_at_100_std
value: 3.4901
- type: nauc_ndcg_at_100_diff1
value: 10.205
- type: nauc_ndcg_at_1000_max
value: 7.7488
- type: nauc_ndcg_at_1000_std
value: 4.9253
- type: nauc_ndcg_at_1000_diff1
value: 10.5103
- type: nauc_map_at_1_max
value: -3.2569
- type: nauc_map_at_1_std
value: -11.8583
- type: nauc_map_at_1_diff1
value: 17.8176
- type: nauc_map_at_3_max
value: -2.3331
- type: nauc_map_at_3_std
value: -10.345500000000001
- type: nauc_map_at_3_diff1
value: 12.136
- type: nauc_map_at_5_max
value: -0.9544
- type: nauc_map_at_5_std
value: -8.3844
- type: nauc_map_at_5_diff1
value: 12.4093
- type: nauc_map_at_10_max
value: -0.2657
- type: nauc_map_at_10_std
value: -6.693200000000001
- type: nauc_map_at_10_diff1
value: 11.6888
- type: nauc_map_at_20_max
value: 0.5226
- type: nauc_map_at_20_std
value: -5.6443
- type: nauc_map_at_20_diff1
value: 11.5943
- type: nauc_map_at_100_max
value: 1.2930000000000001
- type: nauc_map_at_100_std
value: -4.5427
- type: nauc_map_at_100_diff1
value: 11.6536
- type: nauc_map_at_1000_max
value: 1.4096
- type: nauc_map_at_1000_std
value: -4.3770999999999995
- type: nauc_map_at_1000_diff1
value: 11.6872
- type: nauc_recall_at_1_max
value: -3.2569
- type: nauc_recall_at_1_std
value: -11.8583
- type: nauc_recall_at_1_diff1
value: 17.8176
- type: nauc_recall_at_3_max
value: -2.177
- type: nauc_recall_at_3_std
value: -9.099400000000001
- type: nauc_recall_at_3_diff1
value: 7.1512
- type: nauc_recall_at_5_max
value: 1.1292
- type: nauc_recall_at_5_std
value: -4.4891
- type: nauc_recall_at_5_diff1
value: 8.530899999999999
- type: nauc_recall_at_10_max
value: 2.0878
- type: nauc_recall_at_10_std
value: 0.0957
- type: nauc_recall_at_10_diff1
value: 6.149
- type: nauc_recall_at_20_max
value: 5.3045
- type: nauc_recall_at_20_std
value: 4.0603
- type: nauc_recall_at_20_diff1
value: 5.9809
- type: nauc_recall_at_100_max
value: 14.7914
- type: nauc_recall_at_100_std
value: 17.085
- type: nauc_recall_at_100_diff1
value: 7.1123
- type: nauc_recall_at_1000_max
value: 24.1037
- type: nauc_recall_at_1000_std
value: 33.216499999999996
- type: nauc_recall_at_1000_diff1
value: 7.925400000000001
- type: nauc_precision_at_1_max
value: -0.6867
- type: nauc_precision_at_1_std
value: -7.9873
- type: nauc_precision_at_1_diff1
value: 15.4777
- type: nauc_precision_at_3_max
value: 1.8041999999999998
- type: nauc_precision_at_3_std
value: -5.2127
- type: nauc_precision_at_3_diff1
value: 7.3027
- type: nauc_precision_at_5_max
value: 5.5463
- type: nauc_precision_at_5_std
value: 0.8853
- type: nauc_precision_at_5_diff1
value: 7.3189
- type: nauc_precision_at_10_max
value: 8.8561
- type: nauc_precision_at_10_std
value: 7.078900000000001
- type: nauc_precision_at_10_diff1
value: 5.2272
- type: nauc_precision_at_20_max
value: 12.432
- type: nauc_precision_at_20_std
value: 12.006699999999999
- type: nauc_precision_at_20_diff1
value: 5.0414
- type: nauc_precision_at_100_max
value: 20.3992
- type: nauc_precision_at_100_std
value: 23.5259
- type: nauc_precision_at_100_diff1
value: 5.0255
- type: nauc_precision_at_1000_max
value: 22.0358
- type: nauc_precision_at_1000_std
value: 26.360099999999996
- type: nauc_precision_at_1000_diff1
value: 3.1912999999999996
- type: nauc_mrr_at_1_max
value: -0.6867
- type: nauc_mrr_at_1_std
value: -7.9873
- type: nauc_mrr_at_1_diff1
value: 15.4777
- type: nauc_mrr_at_3_max
value: 0.6054999999999999
- type: nauc_mrr_at_3_std
value: -6.7876
- type: nauc_mrr_at_3_diff1
value: 10.6434
- type: nauc_mrr_at_5_max
value: 1.7145000000000001
- type: nauc_mrr_at_5_std
value: -4.9459
- type: nauc_mrr_at_5_diff1
value: 10.731499999999999
- type: nauc_mrr_at_10_max
value: 2.3058
- type: nauc_mrr_at_10_std
value: -3.6794000000000002
- type: nauc_mrr_at_10_diff1
value: 10.328800000000001
- type: nauc_mrr_at_20_max
value: 2.7305
- type: nauc_mrr_at_20_std
value: -3.3355999999999995
- type: nauc_mrr_at_20_diff1
value: 10.5801
- type: nauc_mrr_at_100_max
value: 3.1306000000000003
- type: nauc_mrr_at_100_std
value: -2.8568
- type: nauc_mrr_at_100_diff1
value: 10.6458
- type: nauc_mrr_at_1000_max
value: 3.0882
- type: nauc_mrr_at_1000_std
value: -2.8863
- type: nauc_mrr_at_1000_diff1
value: 10.6507
- type: main_score
value: 17.709
- task:
type: Retrieval
dataset:
name: MTEB MIRACLRetrievalHardNegatives (ar)
type: mteb/miracl-hard-negatives
config: ar
split: dev
revision: 95c8db7d4a6e9c1d8a60601afd63d553ae20a2eb
metrics:
- type: ndcg_at_1
value: 12.5
- type: ndcg_at_3
value: 16.058
- type: ndcg_at_5
value: 18.833
- type: ndcg_at_10
value: 22.583000000000002
- type: ndcg_at_20
value: 25.974000000000004
- type: ndcg_at_100
value: 32.359
- type: ndcg_at_1000
value: 35.278999999999996
- type: map_at_1
value: 8.211
- type: map_at_3
value: 12.913
- type: map_at_5
value: 14.621999999999998
- type: map_at_10
value: 16.314999999999998
- type: map_at_20
value: 17.423
- type: map_at_100
value: 18.522
- type: map_at_1000
value: 18.693
- type: recall_at_1
value: 8.211
- type: recall_at_3
value: 18.474
- type: recall_at_5
value: 24.969
- type: recall_at_10
value: 34.894
- type: recall_at_20
value: 45.672000000000004
- type: recall_at_100
value: 74.453
- type: recall_at_1000
value: 93.162
- type: precision_at_1
value: 12.5
- type: precision_at_3
value: 9.700000000000001
- type: precision_at_5
value: 8.24
- type: precision_at_10
value: 6.069999999999999
- type: precision_at_20
value: 4.22
- type: precision_at_100
value: 1.456
- type: precision_at_1000
value: 0.186
- type: mrr_at_1
value: 12.5
- type: mrr_at_3
value: 18.5333
- type: mrr_at_5
value: 20.5983
- type: mrr_at_10
value: 22.165000000000003
- type: mrr_at_20
value: 23.0466
- type: mrr_at_100
value: 23.6519
- type: mrr_at_1000
value: 23.7052
- type: nauc_ndcg_at_1_max
value: -4.1030999999999995
- type: nauc_ndcg_at_1_std
value: -5.306
- type: nauc_ndcg_at_1_diff1
value: 14.2036
- type: nauc_ndcg_at_3_max
value: -2.0753
- type: nauc_ndcg_at_3_std
value: -8.209800000000001
- type: nauc_ndcg_at_3_diff1
value: 13.8408
- type: nauc_ndcg_at_5_max
value: -0.3815
- type: nauc_ndcg_at_5_std
value: -6.2694
- type: nauc_ndcg_at_5_diff1
value: 13.717699999999999
- type: nauc_ndcg_at_10_max
value: 0.11460000000000001
- type: nauc_ndcg_at_10_std
value: -4.737
- type: nauc_ndcg_at_10_diff1
value: 13.524
- type: nauc_ndcg_at_20_max
value: 1.7666000000000002
- type: nauc_ndcg_at_20_std
value: -3.8832
- type: nauc_ndcg_at_20_diff1
value: 13.2796
- type: nauc_ndcg_at_100_max
value: 5.0085
- type: nauc_ndcg_at_100_std
value: 0.4544
- type: nauc_ndcg_at_100_diff1
value: 12.401
- type: nauc_ndcg_at_1000_max
value: 5.0894
- type: nauc_ndcg_at_1000_std
value: 0.5319
- type: nauc_ndcg_at_1000_diff1
value: 13.4741
- type: nauc_map_at_1_max
value: -5.8795
- type: nauc_map_at_1_std
value: -9.908999999999999
- type: nauc_map_at_1_diff1
value: 17.0078
- type: nauc_map_at_3_max
value: -3.5836
- type: nauc_map_at_3_std
value: -9.495000000000001
- type: nauc_map_at_3_diff1
value: 14.9483
- type: nauc_map_at_5_max
value: -2.3598
- type: nauc_map_at_5_std
value: -8.098600000000001
- type: nauc_map_at_5_diff1
value: 14.963899999999999
- type: nauc_map_at_10_max
value: -2.0040999999999998
- type: nauc_map_at_10_std
value: -7.2158
- type: nauc_map_at_10_diff1
value: 14.786299999999999
- type: nauc_map_at_20_max
value: -1.3743
- type: nauc_map_at_20_std
value: -6.732
- type: nauc_map_at_20_diff1
value: 14.454600000000001
- type: nauc_map_at_100_max
value: -0.5892
- type: nauc_map_at_100_std
value: -5.782
- type: nauc_map_at_100_diff1
value: 14.1523
- type: nauc_map_at_1000_max
value: -0.47939999999999994
- type: nauc_map_at_1000_std
value: -5.6652000000000005
- type: nauc_map_at_1000_diff1
value: 14.236099999999999
- type: nauc_recall_at_1_max
value: -5.8795
- type: nauc_recall_at_1_std
value: -9.908999999999999
- type: nauc_recall_at_1_diff1
value: 17.0078
- type: nauc_recall_at_3_max
value: -1.9456999999999998
- type: nauc_recall_at_3_std
value: -9.8194
- type: nauc_recall_at_3_diff1
value: 12.6027
- type: nauc_recall_at_5_max
value: 0.8479000000000001
- type: nauc_recall_at_5_std
value: -6.040100000000001
- type: nauc_recall_at_5_diff1
value: 12.3169
- type: nauc_recall_at_10_max
value: 1.4895
- type: nauc_recall_at_10_std
value: -2.6813
- type: nauc_recall_at_10_diff1
value: 12.182500000000001
- type: nauc_recall_at_20_max
value: 4.8476
- type: nauc_recall_at_20_std
value: -1.2982
- type: nauc_recall_at_20_diff1
value: 12.1922
- type: nauc_recall_at_100_max
value: 16.8711
- type: nauc_recall_at_100_std
value: 15.7488
- type: nauc_recall_at_100_diff1
value: 8.4922
- type: nauc_recall_at_1000_max
value: 34.6438
- type: nauc_recall_at_1000_std
value: 37.9865
- type: nauc_recall_at_1000_diff1
value: 24.320800000000002
- type: nauc_precision_at_1_max
value: -4.1030999999999995
- type: nauc_precision_at_1_std
value: -5.306
- type: nauc_precision_at_1_diff1
value: 14.2036
- type: nauc_precision_at_3_max
value: 1.2384
- type: nauc_precision_at_3_std
value: -4.7199
- type: nauc_precision_at_3_diff1
value: 12.5113
- type: nauc_precision_at_5_max
value: 5.4619
- type: nauc_precision_at_5_std
value: 0.9998999999999999
- type: nauc_precision_at_5_diff1
value: 10.330599999999999
- type: nauc_precision_at_10_max
value: 8.101600000000001
- type: nauc_precision_at_10_std
value: 5.212
- type: nauc_precision_at_10_diff1
value: 8.1145
- type: nauc_precision_at_20_max
value: 11.136
- type: nauc_precision_at_20_std
value: 7.5323
- type: nauc_precision_at_20_diff1
value: 5.2908
- type: nauc_precision_at_100_max
value: 20.4696
- type: nauc_precision_at_100_std
value: 21.928800000000003
- type: nauc_precision_at_100_diff1
value: -0.5745
- type: nauc_precision_at_1000_max
value: 18.8294
- type: nauc_precision_at_1000_std
value: 20.999699999999997
- type: nauc_precision_at_1000_diff1
value: 0.40340000000000004
- type: nauc_mrr_at_1_max
value: -4.1030999999999995
- type: nauc_mrr_at_1_std
value: -5.306
- type: nauc_mrr_at_1_diff1
value: 14.2036
- type: nauc_mrr_at_3_max
value: -1.5056999999999998
- type: nauc_mrr_at_3_std
value: -6.223
- type: nauc_mrr_at_3_diff1
value: 12.9131
- type: nauc_mrr_at_5_max
value: 0.1196
- type: nauc_mrr_at_5_std
value: -4.1637
- type: nauc_mrr_at_5_diff1
value: 12.3498
- type: nauc_mrr_at_10_max
value: 0.2111
- type: nauc_mrr_at_10_std
value: -3.6927000000000003
- type: nauc_mrr_at_10_diff1
value: 12.179
- type: nauc_mrr_at_20_max
value: 0.9067999999999999
- type: nauc_mrr_at_20_std
value: -3.5138999999999996
- type: nauc_mrr_at_20_diff1
value: 12.313
- type: nauc_mrr_at_100_max
value: 1.0301
- type: nauc_mrr_at_100_std
value: -3.3045999999999998
- type: nauc_mrr_at_100_diff1
value: 12.3773
- type: nauc_mrr_at_1000_max
value: 0.9991
- type: nauc_mrr_at_1000_std
value: -3.3230000000000004
- type: nauc_mrr_at_1000_diff1
value: 12.376800000000001
- type: main_score
value: 22.583000000000002
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-ara)
type: facebook/mlqa
config: ara-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 50.29
- type: ndcg_at_3
value: 60.972
- type: ndcg_at_5
value: 63.102000000000004
- type: ndcg_at_10
value: 65.23400000000001
- type: ndcg_at_20
value: 66.506
- type: ndcg_at_100
value: 68.66
- type: ndcg_at_1000
value: 69.055
- type: map_at_1
value: 50.29
- type: map_at_3
value: 58.31699999999999
- type: map_at_5
value: 59.487
- type: map_at_10
value: 60.370000000000005
- type: map_at_20
value: 60.719
- type: map_at_100
value: 61.015
- type: map_at_1000
value: 61.034
- type: recall_at_1
value: 50.29
- type: recall_at_3
value: 68.66499999999999
- type: recall_at_5
value: 73.888
- type: recall_at_10
value: 80.464
- type: recall_at_20
value: 85.493
- type: recall_at_100
value: 97.099
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 50.29
- type: precision_at_3
value: 22.888
- type: precision_at_5
value: 14.777999999999999
- type: precision_at_10
value: 8.046000000000001
- type: precision_at_20
value: 4.275
- type: precision_at_100
value: 0.971
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 50.2901
- type: mrr_at_3
value: 58.3172
- type: mrr_at_5
value: 59.4874
- type: mrr_at_10
value: 60.3699
- type: mrr_at_20
value: 60.719
- type: mrr_at_100
value: 61.015299999999996
- type: mrr_at_1000
value: 61.0344
- type: nauc_ndcg_at_1_max
value: 45.2805
- type: nauc_ndcg_at_1_std
value: 0.0181
- type: nauc_ndcg_at_1_diff1
value: 65.3259
- type: nauc_ndcg_at_3_max
value: 52.225
- type: nauc_ndcg_at_3_std
value: 5.8812999999999995
- type: nauc_ndcg_at_3_diff1
value: 61.60679999999999
- type: nauc_ndcg_at_5_max
value: 53.290400000000005
- type: nauc_ndcg_at_5_std
value: 7.0203
- type: nauc_ndcg_at_5_diff1
value: 61.437
- type: nauc_ndcg_at_10_max
value: 54.74400000000001
- type: nauc_ndcg_at_10_std
value: 9.7049
- type: nauc_ndcg_at_10_diff1
value: 61.094899999999996
- type: nauc_ndcg_at_20_max
value: 54.3655
- type: nauc_ndcg_at_20_std
value: 9.504999999999999
- type: nauc_ndcg_at_20_diff1
value: 62.002500000000005
- type: nauc_ndcg_at_100_max
value: 53.162699999999994
- type: nauc_ndcg_at_100_std
value: 8.163
- type: nauc_ndcg_at_100_diff1
value: 62.004999999999995
- type: nauc_ndcg_at_1000_max
value: 52.550399999999996
- type: nauc_ndcg_at_1000_std
value: 7.113700000000001
- type: nauc_ndcg_at_1000_diff1
value: 62.16009999999999
- type: nauc_map_at_1_max
value: 45.2805
- type: nauc_map_at_1_std
value: 0.0181
- type: nauc_map_at_1_diff1
value: 65.3259
- type: nauc_map_at_3_max
value: 50.4866
- type: nauc_map_at_3_std
value: 4.1894
- type: nauc_map_at_3_diff1
value: 62.62520000000001
- type: nauc_map_at_5_max
value: 51.047399999999996
- type: nauc_map_at_5_std
value: 4.7825
- type: nauc_map_at_5_diff1
value: 62.5698
- type: nauc_map_at_10_max
value: 51.505100000000006
- type: nauc_map_at_10_std
value: 5.6847
- type: nauc_map_at_10_diff1
value: 62.40710000000001
- type: nauc_map_at_20_max
value: 51.3852
- type: nauc_map_at_20_std
value: 5.5943
- type: nauc_map_at_20_diff1
value: 62.6332
- type: nauc_map_at_100_max
value: 51.2446
- type: nauc_map_at_100_std
value: 5.4548
- type: nauc_map_at_100_diff1
value: 62.6288
- type: nauc_map_at_1000_max
value: 51.2191
- type: nauc_map_at_1000_std
value: 5.4109
- type: nauc_map_at_1000_diff1
value: 62.634299999999996
- type: nauc_recall_at_1_max
value: 45.2805
- type: nauc_recall_at_1_std
value: 0.0181
- type: nauc_recall_at_1_diff1
value: 65.3259
- type: nauc_recall_at_3_max
value: 58.0831
- type: nauc_recall_at_3_std
value: 11.6994
- type: nauc_recall_at_3_diff1
value: 58.1295
- type: nauc_recall_at_5_max
value: 61.925799999999995
- type: nauc_recall_at_5_std
value: 15.798799999999998
- type: nauc_recall_at_5_diff1
value: 57.044799999999995
- type: nauc_recall_at_10_max
value: 71.2178
- type: nauc_recall_at_10_std
value: 30.915
- type: nauc_recall_at_10_diff1
value: 54.850100000000005
- type: nauc_recall_at_20_max
value: 73.5109
- type: nauc_recall_at_20_std
value: 36.0963
- type: nauc_recall_at_20_diff1
value: 59.7367
- type: nauc_recall_at_100_max
value: 89.58930000000001
- type: nauc_recall_at_100_std
value: 70.52619999999999
- type: nauc_recall_at_100_diff1
value: 52.489799999999995
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 45.2805
- type: nauc_precision_at_1_std
value: 0.0181
- type: nauc_precision_at_1_diff1
value: 65.3259
- type: nauc_precision_at_3_max
value: 58.0831
- type: nauc_precision_at_3_std
value: 11.6994
- type: nauc_precision_at_3_diff1
value: 58.1295
- type: nauc_precision_at_5_max
value: 61.925799999999995
- type: nauc_precision_at_5_std
value: 15.798799999999998
- type: nauc_precision_at_5_diff1
value: 57.044799999999995
- type: nauc_precision_at_10_max
value: 71.2178
- type: nauc_precision_at_10_std
value: 30.915
- type: nauc_precision_at_10_diff1
value: 54.850100000000005
- type: nauc_precision_at_20_max
value: 73.5109
- type: nauc_precision_at_20_std
value: 36.0963
- type: nauc_precision_at_20_diff1
value: 59.7367
- type: nauc_precision_at_100_max
value: 89.58930000000001
- type: nauc_precision_at_100_std
value: 70.52619999999999
- type: nauc_precision_at_100_diff1
value: 52.489799999999995
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 45.2805
- type: nauc_mrr_at_1_std
value: 0.0181
- type: nauc_mrr_at_1_diff1
value: 65.3259
- type: nauc_mrr_at_3_max
value: 50.4866
- type: nauc_mrr_at_3_std
value: 4.1894
- type: nauc_mrr_at_3_diff1
value: 62.62520000000001
- type: nauc_mrr_at_5_max
value: 51.047399999999996
- type: nauc_mrr_at_5_std
value: 4.7825
- type: nauc_mrr_at_5_diff1
value: 62.5698
- type: nauc_mrr_at_10_max
value: 51.505100000000006
- type: nauc_mrr_at_10_std
value: 5.6847
- type: nauc_mrr_at_10_diff1
value: 62.40710000000001
- type: nauc_mrr_at_20_max
value: 51.3852
- type: nauc_mrr_at_20_std
value: 5.5943
- type: nauc_mrr_at_20_diff1
value: 62.6332
- type: nauc_mrr_at_100_max
value: 51.2446
- type: nauc_mrr_at_100_std
value: 5.4548
- type: nauc_mrr_at_100_diff1
value: 62.6288
- type: nauc_mrr_at_1000_max
value: 51.2191
- type: nauc_mrr_at_1000_std
value: 5.4109
- type: nauc_mrr_at_1000_diff1
value: 62.634299999999996
- type: main_score
value: 65.23400000000001
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-deu)
type: facebook/mlqa
config: ara-deu
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 0.966
- type: ndcg_at_3
value: 3.6229999999999998
- type: ndcg_at_5
value: 5.64
- type: ndcg_at_10
value: 7.678
- type: ndcg_at_20
value: 10.109
- type: ndcg_at_100
value: 19.001
- type: ndcg_at_1000
value: 22.148
- type: map_at_1
value: 0.966
- type: map_at_3
value: 2.738
- type: map_at_5
value: 3.873
- type: map_at_10
value: 4.718
- type: map_at_20
value: 5.379
- type: map_at_100
value: 6.425
- type: map_at_1000
value: 6.593999999999999
- type: recall_at_1
value: 0.966
- type: recall_at_3
value: 6.279999999999999
- type: recall_at_5
value: 11.111
- type: recall_at_10
value: 17.391000000000002
- type: recall_at_20
value: 27.053
- type: recall_at_100
value: 77.778
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 0.966
- type: precision_at_3
value: 2.093
- type: precision_at_5
value: 2.222
- type: precision_at_10
value: 1.7389999999999999
- type: precision_at_20
value: 1.353
- type: precision_at_100
value: 0.7779999999999999
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 0.9662000000000001
- type: mrr_at_3
value: 2.7375
- type: mrr_at_5
value: 3.8728
- type: mrr_at_10
value: 4.718
- type: mrr_at_20
value: 5.379
- type: mrr_at_100
value: 6.4253
- type: mrr_at_1000
value: 6.5942
- type: nauc_ndcg_at_1_max
value: -44.7077
- type: nauc_ndcg_at_1_std
value: -44.7077
- type: nauc_ndcg_at_1_diff1
value: -4.5372
- type: nauc_ndcg_at_3_max
value: -30.044900000000002
- type: nauc_ndcg_at_3_std
value: -16.3138
- type: nauc_ndcg_at_3_diff1
value: 4.616499999999999
- type: nauc_ndcg_at_5_max
value: -34.3111
- type: nauc_ndcg_at_5_std
value: -22.1049
- type: nauc_ndcg_at_5_diff1
value: -1.9365
- type: nauc_ndcg_at_10_max
value: -33.617599999999996
- type: nauc_ndcg_at_10_std
value: -19.0105
- type: nauc_ndcg_at_10_diff1
value: 0.8742
- type: nauc_ndcg_at_20_max
value: -26.177099999999996
- type: nauc_ndcg_at_20_std
value: -12.6937
- type: nauc_ndcg_at_20_diff1
value: 5.4471
- type: nauc_ndcg_at_100_max
value: -23.236
- type: nauc_ndcg_at_100_std
value: -9.762500000000001
- type: nauc_ndcg_at_100_diff1
value: 2.9798
- type: nauc_ndcg_at_1000_max
value: -26.982699999999998
- type: nauc_ndcg_at_1000_std
value: -14.061399999999999
- type: nauc_ndcg_at_1000_diff1
value: 3.9429
- type: nauc_map_at_1_max
value: -44.7077
- type: nauc_map_at_1_std
value: -44.7077
- type: nauc_map_at_1_diff1
value: -4.5372
- type: nauc_map_at_3_max
value: -31.7699
- type: nauc_map_at_3_std
value: -19.6543
- type: nauc_map_at_3_diff1
value: 3.5395999999999996
- type: nauc_map_at_5_max
value: -34.6254
- type: nauc_map_at_5_std
value: -23.3293
- type: nauc_map_at_5_diff1
value: -1.3139
- type: nauc_map_at_10_max
value: -34.044000000000004
- type: nauc_map_at_10_std
value: -21.4667
- type: nauc_map_at_10_diff1
value: 0.6301
- type: nauc_map_at_20_max
value: -30.3898
- type: nauc_map_at_20_std
value: -18.2854
- type: nauc_map_at_20_diff1
value: 2.9196
- type: nauc_map_at_100_max
value: -29.4922
- type: nauc_map_at_100_std
value: -17.3755
- type: nauc_map_at_100_diff1
value: 2.7664999999999997
- type: nauc_map_at_1000_max
value: -29.830000000000002
- type: nauc_map_at_1000_std
value: -17.7603
- type: nauc_map_at_1000_diff1
value: 2.8049
- type: nauc_recall_at_1_max
value: -44.7077
- type: nauc_recall_at_1_std
value: -44.7077
- type: nauc_recall_at_1_diff1
value: -4.5372
- type: nauc_recall_at_3_max
value: -27.7891
- type: nauc_recall_at_3_std
value: -11.9456
- type: nauc_recall_at_3_diff1
value: 6.0247
- type: nauc_recall_at_5_max
value: -34.1557
- type: nauc_recall_at_5_std
value: -21.0171
- type: nauc_recall_at_5_diff1
value: -2.8583999999999996
- type: nauc_recall_at_10_max
value: -33.3562
- type: nauc_recall_at_10_std
value: -16.436700000000002
- type: nauc_recall_at_10_diff1
value: 1.0688
- type: nauc_recall_at_20_max
value: -21.4644
- type: nauc_recall_at_20_std
value: -6.7522
- type: nauc_recall_at_20_diff1
value: 8.3037
- type: nauc_recall_at_100_max
value: -11.3494
- type: nauc_recall_at_100_std
value: 4.0219
- type: nauc_recall_at_100_diff1
value: -0.2595
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: -44.7077
- type: nauc_precision_at_1_std
value: -44.7077
- type: nauc_precision_at_1_diff1
value: -4.5372
- type: nauc_precision_at_3_max
value: -27.7891
- type: nauc_precision_at_3_std
value: -11.9456
- type: nauc_precision_at_3_diff1
value: 6.0247
- type: nauc_precision_at_5_max
value: -34.1557
- type: nauc_precision_at_5_std
value: -21.0171
- type: nauc_precision_at_5_diff1
value: -2.8583999999999996
- type: nauc_precision_at_10_max
value: -33.3562
- type: nauc_precision_at_10_std
value: -16.436700000000002
- type: nauc_precision_at_10_diff1
value: 1.0688
- type: nauc_precision_at_20_max
value: -21.4644
- type: nauc_precision_at_20_std
value: -6.7522
- type: nauc_precision_at_20_diff1
value: 8.3037
- type: nauc_precision_at_100_max
value: -11.3494
- type: nauc_precision_at_100_std
value: 4.0219
- type: nauc_precision_at_100_diff1
value: -0.2595
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: -44.7077
- type: nauc_mrr_at_1_std
value: -44.7077
- type: nauc_mrr_at_1_diff1
value: -4.5372
- type: nauc_mrr_at_3_max
value: -31.7699
- type: nauc_mrr_at_3_std
value: -19.6543
- type: nauc_mrr_at_3_diff1
value: 3.5395999999999996
- type: nauc_mrr_at_5_max
value: -34.6254
- type: nauc_mrr_at_5_std
value: -23.3293
- type: nauc_mrr_at_5_diff1
value: -1.3139
- type: nauc_mrr_at_10_max
value: -34.044000000000004
- type: nauc_mrr_at_10_std
value: -21.4667
- type: nauc_mrr_at_10_diff1
value: 0.6301
- type: nauc_mrr_at_20_max
value: -30.3898
- type: nauc_mrr_at_20_std
value: -18.2854
- type: nauc_mrr_at_20_diff1
value: 2.9196
- type: nauc_mrr_at_100_max
value: -29.4922
- type: nauc_mrr_at_100_std
value: -17.3755
- type: nauc_mrr_at_100_diff1
value: 2.7664999999999997
- type: nauc_mrr_at_1000_max
value: -29.830000000000002
- type: nauc_mrr_at_1000_std
value: -17.7603
- type: nauc_mrr_at_1000_diff1
value: 2.8049
- type: main_score
value: 7.678
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-eng)
type: facebook/mlqa
config: ara-eng
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 2.708
- type: ndcg_at_3
value: 4.2139999999999995
- type: ndcg_at_5
value: 6.827
- type: ndcg_at_10
value: 10.234
- type: ndcg_at_20
value: 13.202
- type: ndcg_at_100
value: 18.62
- type: ndcg_at_1000
value: 23.307
- type: map_at_1
value: 2.708
- type: map_at_3
value: 3.804
- type: map_at_5
value: 5.244999999999999
- type: map_at_10
value: 6.666999999999999
- type: map_at_20
value: 7.5
- type: map_at_100
value: 8.169
- type: map_at_1000
value: 8.36
- type: recall_at_1
value: 2.708
- type: recall_at_3
value: 5.416
- type: recall_at_5
value: 11.799
- type: recall_at_10
value: 22.244
- type: recall_at_20
value: 33.849000000000004
- type: recall_at_100
value: 64.217
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 2.708
- type: precision_at_3
value: 1.805
- type: precision_at_5
value: 2.36
- type: precision_at_10
value: 2.2239999999999998
- type: precision_at_20
value: 1.6920000000000002
- type: precision_at_100
value: 0.642
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 2.7079
- type: mrr_at_3
value: 3.804
- type: mrr_at_5
value: 5.244999999999999
- type: mrr_at_10
value: 6.6674
- type: mrr_at_20
value: 7.5001999999999995
- type: mrr_at_100
value: 8.1688
- type: mrr_at_1000
value: 8.3597
- type: nauc_ndcg_at_1_max
value: 10.6266
- type: nauc_ndcg_at_1_std
value: 5.2812
- type: nauc_ndcg_at_1_diff1
value: 23.1004
- type: nauc_ndcg_at_3_max
value: 4.6738
- type: nauc_ndcg_at_3_std
value: 2.7851999999999997
- type: nauc_ndcg_at_3_diff1
value: 19.3925
- type: nauc_ndcg_at_5_max
value: 4.5083
- type: nauc_ndcg_at_5_std
value: 0.7295
- type: nauc_ndcg_at_5_diff1
value: 16.6812
- type: nauc_ndcg_at_10_max
value: 1.7111
- type: nauc_ndcg_at_10_std
value: 2.616
- type: nauc_ndcg_at_10_diff1
value: 11.7058
- type: nauc_ndcg_at_20_max
value: 2.1995
- type: nauc_ndcg_at_20_std
value: 5.2672
- type: nauc_ndcg_at_20_diff1
value: 11.9285
- type: nauc_ndcg_at_100_max
value: 2.2007
- type: nauc_ndcg_at_100_std
value: 9.5383
- type: nauc_ndcg_at_100_diff1
value: 11.5884
- type: nauc_ndcg_at_1000_max
value: 3.1725000000000003
- type: nauc_ndcg_at_1000_std
value: 6.281299999999999
- type: nauc_ndcg_at_1000_diff1
value: 13.100700000000002
- type: nauc_map_at_1_max
value: 10.6266
- type: nauc_map_at_1_std
value: 5.2812
- type: nauc_map_at_1_diff1
value: 23.1004
- type: nauc_map_at_3_max
value: 5.5484
- type: nauc_map_at_3_std
value: 3.3171
- type: nauc_map_at_3_diff1
value: 20.255200000000002
- type: nauc_map_at_5_max
value: 5.0303
- type: nauc_map_at_5_std
value: 1.4756
- type: nauc_map_at_5_diff1
value: 17.9658
- type: nauc_map_at_10_max
value: 3.3158
- type: nauc_map_at_10_std
value: 2.4996
- type: nauc_map_at_10_diff1
value: 14.785400000000001
- type: nauc_map_at_20_max
value: 3.5715999999999997
- type: nauc_map_at_20_std
value: 3.7656
- type: nauc_map_at_20_diff1
value: 14.791199999999998
- type: nauc_map_at_100_max
value: 3.605
- type: nauc_map_at_100_std
value: 4.457
- type: nauc_map_at_100_diff1
value: 14.636
- type: nauc_map_at_1000_max
value: 3.714
- type: nauc_map_at_1000_std
value: 4.3167
- type: nauc_map_at_1000_diff1
value: 14.784500000000001
- type: nauc_recall_at_1_max
value: 10.6266
- type: nauc_recall_at_1_std
value: 5.2812
- type: nauc_recall_at_1_diff1
value: 23.1004
- type: nauc_recall_at_3_max
value: 2.9438
- type: nauc_recall_at_3_std
value: 1.6771
- type: nauc_recall_at_3_diff1
value: 17.5783
- type: nauc_recall_at_5_max
value: 3.9315
- type: nauc_recall_at_5_std
value: -0.2412
- type: nauc_recall_at_5_diff1
value: 14.8877
- type: nauc_recall_at_10_max
value: -0.20309999999999997
- type: nauc_recall_at_10_std
value: 2.9946
- type: nauc_recall_at_10_diff1
value: 7.942399999999999
- type: nauc_recall_at_20_max
value: 0.7283000000000001
- type: nauc_recall_at_20_std
value: 7.439
- type: nauc_recall_at_20_diff1
value: 8.8412
- type: nauc_recall_at_100_max
value: 0.0955
- type: nauc_recall_at_100_std
value: 20.7782
- type: nauc_recall_at_100_diff1
value: 7.725600000000001
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 10.6266
- type: nauc_precision_at_1_std
value: 5.2812
- type: nauc_precision_at_1_diff1
value: 23.1004
- type: nauc_precision_at_3_max
value: 2.9438
- type: nauc_precision_at_3_std
value: 1.6771
- type: nauc_precision_at_3_diff1
value: 17.5783
- type: nauc_precision_at_5_max
value: 3.9315
- type: nauc_precision_at_5_std
value: -0.2412
- type: nauc_precision_at_5_diff1
value: 14.8877
- type: nauc_precision_at_10_max
value: -0.20309999999999997
- type: nauc_precision_at_10_std
value: 2.9946
- type: nauc_precision_at_10_diff1
value: 7.942399999999999
- type: nauc_precision_at_20_max
value: 0.7283000000000001
- type: nauc_precision_at_20_std
value: 7.439
- type: nauc_precision_at_20_diff1
value: 8.8412
- type: nauc_precision_at_100_max
value: 0.0955
- type: nauc_precision_at_100_std
value: 20.7782
- type: nauc_precision_at_100_diff1
value: 7.725600000000001
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 10.6266
- type: nauc_mrr_at_1_std
value: 5.2812
- type: nauc_mrr_at_1_diff1
value: 23.1004
- type: nauc_mrr_at_3_max
value: 5.5484
- type: nauc_mrr_at_3_std
value: 3.3171
- type: nauc_mrr_at_3_diff1
value: 20.255200000000002
- type: nauc_mrr_at_5_max
value: 5.0303
- type: nauc_mrr_at_5_std
value: 1.4756
- type: nauc_mrr_at_5_diff1
value: 17.9658
- type: nauc_mrr_at_10_max
value: 3.3158
- type: nauc_mrr_at_10_std
value: 2.4996
- type: nauc_mrr_at_10_diff1
value: 14.785400000000001
- type: nauc_mrr_at_20_max
value: 3.5715999999999997
- type: nauc_mrr_at_20_std
value: 3.7656
- type: nauc_mrr_at_20_diff1
value: 14.791199999999998
- type: nauc_mrr_at_100_max
value: 3.605
- type: nauc_mrr_at_100_std
value: 4.457
- type: nauc_mrr_at_100_diff1
value: 14.636
- type: nauc_mrr_at_1000_max
value: 3.714
- type: nauc_mrr_at_1000_std
value: 4.3167
- type: nauc_mrr_at_1000_diff1
value: 14.784500000000001
- type: main_score
value: 10.234
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-spa)
type: facebook/mlqa
config: ara-spa
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 1.242
- type: ndcg_at_3
value: 3.497
- type: ndcg_at_5
value: 5.583
- type: ndcg_at_10
value: 7.55
- type: ndcg_at_20
value: 9.883000000000001
- type: ndcg_at_100
value: 19.747999999999998
- type: ndcg_at_1000
value: 22.457
- type: map_at_1
value: 1.242
- type: map_at_3
value: 2.795
- type: map_at_5
value: 3.975
- type: map_at_10
value: 4.7620000000000005
- type: map_at_20
value: 5.389
- type: map_at_100
value: 6.618
- type: map_at_1000
value: 6.7780000000000005
- type: recall_at_1
value: 1.242
- type: recall_at_3
value: 5.59
- type: recall_at_5
value: 10.559000000000001
- type: recall_at_10
value: 16.77
- type: recall_at_20
value: 26.087
- type: recall_at_100
value: 81.366
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 1.242
- type: precision_at_3
value: 1.863
- type: precision_at_5
value: 2.112
- type: precision_at_10
value: 1.677
- type: precision_at_20
value: 1.304
- type: precision_at_100
value: 0.814
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 1.2422
- type: mrr_at_3
value: 2.795
- type: mrr_at_5
value: 3.9752
- type: mrr_at_10
value: 4.7623999999999995
- type: mrr_at_20
value: 5.3894
- type: mrr_at_100
value: 6.6175999999999995
- type: mrr_at_1000
value: 6.777800000000001
- type: nauc_ndcg_at_1_max
value: -12.445599999999999
- type: nauc_ndcg_at_1_std
value: -44.4624
- type: nauc_ndcg_at_1_diff1
value: 29.339199999999998
- type: nauc_ndcg_at_3_max
value: 11.4312
- type: nauc_ndcg_at_3_std
value: 0.993
- type: nauc_ndcg_at_3_diff1
value: 24.1361
- type: nauc_ndcg_at_5_max
value: 21.9937
- type: nauc_ndcg_at_5_std
value: 14.4561
- type: nauc_ndcg_at_5_diff1
value: 18.956999999999997
- type: nauc_ndcg_at_10_max
value: 29.3543
- type: nauc_ndcg_at_10_std
value: 16.750300000000003
- type: nauc_ndcg_at_10_diff1
value: 18.3077
- type: nauc_ndcg_at_20_max
value: 23.2834
- type: nauc_ndcg_at_20_std
value: 13.678399999999998
- type: nauc_ndcg_at_20_diff1
value: 16.358800000000002
- type: nauc_ndcg_at_100_max
value: 19.9569
- type: nauc_ndcg_at_100_std
value: 11.7888
- type: nauc_ndcg_at_100_diff1
value: 15.0894
- type: nauc_ndcg_at_1000_max
value: 20.7381
- type: nauc_ndcg_at_1000_std
value: 11.4354
- type: nauc_ndcg_at_1000_diff1
value: 15.881200000000002
- type: nauc_map_at_1_max
value: -12.445599999999999
- type: nauc_map_at_1_std
value: -44.4624
- type: nauc_map_at_1_diff1
value: 29.339199999999998
- type: nauc_map_at_3_max
value: 6.815200000000001
- type: nauc_map_at_3_std
value: -6.6357
- type: nauc_map_at_3_diff1
value: 24.1184
- type: nauc_map_at_5_max
value: 16.5725
- type: nauc_map_at_5_std
value: 6.4346
- type: nauc_map_at_5_diff1
value: 20.0389
- type: nauc_map_at_10_max
value: 21.2176
- type: nauc_map_at_10_std
value: 8.402
- type: nauc_map_at_10_diff1
value: 19.217000000000002
- type: nauc_map_at_20_max
value: 19.0886
- type: nauc_map_at_20_std
value: 7.749300000000001
- type: nauc_map_at_20_diff1
value: 18.1056
- type: nauc_map_at_100_max
value: 18.306
- type: nauc_map_at_100_std
value: 7.4771
- type: nauc_map_at_100_diff1
value: 17.4587
- type: nauc_map_at_1000_max
value: 18.3366
- type: nauc_map_at_1000_std
value: 7.4089
- type: nauc_map_at_1000_diff1
value: 17.5205
- type: nauc_recall_at_1_max
value: -12.445599999999999
- type: nauc_recall_at_1_std
value: -44.4624
- type: nauc_recall_at_1_diff1
value: 29.339199999999998
- type: nauc_recall_at_3_max
value: 18.5164
- type: nauc_recall_at_3_std
value: 12.569700000000001
- type: nauc_recall_at_3_diff1
value: 24.2806
- type: nauc_recall_at_5_max
value: 28.5408
- type: nauc_recall_at_5_std
value: 23.9741
- type: nauc_recall_at_5_diff1
value: 17.6308
- type: nauc_recall_at_10_max
value: 38.4262
- type: nauc_recall_at_10_std
value: 25.292399999999997
- type: nauc_recall_at_10_diff1
value: 17.5435
- type: nauc_recall_at_20_max
value: 26.0267
- type: nauc_recall_at_20_std
value: 17.8247
- type: nauc_recall_at_20_diff1
value: 14.788100000000002
- type: nauc_recall_at_100_max
value: 17.3545
- type: nauc_recall_at_100_std
value: 13.5356
- type: nauc_recall_at_100_diff1
value: 11.8308
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: -12.445599999999999
- type: nauc_precision_at_1_std
value: -44.4624
- type: nauc_precision_at_1_diff1
value: 29.339199999999998
- type: nauc_precision_at_3_max
value: 18.5164
- type: nauc_precision_at_3_std
value: 12.569700000000001
- type: nauc_precision_at_3_diff1
value: 24.2806
- type: nauc_precision_at_5_max
value: 28.5408
- type: nauc_precision_at_5_std
value: 23.9741
- type: nauc_precision_at_5_diff1
value: 17.6308
- type: nauc_precision_at_10_max
value: 38.4262
- type: nauc_precision_at_10_std
value: 25.292399999999997
- type: nauc_precision_at_10_diff1
value: 17.5435
- type: nauc_precision_at_20_max
value: 26.0267
- type: nauc_precision_at_20_std
value: 17.8247
- type: nauc_precision_at_20_diff1
value: 14.788100000000002
- type: nauc_precision_at_100_max
value: 17.3545
- type: nauc_precision_at_100_std
value: 13.5356
- type: nauc_precision_at_100_diff1
value: 11.8308
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: -12.445599999999999
- type: nauc_mrr_at_1_std
value: -44.4624
- type: nauc_mrr_at_1_diff1
value: 29.339199999999998
- type: nauc_mrr_at_3_max
value: 6.815200000000001
- type: nauc_mrr_at_3_std
value: -6.6357
- type: nauc_mrr_at_3_diff1
value: 24.1184
- type: nauc_mrr_at_5_max
value: 16.5725
- type: nauc_mrr_at_5_std
value: 6.4346
- type: nauc_mrr_at_5_diff1
value: 20.0389
- type: nauc_mrr_at_10_max
value: 21.2176
- type: nauc_mrr_at_10_std
value: 8.402
- type: nauc_mrr_at_10_diff1
value: 19.217000000000002
- type: nauc_mrr_at_20_max
value: 19.0886
- type: nauc_mrr_at_20_std
value: 7.749300000000001
- type: nauc_mrr_at_20_diff1
value: 18.1056
- type: nauc_mrr_at_100_max
value: 18.306
- type: nauc_mrr_at_100_std
value: 7.4771
- type: nauc_mrr_at_100_diff1
value: 17.4587
- type: nauc_mrr_at_1000_max
value: 18.3366
- type: nauc_mrr_at_1000_std
value: 7.4089
- type: nauc_mrr_at_1000_diff1
value: 17.5205
- type: main_score
value: 7.55
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-hin)
type: facebook/mlqa
config: ara-hin
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 1.6129999999999998
- type: ndcg_at_3
value: 2.899
- type: ndcg_at_5
value: 3.547
- type: ndcg_at_10
value: 4.782
- type: ndcg_at_20
value: 6.419999999999999
- type: ndcg_at_100
value: 15.101999999999999
- type: ndcg_at_1000
value: 20.041999999999998
- type: map_at_1
value: 1.6129999999999998
- type: map_at_3
value: 2.5989999999999998
- type: map_at_5
value: 2.948
- type: map_at_10
value: 3.4680000000000004
- type: map_at_20
value: 3.9210000000000003
- type: map_at_100
value: 4.914000000000001
- type: map_at_1000
value: 5.192
- type: recall_at_1
value: 1.6129999999999998
- type: recall_at_3
value: 3.763
- type: recall_at_5
value: 5.376
- type: recall_at_10
value: 9.139999999999999
- type: recall_at_20
value: 15.591
- type: recall_at_100
value: 65.591
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 1.6129999999999998
- type: precision_at_3
value: 1.254
- type: precision_at_5
value: 1.075
- type: precision_at_10
value: 0.914
- type: precision_at_20
value: 0.7799999999999999
- type: precision_at_100
value: 0.656
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 1.6129
- type: mrr_at_3
value: 2.5986
- type: mrr_at_5
value: 2.948
- type: mrr_at_10
value: 3.4675
- type: mrr_at_20
value: 3.9209
- type: mrr_at_100
value: 4.9135
- type: mrr_at_1000
value: 5.1921
- type: nauc_ndcg_at_1_max
value: 47.5085
- type: nauc_ndcg_at_1_std
value: 34.2866
- type: nauc_ndcg_at_1_diff1
value: 52.7582
- type: nauc_ndcg_at_3_max
value: 9.8372
- type: nauc_ndcg_at_3_std
value: 2.7338999999999998
- type: nauc_ndcg_at_3_diff1
value: 24.908
- type: nauc_ndcg_at_5_max
value: 11.766
- type: nauc_ndcg_at_5_std
value: -1.0312
- type: nauc_ndcg_at_5_diff1
value: 32.4895
- type: nauc_ndcg_at_10_max
value: 10.4204
- type: nauc_ndcg_at_10_std
value: 0.47479999999999994
- type: nauc_ndcg_at_10_diff1
value: 27.427
- type: nauc_ndcg_at_20_max
value: 6.3569
- type: nauc_ndcg_at_20_std
value: -0.7947
- type: nauc_ndcg_at_20_diff1
value: 16.6717
- type: nauc_ndcg_at_100_max
value: 12.878200000000001
- type: nauc_ndcg_at_100_std
value: 8.6943
- type: nauc_ndcg_at_100_diff1
value: 15.512300000000002
- type: nauc_ndcg_at_1000_max
value: 11.164399999999999
- type: nauc_ndcg_at_1000_std
value: 3.8767000000000005
- type: nauc_ndcg_at_1000_diff1
value: 21.2167
- type: nauc_map_at_1_max
value: 47.5085
- type: nauc_map_at_1_std
value: 34.2866
- type: nauc_map_at_1_diff1
value: 52.7582
- type: nauc_map_at_3_max
value: 14.6876
- type: nauc_map_at_3_std
value: 6.7038
- type: nauc_map_at_3_diff1
value: 29.472900000000003
- type: nauc_map_at_5_max
value: 15.762
- type: nauc_map_at_5_std
value: 4.04
- type: nauc_map_at_5_diff1
value: 33.8561
- type: nauc_map_at_10_max
value: 14.46
- type: nauc_map_at_10_std
value: 4.1512
- type: nauc_map_at_10_diff1
value: 31.0161
- type: nauc_map_at_20_max
value: 12.2367
- type: nauc_map_at_20_std
value: 3.2522
- type: nauc_map_at_20_diff1
value: 26.2027
- type: nauc_map_at_100_max
value: 13.314699999999998
- type: nauc_map_at_100_std
value: 5.0341
- type: nauc_map_at_100_diff1
value: 25.3857
- type: nauc_map_at_1000_max
value: 13.237599999999999
- type: nauc_map_at_1000_std
value: 4.620699999999999
- type: nauc_map_at_1000_diff1
value: 26.075300000000002
- type: nauc_recall_at_1_max
value: 47.5085
- type: nauc_recall_at_1_std
value: 34.2866
- type: nauc_recall_at_1_diff1
value: 52.7582
- type: nauc_recall_at_3_max
value: 0.39709999999999995
- type: nauc_recall_at_3_std
value: -4.9616
- type: nauc_recall_at_3_diff1
value: 15.699
- type: nauc_recall_at_5_max
value: 5.497
- type: nauc_recall_at_5_std
value: -9.4116
- type: nauc_recall_at_5_diff1
value: 30.917099999999998
- type: nauc_recall_at_10_max
value: 5.7965
- type: nauc_recall_at_10_std
value: -3.5463
- type: nauc_recall_at_10_diff1
value: 22.8954
- type: nauc_recall_at_20_max
value: 0.6188
- type: nauc_recall_at_20_std
value: -4.326
- type: nauc_recall_at_20_diff1
value: 5.7056000000000004
- type: nauc_recall_at_100_max
value: 16.1744
- type: nauc_recall_at_100_std
value: 17.721700000000002
- type: nauc_recall_at_100_diff1
value: 4.917400000000001
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 47.5085
- type: nauc_precision_at_1_std
value: 34.2866
- type: nauc_precision_at_1_diff1
value: 52.7582
- type: nauc_precision_at_3_max
value: 0.39709999999999995
- type: nauc_precision_at_3_std
value: -4.9616
- type: nauc_precision_at_3_diff1
value: 15.699
- type: nauc_precision_at_5_max
value: 5.497
- type: nauc_precision_at_5_std
value: -9.4116
- type: nauc_precision_at_5_diff1
value: 30.917099999999998
- type: nauc_precision_at_10_max
value: 5.7965
- type: nauc_precision_at_10_std
value: -3.5463
- type: nauc_precision_at_10_diff1
value: 22.8954
- type: nauc_precision_at_20_max
value: 0.6188
- type: nauc_precision_at_20_std
value: -4.326
- type: nauc_precision_at_20_diff1
value: 5.7056000000000004
- type: nauc_precision_at_100_max
value: 16.1744
- type: nauc_precision_at_100_std
value: 17.721700000000002
- type: nauc_precision_at_100_diff1
value: 4.917400000000001
- type: nauc_precision_at_1000_max
value: .nan
- type: nauc_precision_at_1000_std
value: .nan
- type: nauc_precision_at_1000_diff1
value: .nan
- type: nauc_mrr_at_1_max
value: 47.5085
- type: nauc_mrr_at_1_std
value: 34.2866
- type: nauc_mrr_at_1_diff1
value: 52.7582
- type: nauc_mrr_at_3_max
value: 14.6876
- type: nauc_mrr_at_3_std
value: 6.7038
- type: nauc_mrr_at_3_diff1
value: 29.472900000000003
- type: nauc_mrr_at_5_max
value: 15.762
- type: nauc_mrr_at_5_std
value: 4.04
- type: nauc_mrr_at_5_diff1
value: 33.8561
- type: nauc_mrr_at_10_max
value: 14.46
- type: nauc_mrr_at_10_std
value: 4.1512
- type: nauc_mrr_at_10_diff1
value: 31.0161
- type: nauc_mrr_at_20_max
value: 12.2367
- type: nauc_mrr_at_20_std
value: 3.2522
- type: nauc_mrr_at_20_diff1
value: 26.2027
- type: nauc_mrr_at_100_max
value: 13.314699999999998
- type: nauc_mrr_at_100_std
value: 5.0341
- type: nauc_mrr_at_100_diff1
value: 25.3857
- type: nauc_mrr_at_1000_max
value: 13.237599999999999
- type: nauc_mrr_at_1000_std
value: 4.620699999999999
- type: nauc_mrr_at_1000_diff1
value: 26.075300000000002
- type: main_score
value: 4.782
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-vie)
type: facebook/mlqa
config: ara-vie
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 1.8399999999999999
- type: ndcg_at_3
value: 6.084
- type: ndcg_at_5
value: 7.88
- type: ndcg_at_10
value: 10.208
- type: ndcg_at_20
value: 12.341000000000001
- type: ndcg_at_100
value: 21.467
- type: ndcg_at_1000
value: 24.204
- type: map_at_1
value: 1.8399999999999999
- type: map_at_3
value: 5.01
- type: map_at_5
value: 6.022
- type: map_at_10
value: 6.952999999999999
- type: map_at_20
value: 7.519000000000001
- type: map_at_100
value: 8.627
- type: map_at_1000
value: 8.783000000000001
- type: recall_at_1
value: 1.8399999999999999
- type: recall_at_3
value: 9.202
- type: recall_at_5
value: 13.497
- type: recall_at_10
value: 20.858999999999998
- type: recall_at_20
value: 29.448
- type: recall_at_100
value: 80.982
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 1.8399999999999999
- type: precision_at_3
value: 3.0669999999999997
- type: precision_at_5
value: 2.699
- type: precision_at_10
value: 2.086
- type: precision_at_20
value: 1.472
- type: precision_at_100
value: 0.8099999999999999
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 1.8405
- type: mrr_at_3
value: 5.0102
- type: mrr_at_5
value: 6.0225
- type: mrr_at_10
value: 6.9527
- type: mrr_at_20
value: 7.519099999999999
- type: mrr_at_100
value: 8.6274
- type: mrr_at_1000
value: 8.783299999999999
- type: nauc_ndcg_at_1_max
value: 52.876999999999995
- type: nauc_ndcg_at_1_std
value: 18.8889
- type: nauc_ndcg_at_1_diff1
value: 52.876999999999995
- type: nauc_ndcg_at_3_max
value: 38.5665
- type: nauc_ndcg_at_3_std
value: 22.0193
- type: nauc_ndcg_at_3_diff1
value: 41.2907
- type: nauc_ndcg_at_5_max
value: 44.3423
- type: nauc_ndcg_at_5_std
value: 19.5666
- type: nauc_ndcg_at_5_diff1
value: 49.2458
- type: nauc_ndcg_at_10_max
value: 34.1614
- type: nauc_ndcg_at_10_std
value: 12.8171
- type: nauc_ndcg_at_10_diff1
value: 42.0935
- type: nauc_ndcg_at_20_max
value: 31.5043
- type: nauc_ndcg_at_20_std
value: 21.6028
- type: nauc_ndcg_at_20_diff1
value: 37.4641
- type: nauc_ndcg_at_100_max
value: 32.8116
- type: nauc_ndcg_at_100_std
value: 21.9274
- type: nauc_ndcg_at_100_diff1
value: 32.9501
- type: nauc_ndcg_at_1000_max
value: 33.9661
- type: nauc_ndcg_at_1000_std
value: 20.170199999999998
- type: nauc_ndcg_at_1000_diff1
value: 38.0503
- type: nauc_map_at_1_max
value: 52.876999999999995
- type: nauc_map_at_1_std
value: 18.8889
- type: nauc_map_at_1_diff1
value: 52.876999999999995
- type: nauc_map_at_3_max
value: 40.726600000000005
- type: nauc_map_at_3_std
value: 22.6993
- type: nauc_map_at_3_diff1
value: 42.1939
- type: nauc_map_at_5_max
value: 45.0313
- type: nauc_map_at_5_std
value: 21.144099999999998
- type: nauc_map_at_5_diff1
value: 48.0884
- type: nauc_map_at_10_max
value: 38.9346
- type: nauc_map_at_10_std
value: 17.3547
- type: nauc_map_at_10_diff1
value: 43.9371
- type: nauc_map_at_20_max
value: 37.8438
- type: nauc_map_at_20_std
value: 20.8716
- type: nauc_map_at_20_diff1
value: 41.9294
- type: nauc_map_at_100_max
value: 37.419999999999995
- type: nauc_map_at_100_std
value: 20.6405
- type: nauc_map_at_100_diff1
value: 40.8201
- type: nauc_map_at_1000_max
value: 37.5517
- type: nauc_map_at_1000_std
value: 20.515
- type: nauc_map_at_1000_diff1
value: 41.2154
- type: nauc_recall_at_1_max
value: 52.876999999999995
- type: nauc_recall_at_1_std
value: 18.8889
- type: nauc_recall_at_1_diff1
value: 52.876999999999995
- type: nauc_recall_at_3_max
value: 34.9721
- type: nauc_recall_at_3_std
value: 20.7357
- type: nauc_recall_at_3_diff1
value: 39.8992
- type: nauc_recall_at_5_max
value: 43.399100000000004
- type: nauc_recall_at_5_std
value: 16.9361
- type: nauc_recall_at_5_diff1
value: 51.194799999999994
- type: nauc_recall_at_10_max
value: 27.520699999999998
- type: nauc_recall_at_10_std
value: 6.251900000000001
- type: nauc_recall_at_10_diff1
value: 39.3665
- type: nauc_recall_at_20_max
value: 23.0855
- type: nauc_recall_at_20_std
value: 23.717299999999998
- type: nauc_recall_at_20_diff1
value: 31.1618
- type: nauc_recall_at_100_max
value: 27.691100000000002
- type: nauc_recall_at_100_std
value: 29.7084
- type: nauc_recall_at_100_diff1
value: 9.9303
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 52.876999999999995
- type: nauc_precision_at_1_std
value: 18.8889
- type: nauc_precision_at_1_diff1
value: 52.876999999999995
- type: nauc_precision_at_3_max
value: 34.9721
- type: nauc_precision_at_3_std
value: 20.7357
- type: nauc_precision_at_3_diff1
value: 39.8992
- type: nauc_precision_at_5_max
value: 43.399100000000004
- type: nauc_precision_at_5_std
value: 16.9361
- type: nauc_precision_at_5_diff1
value: 51.194799999999994
- type: nauc_precision_at_10_max
value: 27.520699999999998
- type: nauc_precision_at_10_std
value: 6.251900000000001
- type: nauc_precision_at_10_diff1
value: 39.3665
- type: nauc_precision_at_20_max
value: 23.0855
- type: nauc_precision_at_20_std
value: 23.717299999999998
- type: nauc_precision_at_20_diff1
value: 31.1618
- type: nauc_precision_at_100_max
value: 27.691100000000002
- type: nauc_precision_at_100_std
value: 29.7084
- type: nauc_precision_at_100_diff1
value: 9.9303
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 52.876999999999995
- type: nauc_mrr_at_1_std
value: 18.8889
- type: nauc_mrr_at_1_diff1
value: 52.876999999999995
- type: nauc_mrr_at_3_max
value: 40.726600000000005
- type: nauc_mrr_at_3_std
value: 22.6993
- type: nauc_mrr_at_3_diff1
value: 42.1939
- type: nauc_mrr_at_5_max
value: 45.0313
- type: nauc_mrr_at_5_std
value: 21.144099999999998
- type: nauc_mrr_at_5_diff1
value: 48.0884
- type: nauc_mrr_at_10_max
value: 38.9346
- type: nauc_mrr_at_10_std
value: 17.3547
- type: nauc_mrr_at_10_diff1
value: 43.9371
- type: nauc_mrr_at_20_max
value: 37.8438
- type: nauc_mrr_at_20_std
value: 20.8716
- type: nauc_mrr_at_20_diff1
value: 41.9294
- type: nauc_mrr_at_100_max
value: 37.419999999999995
- type: nauc_mrr_at_100_std
value: 20.6405
- type: nauc_mrr_at_100_diff1
value: 40.8201
- type: nauc_mrr_at_1000_max
value: 37.5517
- type: nauc_mrr_at_1000_std
value: 20.515
- type: nauc_mrr_at_1000_diff1
value: 41.2154
- type: main_score
value: 10.208
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-zho)
type: facebook/mlqa
config: ara-zho
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 1.5959999999999999
- type: ndcg_at_3
value: 2.7289999999999996
- type: ndcg_at_5
value: 2.935
- type: ndcg_at_10
value: 4.668
- type: ndcg_at_20
value: 6.487
- type: ndcg_at_100
value: 15.845999999999998
- type: ndcg_at_1000
value: 19.963
- type: map_at_1
value: 1.5959999999999999
- type: map_at_3
value: 2.394
- type: map_at_5
value: 2.5
- type: map_at_10
value: 3.222
- type: map_at_20
value: 3.688
- type: map_at_100
value: 4.731
- type: map_at_1000
value: 4.962
- type: recall_at_1
value: 1.5959999999999999
- type: recall_at_3
value: 3.723
- type: recall_at_5
value: 4.255
- type: recall_at_10
value: 9.574
- type: recall_at_20
value: 17.021
- type: recall_at_100
value: 71.277
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 1.5959999999999999
- type: precision_at_3
value: 1.2409999999999999
- type: precision_at_5
value: 0.851
- type: precision_at_10
value: 0.9570000000000001
- type: precision_at_20
value: 0.851
- type: precision_at_100
value: 0.713
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 1.5957
- type: mrr_at_3
value: 2.3935999999999997
- type: mrr_at_5
value: 2.5
- type: mrr_at_10
value: 3.2223
- type: mrr_at_20
value: 3.6881999999999997
- type: mrr_at_100
value: 4.7308
- type: mrr_at_1000
value: 4.9618
- type: nauc_ndcg_at_1_max
value: 77.5817
- type: nauc_ndcg_at_1_std
value: 77.5817
- type: nauc_ndcg_at_1_diff1
value: 88.7908
- type: nauc_ndcg_at_3_max
value: 44.5384
- type: nauc_ndcg_at_3_std
value: 43.708200000000005
- type: nauc_ndcg_at_3_diff1
value: 43.5215
- type: nauc_ndcg_at_5_max
value: 46.0692
- type: nauc_ndcg_at_5_std
value: 42.9396
- type: nauc_ndcg_at_5_diff1
value: 41.166199999999996
- type: nauc_ndcg_at_10_max
value: 30.946800000000003
- type: nauc_ndcg_at_10_std
value: 32.2119
- type: nauc_ndcg_at_10_diff1
value: 30.8354
- type: nauc_ndcg_at_20_max
value: 21.0281
- type: nauc_ndcg_at_20_std
value: 22.289
- type: nauc_ndcg_at_20_diff1
value: 31.3122
- type: nauc_ndcg_at_100_max
value: 17.1413
- type: nauc_ndcg_at_100_std
value: 15.3116
- type: nauc_ndcg_at_100_diff1
value: 17.156299999999998
- type: nauc_ndcg_at_1000_max
value: 24.814700000000002
- type: nauc_ndcg_at_1000_std
value: 24.8968
- type: nauc_ndcg_at_1000_diff1
value: 28.456300000000002
- type: nauc_map_at_1_max
value: 77.5817
- type: nauc_map_at_1_std
value: 77.5817
- type: nauc_map_at_1_diff1
value: 88.7908
- type: nauc_map_at_3_max
value: 50.9702
- type: nauc_map_at_3_std
value: 50.3392
- type: nauc_map_at_3_diff1
value: 52.2489
- type: nauc_map_at_5_max
value: 51.625600000000006
- type: nauc_map_at_5_std
value: 49.5905
- type: nauc_map_at_5_diff1
value: 50.44800000000001
- type: nauc_map_at_10_max
value: 41.103
- type: nauc_map_at_10_std
value: 41.624100000000006
- type: nauc_map_at_10_diff1
value: 41.6516
- type: nauc_map_at_20_max
value: 35.8476
- type: nauc_map_at_20_std
value: 36.3296
- type: nauc_map_at_20_diff1
value: 40.9989
- type: nauc_map_at_100_max
value: 33.3228
- type: nauc_map_at_100_std
value: 33.2988
- type: nauc_map_at_100_diff1
value: 36.5126
- type: nauc_map_at_1000_max
value: 34.405
- type: nauc_map_at_1000_std
value: 34.5349
- type: nauc_map_at_1000_diff1
value: 37.889
- type: nauc_recall_at_1_max
value: 77.5817
- type: nauc_recall_at_1_std
value: 77.5817
- type: nauc_recall_at_1_diff1
value: 88.7908
- type: nauc_recall_at_3_max
value: 32.3091
- type: nauc_recall_at_3_std
value: 31.092100000000002
- type: nauc_recall_at_3_diff1
value: 26.9461
- type: nauc_recall_at_5_max
value: 36.567
- type: nauc_recall_at_5_std
value: 31.2987
- type: nauc_recall_at_5_diff1
value: 24.8186
- type: nauc_recall_at_10_max
value: 19.4747
- type: nauc_recall_at_10_std
value: 21.7032
- type: nauc_recall_at_10_diff1
value: 19.313299999999998
- type: nauc_recall_at_20_max
value: 7.2557
- type: nauc_recall_at_20_std
value: 9.3428
- type: nauc_recall_at_20_diff1
value: 23.842
- type: nauc_recall_at_100_max
value: 2.5262
- type: nauc_recall_at_100_std
value: -3.295
- type: nauc_recall_at_100_diff1
value: -4.9431
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 77.5817
- type: nauc_precision_at_1_std
value: 77.5817
- type: nauc_precision_at_1_diff1
value: 88.7908
- type: nauc_precision_at_3_max
value: 32.3091
- type: nauc_precision_at_3_std
value: 31.092100000000002
- type: nauc_precision_at_3_diff1
value: 26.9461
- type: nauc_precision_at_5_max
value: 36.567
- type: nauc_precision_at_5_std
value: 31.2987
- type: nauc_precision_at_5_diff1
value: 24.8186
- type: nauc_precision_at_10_max
value: 19.4747
- type: nauc_precision_at_10_std
value: 21.7032
- type: nauc_precision_at_10_diff1
value: 19.313299999999998
- type: nauc_precision_at_20_max
value: 7.2557
- type: nauc_precision_at_20_std
value: 9.3428
- type: nauc_precision_at_20_diff1
value: 23.842
- type: nauc_precision_at_100_max
value: 2.5262
- type: nauc_precision_at_100_std
value: -3.295
- type: nauc_precision_at_100_diff1
value: -4.9431
- type: nauc_precision_at_1000_max
value: .nan
- type: nauc_precision_at_1000_std
value: .nan
- type: nauc_precision_at_1000_diff1
value: .nan
- type: nauc_mrr_at_1_max
value: 77.5817
- type: nauc_mrr_at_1_std
value: 77.5817
- type: nauc_mrr_at_1_diff1
value: 88.7908
- type: nauc_mrr_at_3_max
value: 50.9702
- type: nauc_mrr_at_3_std
value: 50.3392
- type: nauc_mrr_at_3_diff1
value: 52.2489
- type: nauc_mrr_at_5_max
value: 51.625600000000006
- type: nauc_mrr_at_5_std
value: 49.5905
- type: nauc_mrr_at_5_diff1
value: 50.44800000000001
- type: nauc_mrr_at_10_max
value: 41.103
- type: nauc_mrr_at_10_std
value: 41.624100000000006
- type: nauc_mrr_at_10_diff1
value: 41.6516
- type: nauc_mrr_at_20_max
value: 35.8476
- type: nauc_mrr_at_20_std
value: 36.3296
- type: nauc_mrr_at_20_diff1
value: 40.9989
- type: nauc_mrr_at_100_max
value: 33.3228
- type: nauc_mrr_at_100_std
value: 33.2988
- type: nauc_mrr_at_100_diff1
value: 36.5126
- type: nauc_mrr_at_1000_max
value: 34.405
- type: nauc_mrr_at_1000_std
value: 34.5349
- type: nauc_mrr_at_1000_diff1
value: 37.889
- type: main_score
value: 4.668
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (deu-ara)
type: facebook/mlqa
config: deu-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 9.661999999999999
- type: ndcg_at_3
value: 13.434
- type: ndcg_at_5
value: 15.18
- type: ndcg_at_10
value: 19.24
- type: ndcg_at_20
value: 21.028
- type: ndcg_at_100
value: 28.998
- type: ndcg_at_1000
value: 31.197000000000003
- type: map_at_1
value: 9.661999999999999
- type: map_at_3
value: 12.559999999999999
- type: map_at_5
value: 13.502
- type: map_at_10
value: 15.179
- type: map_at_20
value: 15.645999999999999
- type: map_at_100
value: 16.639
- type: map_at_1000
value: 16.759
- type: recall_at_1
value: 9.661999999999999
- type: recall_at_3
value: 15.942
- type: recall_at_5
value: 20.29
- type: recall_at_10
value: 32.85
- type: recall_at_20
value: 40.097
- type: recall_at_100
value: 84.541
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 9.661999999999999
- type: precision_at_3
value: 5.314
- type: precision_at_5
value: 4.058
- type: precision_at_10
value: 3.2849999999999997
- type: precision_at_20
value: 2.005
- type: precision_at_100
value: 0.845
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 9.6618
- type: mrr_at_3
value: 12.5604
- type: mrr_at_5
value: 13.5024
- type: mrr_at_10
value: 15.178700000000001
- type: mrr_at_20
value: 15.646099999999999
- type: mrr_at_100
value: 16.639300000000002
- type: mrr_at_1000
value: 16.7593
- type: nauc_ndcg_at_1_max
value: 46.9036
- type: nauc_ndcg_at_1_std
value: 47.3
- type: nauc_ndcg_at_1_diff1
value: 37.804300000000005
- type: nauc_ndcg_at_3_max
value: 42.582
- type: nauc_ndcg_at_3_std
value: 42.4601
- type: nauc_ndcg_at_3_diff1
value: 32.8016
- type: nauc_ndcg_at_5_max
value: 39.785199999999996
- type: nauc_ndcg_at_5_std
value: 43.6797
- type: nauc_ndcg_at_5_diff1
value: 31.4959
- type: nauc_ndcg_at_10_max
value: 39.833400000000005
- type: nauc_ndcg_at_10_std
value: 43.2245
- type: nauc_ndcg_at_10_diff1
value: 29.857699999999998
- type: nauc_ndcg_at_20_max
value: 39.4031
- type: nauc_ndcg_at_20_std
value: 42.9703
- type: nauc_ndcg_at_20_diff1
value: 29.1932
- type: nauc_ndcg_at_100_max
value: 39.5612
- type: nauc_ndcg_at_100_std
value: 43.803399999999996
- type: nauc_ndcg_at_100_diff1
value: 27.535500000000003
- type: nauc_ndcg_at_1000_max
value: 40.466
- type: nauc_ndcg_at_1000_std
value: 44.0194
- type: nauc_ndcg_at_1000_diff1
value: 30.501299999999997
- type: nauc_map_at_1_max
value: 46.9036
- type: nauc_map_at_1_std
value: 47.3
- type: nauc_map_at_1_diff1
value: 37.804300000000005
- type: nauc_map_at_3_max
value: 43.6776
- type: nauc_map_at_3_std
value: 43.648399999999995
- type: nauc_map_at_3_diff1
value: 34.0512
- type: nauc_map_at_5_max
value: 41.994
- type: nauc_map_at_5_std
value: 44.2756
- type: nauc_map_at_5_diff1
value: 33.1186
- type: nauc_map_at_10_max
value: 41.8409
- type: nauc_map_at_10_std
value: 44.0738
- type: nauc_map_at_10_diff1
value: 32.2567
- type: nauc_map_at_20_max
value: 41.7295
- type: nauc_map_at_20_std
value: 44.0689
- type: nauc_map_at_20_diff1
value: 32.096599999999995
- type: nauc_map_at_100_max
value: 41.7376
- type: nauc_map_at_100_std
value: 44.2902
- type: nauc_map_at_100_diff1
value: 32.0627
- type: nauc_map_at_1000_max
value: 41.781800000000004
- type: nauc_map_at_1000_std
value: 44.308
- type: nauc_map_at_1000_diff1
value: 32.2156
- type: nauc_recall_at_1_max
value: 46.9036
- type: nauc_recall_at_1_std
value: 47.3
- type: nauc_recall_at_1_diff1
value: 37.804300000000005
- type: nauc_recall_at_3_max
value: 39.866800000000005
- type: nauc_recall_at_3_std
value: 39.5259
- type: nauc_recall_at_3_diff1
value: 29.7101
- type: nauc_recall_at_5_max
value: 34.6971
- type: nauc_recall_at_5_std
value: 42.5317
- type: nauc_recall_at_5_diff1
value: 27.9304
- type: nauc_recall_at_10_max
value: 35.9878
- type: nauc_recall_at_10_std
value: 41.5877
- type: nauc_recall_at_10_diff1
value: 25.0104
- type: nauc_recall_at_20_max
value: 34.7729
- type: nauc_recall_at_20_std
value: 40.5754
- type: nauc_recall_at_20_diff1
value: 23.058799999999998
- type: nauc_recall_at_100_max
value: 30.4483
- type: nauc_recall_at_100_std
value: 41.924099999999996
- type: nauc_recall_at_100_diff1
value: -1.2919999999999998
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 46.9036
- type: nauc_precision_at_1_std
value: 47.3
- type: nauc_precision_at_1_diff1
value: 37.804300000000005
- type: nauc_precision_at_3_max
value: 39.866800000000005
- type: nauc_precision_at_3_std
value: 39.5259
- type: nauc_precision_at_3_diff1
value: 29.7101
- type: nauc_precision_at_5_max
value: 34.6971
- type: nauc_precision_at_5_std
value: 42.5317
- type: nauc_precision_at_5_diff1
value: 27.9304
- type: nauc_precision_at_10_max
value: 35.9878
- type: nauc_precision_at_10_std
value: 41.5877
- type: nauc_precision_at_10_diff1
value: 25.0104
- type: nauc_precision_at_20_max
value: 34.7729
- type: nauc_precision_at_20_std
value: 40.5754
- type: nauc_precision_at_20_diff1
value: 23.058799999999998
- type: nauc_precision_at_100_max
value: 30.4483
- type: nauc_precision_at_100_std
value: 41.924099999999996
- type: nauc_precision_at_100_diff1
value: -1.2919999999999998
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 46.9036
- type: nauc_mrr_at_1_std
value: 47.3
- type: nauc_mrr_at_1_diff1
value: 37.804300000000005
- type: nauc_mrr_at_3_max
value: 43.6776
- type: nauc_mrr_at_3_std
value: 43.648399999999995
- type: nauc_mrr_at_3_diff1
value: 34.0512
- type: nauc_mrr_at_5_max
value: 41.994
- type: nauc_mrr_at_5_std
value: 44.2756
- type: nauc_mrr_at_5_diff1
value: 33.1186
- type: nauc_mrr_at_10_max
value: 41.8409
- type: nauc_mrr_at_10_std
value: 44.0738
- type: nauc_mrr_at_10_diff1
value: 32.2567
- type: nauc_mrr_at_20_max
value: 41.7295
- type: nauc_mrr_at_20_std
value: 44.0689
- type: nauc_mrr_at_20_diff1
value: 32.096599999999995
- type: nauc_mrr_at_100_max
value: 41.7376
- type: nauc_mrr_at_100_std
value: 44.2902
- type: nauc_mrr_at_100_diff1
value: 32.0627
- type: nauc_mrr_at_1000_max
value: 41.781800000000004
- type: nauc_mrr_at_1000_std
value: 44.308
- type: nauc_mrr_at_1000_diff1
value: 32.2156
- type: main_score
value: 19.24
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (eng-ara)
type: facebook/mlqa
config: eng-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 13.733
- type: ndcg_at_3
value: 20.279
- type: ndcg_at_5
value: 23.384
- type: ndcg_at_10
value: 27.189000000000004
- type: ndcg_at_20
value: 30.29
- type: ndcg_at_100
value: 35.32
- type: ndcg_at_1000
value: 37.425000000000004
- type: map_at_1
value: 13.733
- type: map_at_3
value: 18.665000000000003
- type: map_at_5
value: 20.387
- type: map_at_10
value: 21.951
- type: map_at_20
value: 22.787
- type: map_at_100
value: 23.473
- type: map_at_1000
value: 23.558
- type: recall_at_1
value: 13.733
- type: recall_at_3
value: 24.951999999999998
- type: recall_at_5
value: 32.495000000000005
- type: recall_at_10
value: 44.294
- type: recall_at_20
value: 56.672999999999995
- type: recall_at_100
value: 83.946
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 13.733
- type: precision_at_3
value: 8.317
- type: precision_at_5
value: 6.4990000000000006
- type: precision_at_10
value: 4.429
- type: precision_at_20
value: 2.834
- type: precision_at_100
value: 0.839
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 13.7331
- type: mrr_at_3
value: 18.665399999999998
- type: mrr_at_5
value: 20.3868
- type: mrr_at_10
value: 21.9511
- type: mrr_at_20
value: 22.7873
- type: mrr_at_100
value: 23.4728
- type: mrr_at_1000
value: 23.5579
- type: nauc_ndcg_at_1_max
value: 31.3549
- type: nauc_ndcg_at_1_std
value: 22.8524
- type: nauc_ndcg_at_1_diff1
value: 37.5512
- type: nauc_ndcg_at_3_max
value: 30.3012
- type: nauc_ndcg_at_3_std
value: 21.8318
- type: nauc_ndcg_at_3_diff1
value: 30.4344
- type: nauc_ndcg_at_5_max
value: 26.604499999999998
- type: nauc_ndcg_at_5_std
value: 20.627599999999997
- type: nauc_ndcg_at_5_diff1
value: 27.6343
- type: nauc_ndcg_at_10_max
value: 27.330700000000004
- type: nauc_ndcg_at_10_std
value: 20.8627
- type: nauc_ndcg_at_10_diff1
value: 25.8142
- type: nauc_ndcg_at_20_max
value: 29.027399999999997
- type: nauc_ndcg_at_20_std
value: 21.307100000000002
- type: nauc_ndcg_at_20_diff1
value: 26.6961
- type: nauc_ndcg_at_100_max
value: 29.074499999999997
- type: nauc_ndcg_at_100_std
value: 23.1857
- type: nauc_ndcg_at_100_diff1
value: 26.266099999999998
- type: nauc_ndcg_at_1000_max
value: 28.8016
- type: nauc_ndcg_at_1000_std
value: 21.7539
- type: nauc_ndcg_at_1000_diff1
value: 27.777
- type: nauc_map_at_1_max
value: 31.3549
- type: nauc_map_at_1_std
value: 22.8524
- type: nauc_map_at_1_diff1
value: 37.5512
- type: nauc_map_at_3_max
value: 30.5276
- type: nauc_map_at_3_std
value: 22.0186
- type: nauc_map_at_3_diff1
value: 31.6059
- type: nauc_map_at_5_max
value: 28.3572
- type: nauc_map_at_5_std
value: 21.341099999999997
- type: nauc_map_at_5_diff1
value: 29.9248
- type: nauc_map_at_10_max
value: 28.601100000000002
- type: nauc_map_at_10_std
value: 21.3735
- type: nauc_map_at_10_diff1
value: 29.108800000000002
- type: nauc_map_at_20_max
value: 29.0503
- type: nauc_map_at_20_std
value: 21.4425
- type: nauc_map_at_20_diff1
value: 29.3655
- type: nauc_map_at_100_max
value: 29.0648
- type: nauc_map_at_100_std
value: 21.6384
- type: nauc_map_at_100_diff1
value: 29.315799999999996
- type: nauc_map_at_1000_max
value: 29.0516
- type: nauc_map_at_1000_std
value: 21.5804
- type: nauc_map_at_1000_diff1
value: 29.391000000000002
- type: nauc_recall_at_1_max
value: 31.3549
- type: nauc_recall_at_1_std
value: 22.8524
- type: nauc_recall_at_1_diff1
value: 37.5512
- type: nauc_recall_at_3_max
value: 29.7528
- type: nauc_recall_at_3_std
value: 21.3895
- type: nauc_recall_at_3_diff1
value: 27.7102
- type: nauc_recall_at_5_max
value: 22.2167
- type: nauc_recall_at_5_std
value: 18.8542
- type: nauc_recall_at_5_diff1
value: 22.245
- type: nauc_recall_at_10_max
value: 24.4284
- type: nauc_recall_at_10_std
value: 19.764300000000002
- type: nauc_recall_at_10_diff1
value: 17.7194
- type: nauc_recall_at_20_max
value: 30.353599999999997
- type: nauc_recall_at_20_std
value: 21.593799999999998
- type: nauc_recall_at_20_diff1
value: 20.138
- type: nauc_recall_at_100_max
value: 32.022
- type: nauc_recall_at_100_std
value: 39.9011
- type: nauc_recall_at_100_diff1
value: 9.5189
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 31.3549
- type: nauc_precision_at_1_std
value: 22.8524
- type: nauc_precision_at_1_diff1
value: 37.5512
- type: nauc_precision_at_3_max
value: 29.7528
- type: nauc_precision_at_3_std
value: 21.3895
- type: nauc_precision_at_3_diff1
value: 27.7102
- type: nauc_precision_at_5_max
value: 22.2167
- type: nauc_precision_at_5_std
value: 18.8542
- type: nauc_precision_at_5_diff1
value: 22.245
- type: nauc_precision_at_10_max
value: 24.4284
- type: nauc_precision_at_10_std
value: 19.764300000000002
- type: nauc_precision_at_10_diff1
value: 17.7194
- type: nauc_precision_at_20_max
value: 30.353599999999997
- type: nauc_precision_at_20_std
value: 21.593799999999998
- type: nauc_precision_at_20_diff1
value: 20.138
- type: nauc_precision_at_100_max
value: 32.022
- type: nauc_precision_at_100_std
value: 39.9011
- type: nauc_precision_at_100_diff1
value: 9.5189
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 31.3549
- type: nauc_mrr_at_1_std
value: 22.8524
- type: nauc_mrr_at_1_diff1
value: 37.5512
- type: nauc_mrr_at_3_max
value: 30.5276
- type: nauc_mrr_at_3_std
value: 22.0186
- type: nauc_mrr_at_3_diff1
value: 31.6059
- type: nauc_mrr_at_5_max
value: 28.3572
- type: nauc_mrr_at_5_std
value: 21.341099999999997
- type: nauc_mrr_at_5_diff1
value: 29.9248
- type: nauc_mrr_at_10_max
value: 28.601100000000002
- type: nauc_mrr_at_10_std
value: 21.3735
- type: nauc_mrr_at_10_diff1
value: 29.108800000000002
- type: nauc_mrr_at_20_max
value: 29.0503
- type: nauc_mrr_at_20_std
value: 21.4425
- type: nauc_mrr_at_20_diff1
value: 29.3655
- type: nauc_mrr_at_100_max
value: 29.0648
- type: nauc_mrr_at_100_std
value: 21.6384
- type: nauc_mrr_at_100_diff1
value: 29.315799999999996
- type: nauc_mrr_at_1000_max
value: 29.0516
- type: nauc_mrr_at_1000_std
value: 21.5804
- type: nauc_mrr_at_1000_diff1
value: 29.391000000000002
- type: main_score
value: 27.189000000000004
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (spa-ara)
type: facebook/mlqa
config: spa-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 10.559000000000001
- type: ndcg_at_3
value: 14.071
- type: ndcg_at_5
value: 16.878
- type: ndcg_at_10
value: 18.429000000000002
- type: ndcg_at_20
value: 21.648
- type: ndcg_at_100
value: 29.946
- type: ndcg_at_1000
value: 31.746999999999996
- type: map_at_1
value: 10.559000000000001
- type: map_at_3
value: 13.147
- type: map_at_5
value: 14.7
- type: map_at_10
value: 15.308
- type: map_at_20
value: 16.23
- type: map_at_100
value: 17.25
- type: map_at_1000
value: 17.355
- type: recall_at_1
value: 10.559000000000001
- type: recall_at_3
value: 16.77
- type: recall_at_5
value: 23.602
- type: recall_at_10
value: 28.571
- type: recall_at_20
value: 40.994
- type: recall_at_100
value: 87.578
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 10.559000000000001
- type: precision_at_3
value: 5.59
- type: precision_at_5
value: 4.72
- type: precision_at_10
value: 2.857
- type: precision_at_20
value: 2.0500000000000003
- type: precision_at_100
value: 0.876
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 10.559000000000001
- type: mrr_at_3
value: 13.147
- type: mrr_at_5
value: 14.6998
- type: mrr_at_10
value: 15.307799999999999
- type: mrr_at_20
value: 16.23
- type: mrr_at_100
value: 17.2501
- type: mrr_at_1000
value: 17.3553
- type: nauc_ndcg_at_1_max
value: 21.683
- type: nauc_ndcg_at_1_std
value: 23.9115
- type: nauc_ndcg_at_1_diff1
value: 34.306799999999996
- type: nauc_ndcg_at_3_max
value: 10.9801
- type: nauc_ndcg_at_3_std
value: 17.8432
- type: nauc_ndcg_at_3_diff1
value: 24.3422
- type: nauc_ndcg_at_5_max
value: 12.8492
- type: nauc_ndcg_at_5_std
value: 18.5369
- type: nauc_ndcg_at_5_diff1
value: 24.5013
- type: nauc_ndcg_at_10_max
value: 10.3186
- type: nauc_ndcg_at_10_std
value: 16.8747
- type: nauc_ndcg_at_10_diff1
value: 22.6062
- type: nauc_ndcg_at_20_max
value: 11.910400000000001
- type: nauc_ndcg_at_20_std
value: 18.9906
- type: nauc_ndcg_at_20_diff1
value: 21.0736
- type: nauc_ndcg_at_100_max
value: 13.780000000000001
- type: nauc_ndcg_at_100_std
value: 20.2702
- type: nauc_ndcg_at_100_diff1
value: 23.7899
- type: nauc_ndcg_at_1000_max
value: 12.9736
- type: nauc_ndcg_at_1000_std
value: 19.4173
- type: nauc_ndcg_at_1000_diff1
value: 24.0248
- type: nauc_map_at_1_max
value: 21.683
- type: nauc_map_at_1_std
value: 23.9115
- type: nauc_map_at_1_diff1
value: 34.306799999999996
- type: nauc_map_at_3_max
value: 13.7629
- type: nauc_map_at_3_std
value: 19.4925
- type: nauc_map_at_3_diff1
value: 26.8286
- type: nauc_map_at_5_max
value: 14.602200000000002
- type: nauc_map_at_5_std
value: 19.8349
- type: nauc_map_at_5_diff1
value: 26.6756
- type: nauc_map_at_10_max
value: 13.5297
- type: nauc_map_at_10_std
value: 19.0117
- type: nauc_map_at_10_diff1
value: 25.803900000000002
- type: nauc_map_at_20_max
value: 14.0185
- type: nauc_map_at_20_std
value: 19.667399999999997
- type: nauc_map_at_20_diff1
value: 25.265900000000002
- type: nauc_map_at_100_max
value: 14.1821
- type: nauc_map_at_100_std
value: 19.8468
- type: nauc_map_at_100_diff1
value: 25.7233
- type: nauc_map_at_1000_max
value: 14.1415
- type: nauc_map_at_1000_std
value: 19.8004
- type: nauc_map_at_1000_diff1
value: 25.7339
- type: nauc_recall_at_1_max
value: 21.683
- type: nauc_recall_at_1_std
value: 23.9115
- type: nauc_recall_at_1_diff1
value: 34.306799999999996
- type: nauc_recall_at_3_max
value: 4.0852
- type: nauc_recall_at_3_std
value: 13.7371
- type: nauc_recall_at_3_diff1
value: 18.2104
- type: nauc_recall_at_5_max
value: 9.3363
- type: nauc_recall_at_5_std
value: 15.767500000000002
- type: nauc_recall_at_5_diff1
value: 19.948
- type: nauc_recall_at_10_max
value: 3.3214
- type: nauc_recall_at_10_std
value: 12.2687
- type: nauc_recall_at_10_diff1
value: 15.7946
- type: nauc_recall_at_20_max
value: 8.2034
- type: nauc_recall_at_20_std
value: 18.5331
- type: nauc_recall_at_20_diff1
value: 12.0362
- type: nauc_recall_at_100_max
value: 23.0879
- type: nauc_recall_at_100_std
value: 30.133399999999998
- type: nauc_recall_at_100_diff1
value: 20.4628
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 21.683
- type: nauc_precision_at_1_std
value: 23.9115
- type: nauc_precision_at_1_diff1
value: 34.306799999999996
- type: nauc_precision_at_3_max
value: 4.0852
- type: nauc_precision_at_3_std
value: 13.7371
- type: nauc_precision_at_3_diff1
value: 18.2104
- type: nauc_precision_at_5_max
value: 9.3363
- type: nauc_precision_at_5_std
value: 15.767500000000002
- type: nauc_precision_at_5_diff1
value: 19.948
- type: nauc_precision_at_10_max
value: 3.3214
- type: nauc_precision_at_10_std
value: 12.2687
- type: nauc_precision_at_10_diff1
value: 15.7946
- type: nauc_precision_at_20_max
value: 8.2034
- type: nauc_precision_at_20_std
value: 18.5331
- type: nauc_precision_at_20_diff1
value: 12.0362
- type: nauc_precision_at_100_max
value: 23.0879
- type: nauc_precision_at_100_std
value: 30.133399999999998
- type: nauc_precision_at_100_diff1
value: 20.4628
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 21.683
- type: nauc_mrr_at_1_std
value: 23.9115
- type: nauc_mrr_at_1_diff1
value: 34.306799999999996
- type: nauc_mrr_at_3_max
value: 13.7629
- type: nauc_mrr_at_3_std
value: 19.4925
- type: nauc_mrr_at_3_diff1
value: 26.8286
- type: nauc_mrr_at_5_max
value: 14.602200000000002
- type: nauc_mrr_at_5_std
value: 19.8349
- type: nauc_mrr_at_5_diff1
value: 26.6756
- type: nauc_mrr_at_10_max
value: 13.5297
- type: nauc_mrr_at_10_std
value: 19.0117
- type: nauc_mrr_at_10_diff1
value: 25.803900000000002
- type: nauc_mrr_at_20_max
value: 14.0185
- type: nauc_mrr_at_20_std
value: 19.667399999999997
- type: nauc_mrr_at_20_diff1
value: 25.265900000000002
- type: nauc_mrr_at_100_max
value: 14.1821
- type: nauc_mrr_at_100_std
value: 19.8468
- type: nauc_mrr_at_100_diff1
value: 25.7233
- type: nauc_mrr_at_1000_max
value: 14.1415
- type: nauc_mrr_at_1000_std
value: 19.8004
- type: nauc_mrr_at_1000_diff1
value: 25.7339
- type: main_score
value: 18.429000000000002
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (hin-ara)
type: facebook/mlqa
config: hin-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 8.602
- type: ndcg_at_3
value: 11.105
- type: ndcg_at_5
value: 12.447
- type: ndcg_at_10
value: 14.274999999999999
- type: ndcg_at_20
value: 16.699
- type: ndcg_at_100
value: 24.785
- type: ndcg_at_1000
value: 27.950999999999997
- type: map_at_1
value: 8.602
- type: map_at_3
value: 10.484
- type: map_at_5
value: 11.237
- type: map_at_10
value: 11.943
- type: map_at_20
value: 12.597
- type: map_at_100
value: 13.536999999999999
- type: map_at_1000
value: 13.716000000000001
- type: recall_at_1
value: 8.602
- type: recall_at_3
value: 12.903
- type: recall_at_5
value: 16.128999999999998
- type: recall_at_10
value: 22.043
- type: recall_at_20
value: 31.72
- type: recall_at_100
value: 77.957
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 8.602
- type: precision_at_3
value: 4.301
- type: precision_at_5
value: 3.2259999999999995
- type: precision_at_10
value: 2.204
- type: precision_at_20
value: 1.5859999999999999
- type: precision_at_100
value: 0.7799999999999999
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 8.6022
- type: mrr_at_3
value: 10.4839
- type: mrr_at_5
value: 11.2366
- type: mrr_at_10
value: 11.9427
- type: mrr_at_20
value: 12.5969
- type: mrr_at_100
value: 13.536999999999999
- type: mrr_at_1000
value: 13.7157
- type: nauc_ndcg_at_1_max
value: 43.5676
- type: nauc_ndcg_at_1_std
value: 48.1034
- type: nauc_ndcg_at_1_diff1
value: 34.3343
- type: nauc_ndcg_at_3_max
value: 34.779700000000005
- type: nauc_ndcg_at_3_std
value: 41.8153
- type: nauc_ndcg_at_3_diff1
value: 22.459100000000003
- type: nauc_ndcg_at_5_max
value: 36.9668
- type: nauc_ndcg_at_5_std
value: 41.5695
- type: nauc_ndcg_at_5_diff1
value: 25.2023
- type: nauc_ndcg_at_10_max
value: 31.114399999999996
- type: nauc_ndcg_at_10_std
value: 37.7021
- type: nauc_ndcg_at_10_diff1
value: 17.8647
- type: nauc_ndcg_at_20_max
value: 27.8539
- type: nauc_ndcg_at_20_std
value: 34.7643
- type: nauc_ndcg_at_20_diff1
value: 18.7205
- type: nauc_ndcg_at_100_max
value: 26.2928
- type: nauc_ndcg_at_100_std
value: 33.4221
- type: nauc_ndcg_at_100_diff1
value: 18.186
- type: nauc_ndcg_at_1000_max
value: 30.8904
- type: nauc_ndcg_at_1000_std
value: 37.4835
- type: nauc_ndcg_at_1000_diff1
value: 21.073
- type: nauc_map_at_1_max
value: 43.5676
- type: nauc_map_at_1_std
value: 48.1034
- type: nauc_map_at_1_diff1
value: 34.3343
- type: nauc_map_at_3_max
value: 36.4446
- type: nauc_map_at_3_std
value: 43.3032
- type: nauc_map_at_3_diff1
value: 25.0872
- type: nauc_map_at_5_max
value: 37.5909
- type: nauc_map_at_5_std
value: 42.9831
- type: nauc_map_at_5_diff1
value: 26.600800000000003
- type: nauc_map_at_10_max
value: 35.0221
- type: nauc_map_at_10_std
value: 41.1277
- type: nauc_map_at_10_diff1
value: 23.2872
- type: nauc_map_at_20_max
value: 33.861799999999995
- type: nauc_map_at_20_std
value: 40.1421
- type: nauc_map_at_20_diff1
value: 23.421300000000002
- type: nauc_map_at_100_max
value: 33.6519
- type: nauc_map_at_100_std
value: 39.9834
- type: nauc_map_at_100_diff1
value: 23.427400000000002
- type: nauc_map_at_1000_max
value: 33.949400000000004
- type: nauc_map_at_1000_std
value: 40.2444
- type: nauc_map_at_1000_diff1
value: 23.603099999999998
- type: nauc_recall_at_1_max
value: 43.5676
- type: nauc_recall_at_1_std
value: 48.1034
- type: nauc_recall_at_1_diff1
value: 34.3343
- type: nauc_recall_at_3_max
value: 30.7755
- type: nauc_recall_at_3_std
value: 38.1252
- type: nauc_recall_at_3_diff1
value: 15.996099999999998
- type: nauc_recall_at_5_max
value: 35.975
- type: nauc_recall_at_5_std
value: 38.5188
- type: nauc_recall_at_5_diff1
value: 22.4214
- type: nauc_recall_at_10_max
value: 22.4406
- type: nauc_recall_at_10_std
value: 30.440800000000003
- type: nauc_recall_at_10_diff1
value: 5.9871
- type: nauc_recall_at_20_max
value: 15.343599999999999
- type: nauc_recall_at_20_std
value: 23.7135
- type: nauc_recall_at_20_diff1
value: 10.032
- type: nauc_recall_at_100_max
value: -1.9075000000000002
- type: nauc_recall_at_100_std
value: 8.4695
- type: nauc_recall_at_100_diff1
value: -0.0034999999999999996
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 43.5676
- type: nauc_precision_at_1_std
value: 48.1034
- type: nauc_precision_at_1_diff1
value: 34.3343
- type: nauc_precision_at_3_max
value: 30.7755
- type: nauc_precision_at_3_std
value: 38.1252
- type: nauc_precision_at_3_diff1
value: 15.996099999999998
- type: nauc_precision_at_5_max
value: 35.975
- type: nauc_precision_at_5_std
value: 38.5188
- type: nauc_precision_at_5_diff1
value: 22.4214
- type: nauc_precision_at_10_max
value: 22.4406
- type: nauc_precision_at_10_std
value: 30.440800000000003
- type: nauc_precision_at_10_diff1
value: 5.9871
- type: nauc_precision_at_20_max
value: 15.343599999999999
- type: nauc_precision_at_20_std
value: 23.7135
- type: nauc_precision_at_20_diff1
value: 10.032
- type: nauc_precision_at_100_max
value: -1.9075000000000002
- type: nauc_precision_at_100_std
value: 8.4695
- type: nauc_precision_at_100_diff1
value: -0.0034999999999999996
- type: nauc_precision_at_1000_max
value: .nan
- type: nauc_precision_at_1000_std
value: .nan
- type: nauc_precision_at_1000_diff1
value: .nan
- type: nauc_mrr_at_1_max
value: 43.5676
- type: nauc_mrr_at_1_std
value: 48.1034
- type: nauc_mrr_at_1_diff1
value: 34.3343
- type: nauc_mrr_at_3_max
value: 36.4446
- type: nauc_mrr_at_3_std
value: 43.3032
- type: nauc_mrr_at_3_diff1
value: 25.0872
- type: nauc_mrr_at_5_max
value: 37.5909
- type: nauc_mrr_at_5_std
value: 42.9831
- type: nauc_mrr_at_5_diff1
value: 26.600800000000003
- type: nauc_mrr_at_10_max
value: 35.0221
- type: nauc_mrr_at_10_std
value: 41.1277
- type: nauc_mrr_at_10_diff1
value: 23.2872
- type: nauc_mrr_at_20_max
value: 33.861799999999995
- type: nauc_mrr_at_20_std
value: 40.1421
- type: nauc_mrr_at_20_diff1
value: 23.421300000000002
- type: nauc_mrr_at_100_max
value: 33.6519
- type: nauc_mrr_at_100_std
value: 39.9834
- type: nauc_mrr_at_100_diff1
value: 23.427400000000002
- type: nauc_mrr_at_1000_max
value: 33.949400000000004
- type: nauc_mrr_at_1000_std
value: 40.2444
- type: nauc_mrr_at_1000_diff1
value: 23.603099999999998
- type: main_score
value: 14.274999999999999
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (vie-ara)
type: facebook/mlqa
config: vie-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 9.202
- type: ndcg_at_3
value: 14.219999999999999
- type: ndcg_at_5
value: 17.913999999999998
- type: ndcg_at_10
value: 20.875
- type: ndcg_at_20
value: 23.504
- type: ndcg_at_100
value: 31.275
- type: ndcg_at_1000
value: 32.696999999999996
- type: map_at_1
value: 9.202
- type: map_at_3
value: 12.986
- type: map_at_5
value: 14.979999999999999
- type: map_at_10
value: 16.191
- type: map_at_20
value: 16.909
- type: map_at_100
value: 17.877000000000002
- type: map_at_1000
value: 17.96
- type: recall_at_1
value: 9.202
- type: recall_at_3
value: 17.791
- type: recall_at_5
value: 26.994
- type: recall_at_10
value: 36.196
- type: recall_at_20
value: 46.626
- type: recall_at_100
value: 90.184
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 9.202
- type: precision_at_3
value: 5.93
- type: precision_at_5
value: 5.399
- type: precision_at_10
value: 3.62
- type: precision_at_20
value: 2.331
- type: precision_at_100
value: 0.902
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 9.202499999999999
- type: mrr_at_3
value: 12.9857
- type: mrr_at_5
value: 14.979600000000001
- type: mrr_at_10
value: 16.191
- type: mrr_at_20
value: 16.9095
- type: mrr_at_100
value: 17.877299999999998
- type: mrr_at_1000
value: 17.9603
- type: nauc_ndcg_at_1_max
value: 62.9598
- type: nauc_ndcg_at_1_std
value: 49.065999999999995
- type: nauc_ndcg_at_1_diff1
value: 56.008500000000005
- type: nauc_ndcg_at_3_max
value: 53.9189
- type: nauc_ndcg_at_3_std
value: 44.1455
- type: nauc_ndcg_at_3_diff1
value: 41.287600000000005
- type: nauc_ndcg_at_5_max
value: 49.749500000000005
- type: nauc_ndcg_at_5_std
value: 41.1122
- type: nauc_ndcg_at_5_diff1
value: 40.7353
- type: nauc_ndcg_at_10_max
value: 53.8852
- type: nauc_ndcg_at_10_std
value: 44.7395
- type: nauc_ndcg_at_10_diff1
value: 38.6166
- type: nauc_ndcg_at_20_max
value: 55.237199999999994
- type: nauc_ndcg_at_20_std
value: 46.7695
- type: nauc_ndcg_at_20_diff1
value: 38.804
- type: nauc_ndcg_at_100_max
value: 53.2497
- type: nauc_ndcg_at_100_std
value: 46.9584
- type: nauc_ndcg_at_100_diff1
value: 38.8298
- type: nauc_ndcg_at_1000_max
value: 53.9127
- type: nauc_ndcg_at_1000_std
value: 45.8294
- type: nauc_ndcg_at_1000_diff1
value: 40.0041
- type: nauc_map_at_1_max
value: 62.9598
- type: nauc_map_at_1_std
value: 49.065999999999995
- type: nauc_map_at_1_diff1
value: 56.008500000000005
- type: nauc_map_at_3_max
value: 55.3652
- type: nauc_map_at_3_std
value: 44.9791
- type: nauc_map_at_3_diff1
value: 44.052
- type: nauc_map_at_5_max
value: 52.735200000000006
- type: nauc_map_at_5_std
value: 43.1035
- type: nauc_map_at_5_diff1
value: 43.2012
- type: nauc_map_at_10_max
value: 54.786500000000004
- type: nauc_map_at_10_std
value: 44.8598
- type: nauc_map_at_10_diff1
value: 42.103
- type: nauc_map_at_20_max
value: 55.10620000000001
- type: nauc_map_at_20_std
value: 45.5114
- type: nauc_map_at_20_diff1
value: 42.032799999999995
- type: nauc_map_at_100_max
value: 54.6794
- type: nauc_map_at_100_std
value: 45.5176
- type: nauc_map_at_100_diff1
value: 41.9804
- type: nauc_map_at_1000_max
value: 54.7162
- type: nauc_map_at_1000_std
value: 45.4536
- type: nauc_map_at_1000_diff1
value: 42.0517
- type: nauc_recall_at_1_max
value: 62.9598
- type: nauc_recall_at_1_std
value: 49.065999999999995
- type: nauc_recall_at_1_diff1
value: 56.008500000000005
- type: nauc_recall_at_3_max
value: 50.73180000000001
- type: nauc_recall_at_3_std
value: 42.2909
- type: nauc_recall_at_3_diff1
value: 35.0404
- type: nauc_recall_at_5_max
value: 43.5873
- type: nauc_recall_at_5_std
value: 36.9356
- type: nauc_recall_at_5_diff1
value: 36.1826
- type: nauc_recall_at_10_max
value: 52.7111
- type: nauc_recall_at_10_std
value: 45.025999999999996
- type: nauc_recall_at_10_diff1
value: 32.0134
- type: nauc_recall_at_20_max
value: 57.0465
- type: nauc_recall_at_20_std
value: 50.73839999999999
- type: nauc_recall_at_20_diff1
value: 33.0878
- type: nauc_recall_at_100_max
value: 43.736399999999996
- type: nauc_recall_at_100_std
value: 62.805
- type: nauc_recall_at_100_diff1
value: 22.2379
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 62.9598
- type: nauc_precision_at_1_std
value: 49.065999999999995
- type: nauc_precision_at_1_diff1
value: 56.008500000000005
- type: nauc_precision_at_3_max
value: 50.73180000000001
- type: nauc_precision_at_3_std
value: 42.2909
- type: nauc_precision_at_3_diff1
value: 35.0404
- type: nauc_precision_at_5_max
value: 43.5873
- type: nauc_precision_at_5_std
value: 36.9356
- type: nauc_precision_at_5_diff1
value: 36.1826
- type: nauc_precision_at_10_max
value: 52.7111
- type: nauc_precision_at_10_std
value: 45.025999999999996
- type: nauc_precision_at_10_diff1
value: 32.0134
- type: nauc_precision_at_20_max
value: 57.0465
- type: nauc_precision_at_20_std
value: 50.73839999999999
- type: nauc_precision_at_20_diff1
value: 33.0878
- type: nauc_precision_at_100_max
value: 43.736399999999996
- type: nauc_precision_at_100_std
value: 62.805
- type: nauc_precision_at_100_diff1
value: 22.2379
- type: nauc_precision_at_1000_max
value: 100.0
- type: nauc_precision_at_1000_std
value: 100.0
- type: nauc_precision_at_1000_diff1
value: 100.0
- type: nauc_mrr_at_1_max
value: 62.9598
- type: nauc_mrr_at_1_std
value: 49.065999999999995
- type: nauc_mrr_at_1_diff1
value: 56.008500000000005
- type: nauc_mrr_at_3_max
value: 55.3652
- type: nauc_mrr_at_3_std
value: 44.9791
- type: nauc_mrr_at_3_diff1
value: 44.052
- type: nauc_mrr_at_5_max
value: 52.735200000000006
- type: nauc_mrr_at_5_std
value: 43.1035
- type: nauc_mrr_at_5_diff1
value: 43.2012
- type: nauc_mrr_at_10_max
value: 54.786500000000004
- type: nauc_mrr_at_10_std
value: 44.8598
- type: nauc_mrr_at_10_diff1
value: 42.103
- type: nauc_mrr_at_20_max
value: 55.10620000000001
- type: nauc_mrr_at_20_std
value: 45.5114
- type: nauc_mrr_at_20_diff1
value: 42.032799999999995
- type: nauc_mrr_at_100_max
value: 54.6794
- type: nauc_mrr_at_100_std
value: 45.5176
- type: nauc_mrr_at_100_diff1
value: 41.9804
- type: nauc_mrr_at_1000_max
value: 54.7162
- type: nauc_mrr_at_1000_std
value: 45.4536
- type: nauc_mrr_at_1000_diff1
value: 42.0517
- type: main_score
value: 20.875
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (zho-ara)
type: facebook/mlqa
config: zho-ara
split: validation
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 6.383
- type: ndcg_at_3
value: 10.999
- type: ndcg_at_5
value: 12.762
- type: ndcg_at_10
value: 15.151
- type: ndcg_at_20
value: 17.394000000000002
- type: ndcg_at_100
value: 24.684
- type: ndcg_at_1000
value: 28.025
- type: map_at_1
value: 6.383
- type: map_at_3
value: 9.84
- type: map_at_5
value: 10.824
- type: map_at_10
value: 11.797
- type: map_at_20
value: 12.389999999999999
- type: map_at_100
value: 13.269
- type: map_at_1000
value: 13.453999999999999
- type: recall_at_1
value: 6.383
- type: recall_at_3
value: 14.362
- type: recall_at_5
value: 18.617
- type: recall_at_10
value: 26.064
- type: recall_at_20
value: 35.106
- type: recall_at_100
value: 76.596
- type: recall_at_1000
value: 100.0
- type: precision_at_1
value: 6.383
- type: precision_at_3
value: 4.787
- type: precision_at_5
value: 3.723
- type: precision_at_10
value: 2.606
- type: precision_at_20
value: 1.755
- type: precision_at_100
value: 0.766
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 6.383
- type: mrr_at_3
value: 9.8404
- type: mrr_at_5
value: 10.824499999999999
- type: mrr_at_10
value: 11.7969
- type: mrr_at_20
value: 12.3905
- type: mrr_at_100
value: 13.2692
- type: mrr_at_1000
value: 13.4538
- type: nauc_ndcg_at_1_max
value: 28.7389
- type: nauc_ndcg_at_1_std
value: 64.9286
- type: nauc_ndcg_at_1_diff1
value: 10.673499999999999
- type: nauc_ndcg_at_3_max
value: 19.4744
- type: nauc_ndcg_at_3_std
value: 44.7069
- type: nauc_ndcg_at_3_diff1
value: 6.631099999999999
- type: nauc_ndcg_at_5_max
value: 18.2711
- type: nauc_ndcg_at_5_std
value: 43.5962
- type: nauc_ndcg_at_5_diff1
value: 6.307500000000001
- type: nauc_ndcg_at_10_max
value: 20.0539
- type: nauc_ndcg_at_10_std
value: 43.5587
- type: nauc_ndcg_at_10_diff1
value: 5.6582
- type: nauc_ndcg_at_20_max
value: 22.5386
- type: nauc_ndcg_at_20_std
value: 42.9099
- type: nauc_ndcg_at_20_diff1
value: 7.5015
- type: nauc_ndcg_at_100_max
value: 21.0851
- type: nauc_ndcg_at_100_std
value: 41.966300000000004
- type: nauc_ndcg_at_100_diff1
value: 6.9177
- type: nauc_ndcg_at_1000_max
value: 20.7669
- type: nauc_ndcg_at_1000_std
value: 43.8782
- type: nauc_ndcg_at_1000_diff1
value: 6.9428
- type: nauc_map_at_1_max
value: 28.7389
- type: nauc_map_at_1_std
value: 64.9286
- type: nauc_map_at_1_diff1
value: 10.673499999999999
- type: nauc_map_at_3_max
value: 20.319499999999998
- type: nauc_map_at_3_std
value: 47.6539
- type: nauc_map_at_3_diff1
value: 7.452
- type: nauc_map_at_5_max
value: 19.7223
- type: nauc_map_at_5_std
value: 46.928799999999995
- type: nauc_map_at_5_diff1
value: 7.2603
- type: nauc_map_at_10_max
value: 20.624000000000002
- type: nauc_map_at_10_std
value: 46.9846
- type: nauc_map_at_10_diff1
value: 6.9296999999999995
- type: nauc_map_at_20_max
value: 21.3628
- type: nauc_map_at_20_std
value: 46.7418
- type: nauc_map_at_20_diff1
value: 7.3283000000000005
- type: nauc_map_at_100_max
value: 21.023500000000002
- type: nauc_map_at_100_std
value: 46.319900000000004
- type: nauc_map_at_100_diff1
value: 7.2962
- type: nauc_map_at_1000_max
value: 20.9867
- type: nauc_map_at_1000_std
value: 46.4588
- type: nauc_map_at_1000_diff1
value: 7.281899999999999
- type: nauc_recall_at_1_max
value: 28.7389
- type: nauc_recall_at_1_std
value: 64.9286
- type: nauc_recall_at_1_diff1
value: 10.673499999999999
- type: nauc_recall_at_3_max
value: 17.924100000000003
- type: nauc_recall_at_3_std
value: 38.7062
- type: nauc_recall_at_3_diff1
value: 4.8814
- type: nauc_recall_at_5_max
value: 15.5025
- type: nauc_recall_at_5_std
value: 37.3735
- type: nauc_recall_at_5_diff1
value: 4.4486
- type: nauc_recall_at_10_max
value: 19.336000000000002
- type: nauc_recall_at_10_std
value: 37.6921
- type: nauc_recall_at_10_diff1
value: 3.3455
- type: nauc_recall_at_20_max
value: 25.874799999999997
- type: nauc_recall_at_20_std
value: 36.5078
- type: nauc_recall_at_20_diff1
value: 8.8964
- type: nauc_recall_at_100_max
value: 22.3107
- type: nauc_recall_at_100_std
value: 31.202800000000003
- type: nauc_recall_at_100_diff1
value: 6.2387999999999995
- type: nauc_recall_at_1000_max
value: .nan
- type: nauc_recall_at_1000_std
value: .nan
- type: nauc_recall_at_1000_diff1
value: .nan
- type: nauc_precision_at_1_max
value: 28.7389
- type: nauc_precision_at_1_std
value: 64.9286
- type: nauc_precision_at_1_diff1
value: 10.673499999999999
- type: nauc_precision_at_3_max
value: 17.924100000000003
- type: nauc_precision_at_3_std
value: 38.7062
- type: nauc_precision_at_3_diff1
value: 4.8814
- type: nauc_precision_at_5_max
value: 15.5025
- type: nauc_precision_at_5_std
value: 37.3735
- type: nauc_precision_at_5_diff1
value: 4.4486
- type: nauc_precision_at_10_max
value: 19.336000000000002
- type: nauc_precision_at_10_std
value: 37.6921
- type: nauc_precision_at_10_diff1
value: 3.3455
- type: nauc_precision_at_20_max
value: 25.874799999999997
- type: nauc_precision_at_20_std
value: 36.5078
- type: nauc_precision_at_20_diff1
value: 8.8964
- type: nauc_precision_at_100_max
value: 22.3107
- type: nauc_precision_at_100_std
value: 31.202800000000003
- type: nauc_precision_at_100_diff1
value: 6.2387999999999995
- type: nauc_precision_at_1000_max
value: .nan
- type: nauc_precision_at_1000_std
value: .nan
- type: nauc_precision_at_1000_diff1
value: .nan
- type: nauc_mrr_at_1_max
value: 28.7389
- type: nauc_mrr_at_1_std
value: 64.9286
- type: nauc_mrr_at_1_diff1
value: 10.673499999999999
- type: nauc_mrr_at_3_max
value: 20.319499999999998
- type: nauc_mrr_at_3_std
value: 47.6539
- type: nauc_mrr_at_3_diff1
value: 7.452
- type: nauc_mrr_at_5_max
value: 19.7223
- type: nauc_mrr_at_5_std
value: 46.928799999999995
- type: nauc_mrr_at_5_diff1
value: 7.2603
- type: nauc_mrr_at_10_max
value: 20.624000000000002
- type: nauc_mrr_at_10_std
value: 46.9846
- type: nauc_mrr_at_10_diff1
value: 6.9296999999999995
- type: nauc_mrr_at_20_max
value: 21.3628
- type: nauc_mrr_at_20_std
value: 46.7418
- type: nauc_mrr_at_20_diff1
value: 7.3283000000000005
- type: nauc_mrr_at_100_max
value: 21.0238
- type: nauc_mrr_at_100_std
value: 46.319900000000004
- type: nauc_mrr_at_100_diff1
value: 7.2976
- type: nauc_mrr_at_1000_max
value: 20.987000000000002
- type: nauc_mrr_at_1000_std
value: 46.4588
- type: nauc_mrr_at_1000_diff1
value: 7.2833
- type: main_score
value: 15.151
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-ara)
type: facebook/mlqa
config: ara-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 32.496
- type: ndcg_at_3
value: 40.172000000000004
- type: ndcg_at_5
value: 42.588
- type: ndcg_at_10
value: 45.078
- type: ndcg_at_20
value: 46.814
- type: ndcg_at_100
value: 49.696
- type: ndcg_at_1000
value: 51.466
- type: map_at_1
value: 32.486
- type: map_at_3
value: 38.271
- type: map_at_5
value: 39.606
- type: map_at_10
value: 40.647
- type: map_at_20
value: 41.121
- type: map_at_100
value: 41.512
- type: map_at_1000
value: 41.573
- type: recall_at_1
value: 32.486
- type: recall_at_3
value: 45.668
- type: recall_at_5
value: 51.556000000000004
- type: recall_at_10
value: 59.187999999999995
- type: recall_at_20
value: 66.07
- type: recall_at_100
value: 81.699
- type: recall_at_1000
value: 95.959
- type: precision_at_1
value: 32.496
- type: precision_at_3
value: 15.226
- type: precision_at_5
value: 10.313
- type: precision_at_10
value: 5.92
- type: precision_at_20
value: 3.304
- type: precision_at_100
value: 0.8170000000000001
- type: precision_at_1000
value: 0.096
- type: mrr_at_1
value: 32.4958
- type: mrr_at_3
value: 38.2805
- type: mrr_at_5
value: 39.6156
- type: mrr_at_10
value: 40.6564
- type: mrr_at_20
value: 41.1308
- type: mrr_at_100
value: 41.5219
- type: mrr_at_1000
value: 41.5827
- type: nauc_ndcg_at_1_max
value: 45.3065
- type: nauc_ndcg_at_1_std
value: 8.438600000000001
- type: nauc_ndcg_at_1_diff1
value: 56.5996
- type: nauc_ndcg_at_3_max
value: 45.677800000000005
- type: nauc_ndcg_at_3_std
value: 11.2794
- type: nauc_ndcg_at_3_diff1
value: 49.1837
- type: nauc_ndcg_at_5_max
value: 45.988
- type: nauc_ndcg_at_5_std
value: 12.4386
- type: nauc_ndcg_at_5_diff1
value: 47.3708
- type: nauc_ndcg_at_10_max
value: 46.305800000000005
- type: nauc_ndcg_at_10_std
value: 13.8563
- type: nauc_ndcg_at_10_diff1
value: 46.2161
- type: nauc_ndcg_at_20_max
value: 46.547
- type: nauc_ndcg_at_20_std
value: 14.746500000000001
- type: nauc_ndcg_at_20_diff1
value: 45.8241
- type: nauc_ndcg_at_100_max
value: 46.8223
- type: nauc_ndcg_at_100_std
value: 15.3285
- type: nauc_ndcg_at_100_diff1
value: 46.470099999999995
- type: nauc_ndcg_at_1000_max
value: 46.6777
- type: nauc_ndcg_at_1000_std
value: 14.3656
- type: nauc_ndcg_at_1000_diff1
value: 47.3024
- type: nauc_map_at_1_max
value: 45.277699999999996
- type: nauc_map_at_1_std
value: 8.4486
- type: nauc_map_at_1_diff1
value: 56.5556
- type: nauc_map_at_3_max
value: 45.536100000000005
- type: nauc_map_at_3_std
value: 10.555100000000001
- type: nauc_map_at_3_diff1
value: 50.8511
- type: nauc_map_at_5_max
value: 45.6962
- type: nauc_map_at_5_std
value: 11.1708
- type: nauc_map_at_5_diff1
value: 49.8493
- type: nauc_map_at_10_max
value: 45.83
- type: nauc_map_at_10_std
value: 11.7378
- type: nauc_map_at_10_diff1
value: 49.4193
- type: nauc_map_at_20_max
value: 45.881699999999995
- type: nauc_map_at_20_std
value: 11.9504
- type: nauc_map_at_20_diff1
value: 49.330600000000004
- type: nauc_map_at_100_max
value: 45.923700000000004
- type: nauc_map_at_100_std
value: 12.0218
- type: nauc_map_at_100_diff1
value: 49.4458
- type: nauc_map_at_1000_max
value: 45.9216
- type: nauc_map_at_1000_std
value: 11.9945
- type: nauc_map_at_1000_diff1
value: 49.4724
- type: nauc_recall_at_1_max
value: 45.277699999999996
- type: nauc_recall_at_1_std
value: 8.4486
- type: nauc_recall_at_1_diff1
value: 56.5556
- type: nauc_recall_at_3_max
value: 46.0736
- type: nauc_recall_at_3_std
value: 13.3868
- type: nauc_recall_at_3_diff1
value: 44.3913
- type: nauc_recall_at_5_max
value: 46.8911
- type: nauc_recall_at_5_std
value: 16.392799999999998
- type: nauc_recall_at_5_diff1
value: 39.8177
- type: nauc_recall_at_10_max
value: 47.9748
- type: nauc_recall_at_10_std
value: 21.4029
- type: nauc_recall_at_10_diff1
value: 35.2649
- type: nauc_recall_at_20_max
value: 49.3908
- type: nauc_recall_at_20_std
value: 26.6036
- type: nauc_recall_at_20_diff1
value: 32.0814
- type: nauc_recall_at_100_max
value: 53.539
- type: nauc_recall_at_100_std
value: 39.2579
- type: nauc_recall_at_100_diff1
value: 29.483500000000003
- type: nauc_recall_at_1000_max
value: 65.35640000000001
- type: nauc_recall_at_1000_std
value: 57.158699999999996
- type: nauc_recall_at_1000_diff1
value: 24.557399999999998
- type: nauc_precision_at_1_max
value: 45.3065
- type: nauc_precision_at_1_std
value: 8.438600000000001
- type: nauc_precision_at_1_diff1
value: 56.5996
- type: nauc_precision_at_3_max
value: 46.1054
- type: nauc_precision_at_3_std
value: 13.3778
- type: nauc_precision_at_3_diff1
value: 44.4386
- type: nauc_precision_at_5_max
value: 46.927
- type: nauc_precision_at_5_std
value: 16.3847
- type: nauc_precision_at_5_diff1
value: 39.868900000000004
- type: nauc_precision_at_10_max
value: 48.0138
- type: nauc_precision_at_10_std
value: 21.3945
- type: nauc_precision_at_10_diff1
value: 35.3201
- type: nauc_precision_at_20_max
value: 49.4384
- type: nauc_precision_at_20_std
value: 26.5966
- type: nauc_precision_at_20_diff1
value: 32.1454
- type: nauc_precision_at_100_max
value: 53.60510000000001
- type: nauc_precision_at_100_std
value: 39.245400000000004
- type: nauc_precision_at_100_diff1
value: 29.5996
- type: nauc_precision_at_1000_max
value: 65.31320000000001
- type: nauc_precision_at_1000_std
value: 56.5386
- type: nauc_precision_at_1000_diff1
value: 25.1914
- type: nauc_mrr_at_1_max
value: 45.3065
- type: nauc_mrr_at_1_std
value: 8.438600000000001
- type: nauc_mrr_at_1_diff1
value: 56.5996
- type: nauc_mrr_at_3_max
value: 45.5645
- type: nauc_mrr_at_3_std
value: 10.5451
- type: nauc_mrr_at_3_diff1
value: 50.8949
- type: nauc_mrr_at_5_max
value: 45.7248
- type: nauc_mrr_at_5_std
value: 11.1608
- type: nauc_mrr_at_5_diff1
value: 49.8934
- type: nauc_mrr_at_10_max
value: 45.858900000000006
- type: nauc_mrr_at_10_std
value: 11.7276
- type: nauc_mrr_at_10_diff1
value: 49.464000000000006
- type: nauc_mrr_at_20_max
value: 45.9109
- type: nauc_mrr_at_20_std
value: 11.9401
- type: nauc_mrr_at_20_diff1
value: 49.3755
- type: nauc_mrr_at_100_max
value: 45.953
- type: nauc_mrr_at_100_std
value: 12.0114
- type: nauc_mrr_at_100_diff1
value: 49.4912
- type: nauc_mrr_at_1000_max
value: 45.9504
- type: nauc_mrr_at_1000_std
value: 11.984200000000001
- type: nauc_mrr_at_1000_diff1
value: 49.5171
- type: main_score
value: 45.078
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-deu)
type: facebook/mlqa
config: ara-deu
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 0.364
- type: ndcg_at_3
value: 1.103
- type: ndcg_at_5
value: 1.482
- type: ndcg_at_10
value: 2.275
- type: ndcg_at_20
value: 2.964
- type: ndcg_at_100
value: 5.203
- type: ndcg_at_1000
value: 12.245000000000001
- type: map_at_1
value: 0.364
- type: map_at_3
value: 0.8999999999999999
- type: map_at_5
value: 1.1119999999999999
- type: map_at_10
value: 1.434
- type: map_at_20
value: 1.6129999999999998
- type: map_at_100
value: 1.881
- type: map_at_1000
value: 2.067
- type: recall_at_1
value: 0.364
- type: recall_at_3
value: 1.699
- type: recall_at_5
value: 2.609
- type: recall_at_10
value: 5.097
- type: recall_at_20
value: 7.888000000000001
- type: recall_at_100
value: 20.57
- type: recall_at_1000
value: 80.734
- type: precision_at_1
value: 0.364
- type: precision_at_3
value: 0.5660000000000001
- type: precision_at_5
value: 0.522
- type: precision_at_10
value: 0.51
- type: precision_at_20
value: 0.394
- type: precision_at_100
value: 0.20600000000000002
- type: precision_at_1000
value: 0.08099999999999999
- type: mrr_at_1
value: 0.36410000000000003
- type: mrr_at_3
value: 0.9001
- type: mrr_at_5
value: 1.1125
- type: mrr_at_10
value: 1.4337
- type: mrr_at_20
value: 1.6132
- type: mrr_at_100
value: 1.8812
- type: mrr_at_1000
value: 2.0674
- type: nauc_ndcg_at_1_max
value: -3.7518999999999996
- type: nauc_ndcg_at_1_std
value: -29.5265
- type: nauc_ndcg_at_1_diff1
value: -9.383
- type: nauc_ndcg_at_3_max
value: -12.5243
- type: nauc_ndcg_at_3_std
value: -14.147000000000002
- type: nauc_ndcg_at_3_diff1
value: -26.011400000000002
- type: nauc_ndcg_at_5_max
value: -16.7965
- type: nauc_ndcg_at_5_std
value: -15.1729
- type: nauc_ndcg_at_5_diff1
value: -27.7871
- type: nauc_ndcg_at_10_max
value: -18.912599999999998
- type: nauc_ndcg_at_10_std
value: -10.5837
- type: nauc_ndcg_at_10_diff1
value: -25.6038
- type: nauc_ndcg_at_20_max
value: -16.9819
- type: nauc_ndcg_at_20_std
value: -6.410100000000001
- type: nauc_ndcg_at_20_diff1
value: -23.090700000000002
- type: nauc_ndcg_at_100_max
value: -17.7062
- type: nauc_ndcg_at_100_std
value: -6.7146
- type: nauc_ndcg_at_100_diff1
value: -20.0496
- type: nauc_ndcg_at_1000_max
value: -17.5259
- type: nauc_ndcg_at_1000_std
value: -8.1273
- type: nauc_ndcg_at_1000_diff1
value: -21.9965
- type: nauc_map_at_1_max
value: -3.7518999999999996
- type: nauc_map_at_1_std
value: -29.5265
- type: nauc_map_at_1_diff1
value: -9.383
- type: nauc_map_at_3_max
value: -10.2362
- type: nauc_map_at_3_std
value: -15.088899999999999
- type: nauc_map_at_3_diff1
value: -23.8832
- type: nauc_map_at_5_max
value: -14.013100000000001
- type: nauc_map_at_5_std
value: -15.710099999999999
- type: nauc_map_at_5_diff1
value: -25.674799999999998
- type: nauc_map_at_10_max
value: -15.9443
- type: nauc_map_at_10_std
value: -12.381300000000001
- type: nauc_map_at_10_diff1
value: -24.6344
- type: nauc_map_at_20_max
value: -15.437899999999999
- type: nauc_map_at_20_std
value: -10.1597
- type: nauc_map_at_20_diff1
value: -23.6569
- type: nauc_map_at_100_max
value: -15.8978
- type: nauc_map_at_100_std
value: -10.050699999999999
- type: nauc_map_at_100_diff1
value: -22.7283
- type: nauc_map_at_1000_max
value: -16.0717
- type: nauc_map_at_1000_std
value: -10.3214
- type: nauc_map_at_1000_diff1
value: -22.8858
- type: nauc_recall_at_1_max
value: -3.7518999999999996
- type: nauc_recall_at_1_std
value: -29.5265
- type: nauc_recall_at_1_diff1
value: -9.383
- type: nauc_recall_at_3_max
value: -16.3357
- type: nauc_recall_at_3_std
value: -12.829099999999999
- type: nauc_recall_at_3_diff1
value: -29.3757
- type: nauc_recall_at_5_max
value: -20.5745
- type: nauc_recall_at_5_std
value: -14.627899999999999
- type: nauc_recall_at_5_diff1
value: -30.521700000000003
- type: nauc_recall_at_10_max
value: -21.7653
- type: nauc_recall_at_10_std
value: -8.8471
- type: nauc_recall_at_10_diff1
value: -26.2943
- type: nauc_recall_at_20_max
value: -17.6809
- type: nauc_recall_at_20_std
value: -3.1351999999999998
- type: nauc_recall_at_20_diff1
value: -22.0324
- type: nauc_recall_at_100_max
value: -18.315
- type: nauc_recall_at_100_std
value: -4.9831
- type: nauc_recall_at_100_diff1
value: -17.8229
- type: nauc_recall_at_1000_max
value: -16.108800000000002
- type: nauc_recall_at_1000_std
value: -6.2484
- type: nauc_recall_at_1000_diff1
value: -22.1741
- type: nauc_precision_at_1_max
value: -3.7518999999999996
- type: nauc_precision_at_1_std
value: -29.5265
- type: nauc_precision_at_1_diff1
value: -9.383
- type: nauc_precision_at_3_max
value: -16.3357
- type: nauc_precision_at_3_std
value: -12.829099999999999
- type: nauc_precision_at_3_diff1
value: -29.3757
- type: nauc_precision_at_5_max
value: -20.5745
- type: nauc_precision_at_5_std
value: -14.627899999999999
- type: nauc_precision_at_5_diff1
value: -30.521700000000003
- type: nauc_precision_at_10_max
value: -21.7653
- type: nauc_precision_at_10_std
value: -8.8471
- type: nauc_precision_at_10_diff1
value: -26.2943
- type: nauc_precision_at_20_max
value: -17.6809
- type: nauc_precision_at_20_std
value: -3.1351999999999998
- type: nauc_precision_at_20_diff1
value: -22.0324
- type: nauc_precision_at_100_max
value: -18.315
- type: nauc_precision_at_100_std
value: -4.9831
- type: nauc_precision_at_100_diff1
value: -17.8229
- type: nauc_precision_at_1000_max
value: -16.253899999999998
- type: nauc_precision_at_1000_std
value: -6.2287
- type: nauc_precision_at_1000_diff1
value: -22.2998
- type: nauc_mrr_at_1_max
value: -3.7518999999999996
- type: nauc_mrr_at_1_std
value: -29.5265
- type: nauc_mrr_at_1_diff1
value: -9.383
- type: nauc_mrr_at_3_max
value: -10.2362
- type: nauc_mrr_at_3_std
value: -15.088899999999999
- type: nauc_mrr_at_3_diff1
value: -23.8832
- type: nauc_mrr_at_5_max
value: -14.013100000000001
- type: nauc_mrr_at_5_std
value: -15.710099999999999
- type: nauc_mrr_at_5_diff1
value: -25.674799999999998
- type: nauc_mrr_at_10_max
value: -15.9443
- type: nauc_mrr_at_10_std
value: -12.381300000000001
- type: nauc_mrr_at_10_diff1
value: -24.6344
- type: nauc_mrr_at_20_max
value: -15.437899999999999
- type: nauc_mrr_at_20_std
value: -10.1597
- type: nauc_mrr_at_20_diff1
value: -23.6569
- type: nauc_mrr_at_100_max
value: -15.8978
- type: nauc_mrr_at_100_std
value: -10.050699999999999
- type: nauc_mrr_at_100_diff1
value: -22.7283
- type: nauc_mrr_at_1000_max
value: -16.074099999999998
- type: nauc_mrr_at_1000_std
value: -10.3209
- type: nauc_mrr_at_1000_diff1
value: -22.8877
- type: main_score
value: 2.275
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-eng)
type: facebook/mlqa
config: ara-eng
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 0.8250000000000001
- type: ndcg_at_3
value: 1.3559999999999999
- type: ndcg_at_5
value: 1.833
- type: ndcg_at_10
value: 2.922
- type: ndcg_at_20
value: 3.943
- type: ndcg_at_100
value: 6.492000000000001
- type: ndcg_at_1000
value: 11.162999999999998
- type: map_at_1
value: 0.8250000000000001
- type: map_at_3
value: 1.222
- type: map_at_5
value: 1.481
- type: map_at_10
value: 1.9220000000000002
- type: map_at_20
value: 2.2009999999999996
- type: map_at_100
value: 2.5180000000000002
- type: map_at_1000
value: 2.654
- type: recall_at_1
value: 0.8250000000000001
- type: recall_at_3
value: 1.744
- type: recall_at_5
value: 2.926
- type: recall_at_10
value: 6.339
- type: recall_at_20
value: 10.39
- type: recall_at_100
value: 24.644
- type: recall_at_1000
value: 63.803
- type: precision_at_1
value: 0.8250000000000001
- type: precision_at_3
value: 0.581
- type: precision_at_5
value: 0.585
- type: precision_at_10
value: 0.634
- type: precision_at_20
value: 0.52
- type: precision_at_100
value: 0.246
- type: precision_at_1000
value: 0.064
- type: mrr_at_1
value: 0.8252
- type: mrr_at_3
value: 1.2222
- type: mrr_at_5
value: 1.481
- type: mrr_at_10
value: 1.9224
- type: mrr_at_20
value: 2.2008
- type: mrr_at_100
value: 2.5183
- type: mrr_at_1000
value: 2.6538
- type: nauc_ndcg_at_1_max
value: 0.9053
- type: nauc_ndcg_at_1_std
value: 34.6374
- type: nauc_ndcg_at_1_diff1
value: 27.330900000000003
- type: nauc_ndcg_at_3_max
value: -5.9703
- type: nauc_ndcg_at_3_std
value: 37.4608
- type: nauc_ndcg_at_3_diff1
value: 16.4823
- type: nauc_ndcg_at_5_max
value: -6.1077
- type: nauc_ndcg_at_5_std
value: 36.6763
- type: nauc_ndcg_at_5_diff1
value: 12.4611
- type: nauc_ndcg_at_10_max
value: -4.5079
- type: nauc_ndcg_at_10_std
value: 27.916400000000003
- type: nauc_ndcg_at_10_diff1
value: 10.6386
- type: nauc_ndcg_at_20_max
value: -2.8867
- type: nauc_ndcg_at_20_std
value: 24.9533
- type: nauc_ndcg_at_20_diff1
value: 8.3649
- type: nauc_ndcg_at_100_max
value: -3.7651999999999997
- type: nauc_ndcg_at_100_std
value: 24.0342
- type: nauc_ndcg_at_100_diff1
value: 7.2088
- type: nauc_ndcg_at_1000_max
value: -2.579
- type: nauc_ndcg_at_1000_std
value: 26.253
- type: nauc_ndcg_at_1000_diff1
value: 7.678699999999999
- type: nauc_map_at_1_max
value: 0.9053
- type: nauc_map_at_1_std
value: 34.6374
- type: nauc_map_at_1_diff1
value: 27.330900000000003
- type: nauc_map_at_3_max
value: -4.6315
- type: nauc_map_at_3_std
value: 36.842999999999996
- type: nauc_map_at_3_diff1
value: 18.601200000000002
- type: nauc_map_at_5_max
value: -5.0622
- type: nauc_map_at_5_std
value: 36.5787
- type: nauc_map_at_5_diff1
value: 15.4748
- type: nauc_map_at_10_max
value: -4.2324
- type: nauc_map_at_10_std
value: 31.355300000000003
- type: nauc_map_at_10_diff1
value: 13.7376
- type: nauc_map_at_20_max
value: -3.4449
- type: nauc_map_at_20_std
value: 29.524299999999997
- type: nauc_map_at_20_diff1
value: 12.3653
- type: nauc_map_at_100_max
value: -3.6995
- type: nauc_map_at_100_std
value: 28.8678
- type: nauc_map_at_100_diff1
value: 11.617700000000001
- type: nauc_map_at_1000_max
value: -3.6461
- type: nauc_map_at_1000_std
value: 29.0105
- type: nauc_map_at_1000_diff1
value: 11.6262
- type: nauc_recall_at_1_max
value: 0.9053
- type: nauc_recall_at_1_std
value: 34.6374
- type: nauc_recall_at_1_diff1
value: 27.330900000000003
- type: nauc_recall_at_3_max
value: -8.7411
- type: nauc_recall_at_3_std
value: 38.7558
- type: nauc_recall_at_3_diff1
value: 12.0955
- type: nauc_recall_at_5_max
value: -7.6163
- type: nauc_recall_at_5_std
value: 36.6908
- type: nauc_recall_at_5_diff1
value: 7.7404
- type: nauc_recall_at_10_max
value: -4.6257
- type: nauc_recall_at_10_std
value: 23.798099999999998
- type: nauc_recall_at_10_diff1
value: 7.5243
- type: nauc_recall_at_20_max
value: -2.182
- type: nauc_recall_at_20_std
value: 20.8335
- type: nauc_recall_at_20_diff1
value: 5.0846
- type: nauc_recall_at_100_max
value: -3.8514
- type: nauc_recall_at_100_std
value: 21.1533
- type: nauc_recall_at_100_diff1
value: 4.826
- type: nauc_recall_at_1000_max
value: -0.5378
- type: nauc_recall_at_1000_std
value: 26.6266
- type: nauc_recall_at_1000_diff1
value: 5.8276
- type: nauc_precision_at_1_max
value: 0.9053
- type: nauc_precision_at_1_std
value: 34.6374
- type: nauc_precision_at_1_diff1
value: 27.330900000000003
- type: nauc_precision_at_3_max
value: -8.7411
- type: nauc_precision_at_3_std
value: 38.7558
- type: nauc_precision_at_3_diff1
value: 12.0955
- type: nauc_precision_at_5_max
value: -7.6163
- type: nauc_precision_at_5_std
value: 36.6908
- type: nauc_precision_at_5_diff1
value: 7.7404
- type: nauc_precision_at_10_max
value: -4.6257
- type: nauc_precision_at_10_std
value: 23.798099999999998
- type: nauc_precision_at_10_diff1
value: 7.5243
- type: nauc_precision_at_20_max
value: -2.182
- type: nauc_precision_at_20_std
value: 20.8335
- type: nauc_precision_at_20_diff1
value: 5.0846
- type: nauc_precision_at_100_max
value: -3.8514
- type: nauc_precision_at_100_std
value: 21.1533
- type: nauc_precision_at_100_diff1
value: 4.826
- type: nauc_precision_at_1000_max
value: -0.5238999999999999
- type: nauc_precision_at_1000_std
value: 26.6614
- type: nauc_precision_at_1000_diff1
value: 5.9221
- type: nauc_mrr_at_1_max
value: 0.9053
- type: nauc_mrr_at_1_std
value: 34.6374
- type: nauc_mrr_at_1_diff1
value: 27.330900000000003
- type: nauc_mrr_at_3_max
value: -4.6315
- type: nauc_mrr_at_3_std
value: 36.842999999999996
- type: nauc_mrr_at_3_diff1
value: 18.601200000000002
- type: nauc_mrr_at_5_max
value: -5.0622
- type: nauc_mrr_at_5_std
value: 36.5787
- type: nauc_mrr_at_5_diff1
value: 15.4748
- type: nauc_mrr_at_10_max
value: -4.2324
- type: nauc_mrr_at_10_std
value: 31.355300000000003
- type: nauc_mrr_at_10_diff1
value: 13.7376
- type: nauc_mrr_at_20_max
value: -3.4449
- type: nauc_mrr_at_20_std
value: 29.524299999999997
- type: nauc_mrr_at_20_diff1
value: 12.3653
- type: nauc_mrr_at_100_max
value: -3.6995
- type: nauc_mrr_at_100_std
value: 28.8678
- type: nauc_mrr_at_100_diff1
value: 11.617700000000001
- type: nauc_mrr_at_1000_max
value: -3.6457
- type: nauc_mrr_at_1000_std
value: 29.010799999999996
- type: nauc_mrr_at_1000_diff1
value: 11.6281
- type: main_score
value: 2.922
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-spa)
type: facebook/mlqa
config: ara-spa
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 0.5559999999999999
- type: ndcg_at_3
value: 1.21
- type: ndcg_at_5
value: 1.504
- type: ndcg_at_10
value: 2.051
- type: ndcg_at_20
value: 2.662
- type: ndcg_at_100
value: 4.553
- type: ndcg_at_1000
value: 11.068999999999999
- type: map_at_1
value: 0.5559999999999999
- type: map_at_3
value: 1.036
- type: map_at_5
value: 1.201
- type: map_at_10
value: 1.421
- type: map_at_20
value: 1.587
- type: map_at_100
value: 1.817
- type: map_at_1000
value: 1.9849999999999999
- type: recall_at_1
value: 0.5559999999999999
- type: recall_at_3
value: 1.719
- type: recall_at_5
value: 2.427
- type: recall_at_10
value: 4.146
- type: recall_at_20
value: 6.572
- type: recall_at_100
value: 17.24
- type: recall_at_1000
value: 73.155
- type: precision_at_1
value: 0.5559999999999999
- type: precision_at_3
value: 0.573
- type: precision_at_5
value: 0.485
- type: precision_at_10
value: 0.415
- type: precision_at_20
value: 0.329
- type: precision_at_100
value: 0.172
- type: precision_at_1000
value: 0.073
- type: mrr_at_1
value: 0.5561
- type: mrr_at_3
value: 1.0364
- type: mrr_at_5
value: 1.2007
- type: mrr_at_10
value: 1.4211
- type: mrr_at_20
value: 1.5872000000000002
- type: mrr_at_100
value: 1.8167
- type: mrr_at_1000
value: 1.9851
- type: nauc_ndcg_at_1_max
value: -29.040100000000002
- type: nauc_ndcg_at_1_std
value: 3.4861999999999997
- type: nauc_ndcg_at_1_diff1
value: 4.853
- type: nauc_ndcg_at_3_max
value: -12.983
- type: nauc_ndcg_at_3_std
value: 1.7259
- type: nauc_ndcg_at_3_diff1
value: 8.4265
- type: nauc_ndcg_at_5_max
value: -10.3764
- type: nauc_ndcg_at_5_std
value: 2.8069
- type: nauc_ndcg_at_5_diff1
value: 14.2088
- type: nauc_ndcg_at_10_max
value: -7.5885
- type: nauc_ndcg_at_10_std
value: 0.9875999999999999
- type: nauc_ndcg_at_10_diff1
value: 14.482800000000001
- type: nauc_ndcg_at_20_max
value: -1.1437
- type: nauc_ndcg_at_20_std
value: 4.1508
- type: nauc_ndcg_at_20_diff1
value: 14.4809
- type: nauc_ndcg_at_100_max
value: -2.751
- type: nauc_ndcg_at_100_std
value: 0.6817
- type: nauc_ndcg_at_100_diff1
value: 12.5662
- type: nauc_ndcg_at_1000_max
value: -0.5488999999999999
- type: nauc_ndcg_at_1000_std
value: 0.3646
- type: nauc_ndcg_at_1000_diff1
value: 11.4795
- type: nauc_map_at_1_max
value: -29.040100000000002
- type: nauc_map_at_1_std
value: 3.4861999999999997
- type: nauc_map_at_1_diff1
value: 4.853
- type: nauc_map_at_3_max
value: -15.252199999999998
- type: nauc_map_at_3_std
value: 1.5733000000000001
- type: nauc_map_at_3_diff1
value: 8.1455
- type: nauc_map_at_5_max
value: -12.8825
- type: nauc_map_at_5_std
value: 2.2918000000000003
- type: nauc_map_at_5_diff1
value: 12.5441
- type: nauc_map_at_10_max
value: -10.509
- type: nauc_map_at_10_std
value: 1.3444
- type: nauc_map_at_10_diff1
value: 13.108600000000001
- type: nauc_map_at_20_max
value: -7.0383000000000004
- type: nauc_map_at_20_std
value: 2.9145999999999996
- type: nauc_map_at_20_diff1
value: 13.2725
- type: nauc_map_at_100_max
value: -6.7613
- type: nauc_map_at_100_std
value: 2.1599
- type: nauc_map_at_100_diff1
value: 12.7128
- type: nauc_map_at_1000_max
value: -6.5134
- type: nauc_map_at_1000_std
value: 1.9965
- type: nauc_map_at_1000_diff1
value: 12.581100000000001
- type: nauc_recall_at_1_max
value: -29.040100000000002
- type: nauc_recall_at_1_std
value: 3.4861999999999997
- type: nauc_recall_at_1_diff1
value: 4.853
- type: nauc_recall_at_3_max
value: -8.9869
- type: nauc_recall_at_3_std
value: 2.086
- type: nauc_recall_at_3_diff1
value: 8.8702
- type: nauc_recall_at_5_max
value: -6.737
- type: nauc_recall_at_5_std
value: 3.7180999999999997
- type: nauc_recall_at_5_diff1
value: 16.743199999999998
- type: nauc_recall_at_10_max
value: -4.5687999999999995
- type: nauc_recall_at_10_std
value: 0.45659999999999995
- type: nauc_recall_at_10_diff1
value: 15.862000000000002
- type: nauc_recall_at_20_max
value: 4.2678
- type: nauc_recall_at_20_std
value: 5.4234
- type: nauc_recall_at_20_diff1
value: 15.3079
- type: nauc_recall_at_100_max
value: -1.4296
- type: nauc_recall_at_100_std
value: -0.9698
- type: nauc_recall_at_100_diff1
value: 12.1166
- type: nauc_recall_at_1000_max
value: 4.0125
- type: nauc_recall_at_1000_std
value: -1.0373
- type: nauc_recall_at_1000_diff1
value: 9.934
- type: nauc_precision_at_1_max
value: -29.040100000000002
- type: nauc_precision_at_1_std
value: 3.4861999999999997
- type: nauc_precision_at_1_diff1
value: 4.853
- type: nauc_precision_at_3_max
value: -8.9869
- type: nauc_precision_at_3_std
value: 2.086
- type: nauc_precision_at_3_diff1
value: 8.8702
- type: nauc_precision_at_5_max
value: -6.737
- type: nauc_precision_at_5_std
value: 3.7180999999999997
- type: nauc_precision_at_5_diff1
value: 16.743199999999998
- type: nauc_precision_at_10_max
value: -4.5687999999999995
- type: nauc_precision_at_10_std
value: 0.45659999999999995
- type: nauc_precision_at_10_diff1
value: 15.862000000000002
- type: nauc_precision_at_20_max
value: 4.2678
- type: nauc_precision_at_20_std
value: 5.4234
- type: nauc_precision_at_20_diff1
value: 15.3079
- type: nauc_precision_at_100_max
value: -1.4296
- type: nauc_precision_at_100_std
value: -0.9698
- type: nauc_precision_at_100_diff1
value: 12.1166
- type: nauc_precision_at_1000_max
value: 4.0125
- type: nauc_precision_at_1000_std
value: -1.0373
- type: nauc_precision_at_1000_diff1
value: 9.934
- type: nauc_mrr_at_1_max
value: -29.040100000000002
- type: nauc_mrr_at_1_std
value: 3.4861999999999997
- type: nauc_mrr_at_1_diff1
value: 4.853
- type: nauc_mrr_at_3_max
value: -15.252199999999998
- type: nauc_mrr_at_3_std
value: 1.5733000000000001
- type: nauc_mrr_at_3_diff1
value: 8.1455
- type: nauc_mrr_at_5_max
value: -12.8825
- type: nauc_mrr_at_5_std
value: 2.2918000000000003
- type: nauc_mrr_at_5_diff1
value: 12.5441
- type: nauc_mrr_at_10_max
value: -10.509
- type: nauc_mrr_at_10_std
value: 1.3444
- type: nauc_mrr_at_10_diff1
value: 13.108600000000001
- type: nauc_mrr_at_20_max
value: -7.0383000000000004
- type: nauc_mrr_at_20_std
value: 2.9145999999999996
- type: nauc_mrr_at_20_diff1
value: 13.2725
- type: nauc_mrr_at_100_max
value: -6.7613
- type: nauc_mrr_at_100_std
value: 2.1599
- type: nauc_mrr_at_100_diff1
value: 12.7128
- type: nauc_mrr_at_1000_max
value: -6.5134
- type: nauc_mrr_at_1000_std
value: 1.9965
- type: nauc_mrr_at_1000_diff1
value: 12.581100000000001
- type: main_score
value: 2.051
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-hin)
type: facebook/mlqa
config: ara-hin
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 0.601
- type: ndcg_at_3
value: 0.889
- type: ndcg_at_5
value: 1.026
- type: ndcg_at_10
value: 1.2409999999999999
- type: ndcg_at_20
value: 1.482
- type: ndcg_at_100
value: 2.6599999999999997
- type: ndcg_at_1000
value: 9.371
- type: map_at_1
value: 0.601
- type: map_at_3
value: 0.819
- type: map_at_5
value: 0.8959999999999999
- type: map_at_10
value: 0.9860000000000001
- type: map_at_20
value: 1.048
- type: map_at_100
value: 1.188
- type: map_at_1000
value: 1.345
- type: recall_at_1
value: 0.601
- type: recall_at_3
value: 1.0919999999999999
- type: recall_at_5
value: 1.4200000000000002
- type: recall_at_10
value: 2.075
- type: recall_at_20
value: 3.058
- type: recall_at_100
value: 9.776
- type: recall_at_1000
value: 68.542
- type: precision_at_1
value: 0.601
- type: precision_at_3
value: 0.364
- type: precision_at_5
value: 0.28400000000000003
- type: precision_at_10
value: 0.208
- type: precision_at_20
value: 0.153
- type: precision_at_100
value: 0.098
- type: precision_at_1000
value: 0.06899999999999999
- type: mrr_at_1
value: 0.6008
- type: mrr_at_3
value: 0.8191999999999999
- type: mrr_at_5
value: 0.8956999999999999
- type: mrr_at_10
value: 0.9862
- type: mrr_at_20
value: 1.0482
- type: mrr_at_100
value: 1.1877
- type: mrr_at_1000
value: 1.3445
- type: nauc_ndcg_at_1_max
value: 77.7698
- type: nauc_ndcg_at_1_std
value: 20.3921
- type: nauc_ndcg_at_1_diff1
value: 78.9992
- type: nauc_ndcg_at_3_max
value: 66.8338
- type: nauc_ndcg_at_3_std
value: 17.974300000000003
- type: nauc_ndcg_at_3_diff1
value: 66.3534
- type: nauc_ndcg_at_5_max
value: 60.3363
- type: nauc_ndcg_at_5_std
value: 15.3865
- type: nauc_ndcg_at_5_diff1
value: 65.0806
- type: nauc_ndcg_at_10_max
value: 48.2563
- type: nauc_ndcg_at_10_std
value: 9.5647
- type: nauc_ndcg_at_10_diff1
value: 53.7428
- type: nauc_ndcg_at_20_max
value: 41.3929
- type: nauc_ndcg_at_20_std
value: 7.0908999999999995
- type: nauc_ndcg_at_20_diff1
value: 47.028999999999996
- type: nauc_ndcg_at_100_max
value: 29.4137
- type: nauc_ndcg_at_100_std
value: 7.297
- type: nauc_ndcg_at_100_diff1
value: 33.575
- type: nauc_ndcg_at_1000_max
value: 21.2503
- type: nauc_ndcg_at_1000_std
value: 5.9479999999999995
- type: nauc_ndcg_at_1000_diff1
value: 21.8539
- type: nauc_map_at_1_max
value: 77.7698
- type: nauc_map_at_1_std
value: 20.3921
- type: nauc_map_at_1_diff1
value: 78.9992
- type: nauc_map_at_3_max
value: 68.6336
- type: nauc_map_at_3_std
value: 18.1845
- type: nauc_map_at_3_diff1
value: 68.3602
- type: nauc_map_at_5_max
value: 64.2857
- type: nauc_map_at_5_std
value: 16.4486
- type: nauc_map_at_5_diff1
value: 67.4023
- type: nauc_map_at_10_max
value: 57.523599999999995
- type: nauc_map_at_10_std
value: 13.2337
- type: nauc_map_at_10_diff1
value: 61.1023
- type: nauc_map_at_20_max
value: 54.5881
- type: nauc_map_at_20_std
value: 12.1576
- type: nauc_map_at_20_diff1
value: 58.4532
- type: nauc_map_at_100_max
value: 49.6122
- type: nauc_map_at_100_std
value: 11.368599999999999
- type: nauc_map_at_100_diff1
value: 53.6787
- type: nauc_map_at_1000_max
value: 47.6843
- type: nauc_map_at_1000_std
value: 10.9958
- type: nauc_map_at_1000_diff1
value: 51.4409
- type: nauc_recall_at_1_max
value: 77.7698
- type: nauc_recall_at_1_std
value: 20.3921
- type: nauc_recall_at_1_diff1
value: 78.9992
- type: nauc_recall_at_3_max
value: 62.9798
- type: nauc_recall_at_3_std
value: 17.5866
- type: nauc_recall_at_3_diff1
value: 62.0812
- type: nauc_recall_at_5_max
value: 52.6436
- type: nauc_recall_at_5_std
value: 13.3293
- type: nauc_recall_at_5_diff1
value: 60.7765
- type: nauc_recall_at_10_max
value: 33.076100000000004
- type: nauc_recall_at_10_std
value: 3.4612
- type: nauc_recall_at_10_diff1
value: 41.6937
- type: nauc_recall_at_20_max
value: 24.080099999999998
- type: nauc_recall_at_20_std
value: 0.41279999999999994
- type: nauc_recall_at_20_diff1
value: 31.678299999999997
- type: nauc_recall_at_100_max
value: 17.8562
- type: nauc_recall_at_100_std
value: 5.8204
- type: nauc_recall_at_100_diff1
value: 21.090600000000002
- type: nauc_recall_at_1000_max
value: 8.8523
- type: nauc_recall_at_1000_std
value: 4.2437000000000005
- type: nauc_recall_at_1000_diff1
value: 5.9054
- type: nauc_precision_at_1_max
value: 77.7698
- type: nauc_precision_at_1_std
value: 20.3921
- type: nauc_precision_at_1_diff1
value: 78.9992
- type: nauc_precision_at_3_max
value: 62.9798
- type: nauc_precision_at_3_std
value: 17.5866
- type: nauc_precision_at_3_diff1
value: 62.0812
- type: nauc_precision_at_5_max
value: 52.6436
- type: nauc_precision_at_5_std
value: 13.3293
- type: nauc_precision_at_5_diff1
value: 60.7765
- type: nauc_precision_at_10_max
value: 33.076100000000004
- type: nauc_precision_at_10_std
value: 3.4612
- type: nauc_precision_at_10_diff1
value: 41.6937
- type: nauc_precision_at_20_max
value: 24.080099999999998
- type: nauc_precision_at_20_std
value: 0.41279999999999994
- type: nauc_precision_at_20_diff1
value: 31.678299999999997
- type: nauc_precision_at_100_max
value: 17.8562
- type: nauc_precision_at_100_std
value: 5.8204
- type: nauc_precision_at_100_diff1
value: 21.090600000000002
- type: nauc_precision_at_1000_max
value: 8.8523
- type: nauc_precision_at_1000_std
value: 4.2437000000000005
- type: nauc_precision_at_1000_diff1
value: 5.9054
- type: nauc_mrr_at_1_max
value: 77.7698
- type: nauc_mrr_at_1_std
value: 20.3921
- type: nauc_mrr_at_1_diff1
value: 78.9992
- type: nauc_mrr_at_3_max
value: 68.6336
- type: nauc_mrr_at_3_std
value: 18.1845
- type: nauc_mrr_at_3_diff1
value: 68.3602
- type: nauc_mrr_at_5_max
value: 64.2857
- type: nauc_mrr_at_5_std
value: 16.4486
- type: nauc_mrr_at_5_diff1
value: 67.4023
- type: nauc_mrr_at_10_max
value: 57.523599999999995
- type: nauc_mrr_at_10_std
value: 13.2337
- type: nauc_mrr_at_10_diff1
value: 61.1023
- type: nauc_mrr_at_20_max
value: 54.5881
- type: nauc_mrr_at_20_std
value: 12.1576
- type: nauc_mrr_at_20_diff1
value: 58.4532
- type: nauc_mrr_at_100_max
value: 49.6122
- type: nauc_mrr_at_100_std
value: 11.368599999999999
- type: nauc_mrr_at_100_diff1
value: 53.6787
- type: nauc_mrr_at_1000_max
value: 47.6843
- type: nauc_mrr_at_1000_std
value: 10.9958
- type: nauc_mrr_at_1000_diff1
value: 51.4409
- type: main_score
value: 1.2409999999999999
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-vie)
type: facebook/mlqa
config: ara-vie
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 1.3679999999999999
- type: ndcg_at_3
value: 2.265
- type: ndcg_at_5
value: 2.624
- type: ndcg_at_10
value: 3.145
- type: ndcg_at_20
value: 3.987
- type: ndcg_at_100
value: 5.968
- type: ndcg_at_1000
value: 11.899999999999999
- type: map_at_1
value: 1.3679999999999999
- type: map_at_3
value: 2.035
- type: map_at_5
value: 2.233
- type: map_at_10
value: 2.448
- type: map_at_20
value: 2.68
- type: map_at_100
value: 2.922
- type: map_at_1000
value: 3.073
- type: recall_at_1
value: 1.3679999999999999
- type: recall_at_3
value: 2.931
- type: recall_at_5
value: 3.81
- type: recall_at_10
value: 5.423
- type: recall_at_20
value: 8.745
- type: recall_at_100
value: 19.883
- type: recall_at_1000
value: 70.982
- type: precision_at_1
value: 1.3679999999999999
- type: precision_at_3
value: 0.9769999999999999
- type: precision_at_5
value: 0.762
- type: precision_at_10
value: 0.542
- type: precision_at_20
value: 0.437
- type: precision_at_100
value: 0.199
- type: precision_at_1000
value: 0.07100000000000001
- type: mrr_at_1
value: 1.3679000000000001
- type: mrr_at_3
value: 2.0355000000000003
- type: mrr_at_5
value: 2.2333
- type: mrr_at_10
value: 2.4479
- type: mrr_at_20
value: 2.6803
- type: mrr_at_100
value: 2.9221
- type: mrr_at_1000
value: 3.0726
- type: nauc_ndcg_at_1_max
value: 52.535900000000005
- type: nauc_ndcg_at_1_std
value: 18.306
- type: nauc_ndcg_at_1_diff1
value: 27.1778
- type: nauc_ndcg_at_3_max
value: 38.7016
- type: nauc_ndcg_at_3_std
value: 22.3974
- type: nauc_ndcg_at_3_diff1
value: 16.9236
- type: nauc_ndcg_at_5_max
value: 37.977
- type: nauc_ndcg_at_5_std
value: 21.3218
- type: nauc_ndcg_at_5_diff1
value: 15.260399999999999
- type: nauc_ndcg_at_10_max
value: 30.9767
- type: nauc_ndcg_at_10_std
value: 17.6847
- type: nauc_ndcg_at_10_diff1
value: 10.74
- type: nauc_ndcg_at_20_max
value: 24.979000000000003
- type: nauc_ndcg_at_20_std
value: 14.299500000000002
- type: nauc_ndcg_at_20_diff1
value: 10.2
- type: nauc_ndcg_at_100_max
value: 23.3543
- type: nauc_ndcg_at_100_std
value: 15.660599999999999
- type: nauc_ndcg_at_100_diff1
value: 9.1841
- type: nauc_ndcg_at_1000_max
value: 21.5855
- type: nauc_ndcg_at_1000_std
value: 12.239
- type: nauc_ndcg_at_1000_diff1
value: 8.6965
- type: nauc_map_at_1_max
value: 52.535900000000005
- type: nauc_map_at_1_std
value: 18.306
- type: nauc_map_at_1_diff1
value: 27.1778
- type: nauc_map_at_3_max
value: 40.8393
- type: nauc_map_at_3_std
value: 21.5482
- type: nauc_map_at_3_diff1
value: 18.6006
- type: nauc_map_at_5_max
value: 40.137
- type: nauc_map_at_5_std
value: 20.856099999999998
- type: nauc_map_at_5_diff1
value: 17.3433
- type: nauc_map_at_10_max
value: 36.3228
- type: nauc_map_at_10_std
value: 18.9674
- type: nauc_map_at_10_diff1
value: 14.8143
- type: nauc_map_at_20_max
value: 33.3903
- type: nauc_map_at_20_std
value: 17.4436
- type: nauc_map_at_20_diff1
value: 14.255799999999999
- type: nauc_map_at_100_max
value: 32.6139
- type: nauc_map_at_100_std
value: 17.6827
- type: nauc_map_at_100_diff1
value: 13.9154
- type: nauc_map_at_1000_max
value: 32.3866
- type: nauc_map_at_1000_std
value: 17.4797
- type: nauc_map_at_1000_diff1
value: 13.8247
- type: nauc_recall_at_1_max
value: 52.535900000000005
- type: nauc_recall_at_1_std
value: 18.306
- type: nauc_recall_at_1_diff1
value: 27.1778
- type: nauc_recall_at_3_max
value: 34.4478
- type: nauc_recall_at_3_std
value: 24.1526
- type: nauc_recall_at_3_diff1
value: 13.5584
- type: nauc_recall_at_5_max
value: 34.316
- type: nauc_recall_at_5_std
value: 22.1098
- type: nauc_recall_at_5_diff1
value: 11.6135
- type: nauc_recall_at_10_max
value: 22.6634
- type: nauc_recall_at_10_std
value: 15.3643
- type: nauc_recall_at_10_diff1
value: 4.4830000000000005
- type: nauc_recall_at_20_max
value: 15.0415
- type: nauc_recall_at_20_std
value: 10.205
- type: nauc_recall_at_20_diff1
value: 5.8558
- type: nauc_recall_at_100_max
value: 16.485
- type: nauc_recall_at_100_std
value: 14.364799999999999
- type: nauc_recall_at_100_diff1
value: 5.6514
- type: nauc_recall_at_1000_max
value: 11.0314
- type: nauc_recall_at_1000_std
value: 3.7834
- type: nauc_recall_at_1000_diff1
value: 4.257099999999999
- type: nauc_precision_at_1_max
value: 52.535900000000005
- type: nauc_precision_at_1_std
value: 18.306
- type: nauc_precision_at_1_diff1
value: 27.1778
- type: nauc_precision_at_3_max
value: 34.4478
- type: nauc_precision_at_3_std
value: 24.1526
- type: nauc_precision_at_3_diff1
value: 13.5584
- type: nauc_precision_at_5_max
value: 34.316
- type: nauc_precision_at_5_std
value: 22.1098
- type: nauc_precision_at_5_diff1
value: 11.6135
- type: nauc_precision_at_10_max
value: 22.6634
- type: nauc_precision_at_10_std
value: 15.3643
- type: nauc_precision_at_10_diff1
value: 4.4830000000000005
- type: nauc_precision_at_20_max
value: 15.0415
- type: nauc_precision_at_20_std
value: 10.205
- type: nauc_precision_at_20_diff1
value: 5.8558
- type: nauc_precision_at_100_max
value: 16.485
- type: nauc_precision_at_100_std
value: 14.364799999999999
- type: nauc_precision_at_100_diff1
value: 5.6514
- type: nauc_precision_at_1000_max
value: 11.0314
- type: nauc_precision_at_1000_std
value: 3.7834
- type: nauc_precision_at_1000_diff1
value: 4.257099999999999
- type: nauc_mrr_at_1_max
value: 52.535900000000005
- type: nauc_mrr_at_1_std
value: 18.306
- type: nauc_mrr_at_1_diff1
value: 27.1778
- type: nauc_mrr_at_3_max
value: 40.8393
- type: nauc_mrr_at_3_std
value: 21.5482
- type: nauc_mrr_at_3_diff1
value: 18.6006
- type: nauc_mrr_at_5_max
value: 40.137
- type: nauc_mrr_at_5_std
value: 20.856099999999998
- type: nauc_mrr_at_5_diff1
value: 17.3433
- type: nauc_mrr_at_10_max
value: 36.3228
- type: nauc_mrr_at_10_std
value: 18.9674
- type: nauc_mrr_at_10_diff1
value: 14.8143
- type: nauc_mrr_at_20_max
value: 33.3903
- type: nauc_mrr_at_20_std
value: 17.4436
- type: nauc_mrr_at_20_diff1
value: 14.255799999999999
- type: nauc_mrr_at_100_max
value: 32.6139
- type: nauc_mrr_at_100_std
value: 17.6827
- type: nauc_mrr_at_100_diff1
value: 13.9154
- type: nauc_mrr_at_1000_max
value: 32.3866
- type: nauc_mrr_at_1000_std
value: 17.4797
- type: nauc_mrr_at_1000_diff1
value: 13.8247
- type: main_score
value: 3.145
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (ara-zho)
type: facebook/mlqa
config: ara-zho
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 0.6799999999999999
- type: ndcg_at_3
value: 1.04
- type: ndcg_at_5
value: 1.106
- type: ndcg_at_10
value: 1.3719999999999999
- type: ndcg_at_20
value: 1.593
- type: ndcg_at_100
value: 2.919
- type: ndcg_at_1000
value: 9.011
- type: map_at_1
value: 0.6799999999999999
- type: map_at_3
value: 0.9329999999999999
- type: map_at_5
value: 0.9690000000000001
- type: map_at_10
value: 1.077
- type: map_at_20
value: 1.1360000000000001
- type: map_at_100
value: 1.287
- type: map_at_1000
value: 1.427
- type: recall_at_1
value: 0.6799999999999999
- type: recall_at_3
value: 1.3599999999999999
- type: recall_at_5
value: 1.517
- type: recall_at_10
value: 2.3539999999999996
- type: recall_at_20
value: 3.243
- type: recall_at_100
value: 10.879
- type: recall_at_1000
value: 64.331
- type: precision_at_1
value: 0.6799999999999999
- type: precision_at_3
value: 0.453
- type: precision_at_5
value: 0.303
- type: precision_at_10
value: 0.23500000000000001
- type: precision_at_20
value: 0.16199999999999998
- type: precision_at_100
value: 0.109
- type: precision_at_1000
value: 0.064
- type: mrr_at_1
value: 0.6799000000000001
- type: mrr_at_3
value: 0.9327
- type: mrr_at_5
value: 0.9693
- type: mrr_at_10
value: 1.0768
- type: mrr_at_20
value: 1.1357000000000002
- type: mrr_at_100
value: 1.2868
- type: mrr_at_1000
value: 1.4273
- type: nauc_ndcg_at_1_max
value: 41.249900000000004
- type: nauc_ndcg_at_1_std
value: -33.319900000000004
- type: nauc_ndcg_at_1_diff1
value: 51.519499999999994
- type: nauc_ndcg_at_3_max
value: 34.7164
- type: nauc_ndcg_at_3_std
value: -21.9086
- type: nauc_ndcg_at_3_diff1
value: 35.729
- type: nauc_ndcg_at_5_max
value: 31.593
- type: nauc_ndcg_at_5_std
value: -22.2105
- type: nauc_ndcg_at_5_diff1
value: 32.5021
- type: nauc_ndcg_at_10_max
value: 22.934099999999997
- type: nauc_ndcg_at_10_std
value: -26.092900000000004
- type: nauc_ndcg_at_10_diff1
value: 30.260199999999998
- type: nauc_ndcg_at_20_max
value: 18.6683
- type: nauc_ndcg_at_20_std
value: -25.922800000000002
- type: nauc_ndcg_at_20_diff1
value: 27.7016
- type: nauc_ndcg_at_100_max
value: 8.9347
- type: nauc_ndcg_at_100_std
value: -18.1861
- type: nauc_ndcg_at_100_diff1
value: 16.4918
- type: nauc_ndcg_at_1000_max
value: 9.234399999999999
- type: nauc_ndcg_at_1000_std
value: -10.485
- type: nauc_ndcg_at_1000_diff1
value: 10.838000000000001
- type: nauc_map_at_1_max
value: 41.249900000000004
- type: nauc_map_at_1_std
value: -33.319900000000004
- type: nauc_map_at_1_diff1
value: 51.519499999999994
- type: nauc_map_at_3_max
value: 36.4384
- type: nauc_map_at_3_std
value: -24.341099999999997
- type: nauc_map_at_3_diff1
value: 39.5864
- type: nauc_map_at_5_max
value: 34.3083
- type: nauc_map_at_5_std
value: -24.5211
- type: nauc_map_at_5_diff1
value: 37.406299999999995
- type: nauc_map_at_10_max
value: 29.3792
- type: nauc_map_at_10_std
value: -26.3575
- type: nauc_map_at_10_diff1
value: 35.702
- type: nauc_map_at_20_max
value: 27.6229
- type: nauc_map_at_20_std
value: -26.238699999999998
- type: nauc_map_at_20_diff1
value: 34.6871
- type: nauc_map_at_100_max
value: 24.1785
- type: nauc_map_at_100_std
value: -24.1922
- type: nauc_map_at_100_diff1
value: 31.005399999999998
- type: nauc_map_at_1000_max
value: 23.3614
- type: nauc_map_at_1000_std
value: -23.3509
- type: nauc_map_at_1000_diff1
value: 29.904700000000002
- type: nauc_recall_at_1_max
value: 41.249900000000004
- type: nauc_recall_at_1_std
value: -33.319900000000004
- type: nauc_recall_at_1_diff1
value: 51.519499999999994
- type: nauc_recall_at_3_max
value: 31.1456
- type: nauc_recall_at_3_std
value: -16.9729
- type: nauc_recall_at_3_diff1
value: 27.7874
- type: nauc_recall_at_5_max
value: 26.2196
- type: nauc_recall_at_5_std
value: -17.8042
- type: nauc_recall_at_5_diff1
value: 22.875799999999998
- type: nauc_recall_at_10_max
value: 12.779399999999999
- type: nauc_recall_at_10_std
value: -26.302300000000002
- type: nauc_recall_at_10_diff1
value: 22.3362
- type: nauc_recall_at_20_max
value: 6.689100000000001
- type: nauc_recall_at_20_std
value: -26.028200000000002
- type: nauc_recall_at_20_diff1
value: 18.8748
- type: nauc_recall_at_100_max
value: -0.3163
- type: nauc_recall_at_100_std
value: -13.942499999999999
- type: nauc_recall_at_100_diff1
value: 7.6121
- type: nauc_recall_at_1000_max
value: 5.161099999999999
- type: nauc_recall_at_1000_std
value: -1.2834999999999999
- type: nauc_recall_at_1000_diff1
value: 1.1552
- type: nauc_precision_at_1_max
value: 41.249900000000004
- type: nauc_precision_at_1_std
value: -33.319900000000004
- type: nauc_precision_at_1_diff1
value: 51.519499999999994
- type: nauc_precision_at_3_max
value: 31.1456
- type: nauc_precision_at_3_std
value: -16.9729
- type: nauc_precision_at_3_diff1
value: 27.7874
- type: nauc_precision_at_5_max
value: 26.2196
- type: nauc_precision_at_5_std
value: -17.8042
- type: nauc_precision_at_5_diff1
value: 22.875799999999998
- type: nauc_precision_at_10_max
value: 12.779399999999999
- type: nauc_precision_at_10_std
value: -26.302300000000002
- type: nauc_precision_at_10_diff1
value: 22.3362
- type: nauc_precision_at_20_max
value: 6.689100000000001
- type: nauc_precision_at_20_std
value: -26.028200000000002
- type: nauc_precision_at_20_diff1
value: 18.8748
- type: nauc_precision_at_100_max
value: -0.3163
- type: nauc_precision_at_100_std
value: -13.942499999999999
- type: nauc_precision_at_100_diff1
value: 7.6121
- type: nauc_precision_at_1000_max
value: 5.161099999999999
- type: nauc_precision_at_1000_std
value: -1.2834999999999999
- type: nauc_precision_at_1000_diff1
value: 1.1552
- type: nauc_mrr_at_1_max
value: 41.249900000000004
- type: nauc_mrr_at_1_std
value: -33.319900000000004
- type: nauc_mrr_at_1_diff1
value: 51.519499999999994
- type: nauc_mrr_at_3_max
value: 36.4384
- type: nauc_mrr_at_3_std
value: -24.341099999999997
- type: nauc_mrr_at_3_diff1
value: 39.5864
- type: nauc_mrr_at_5_max
value: 34.3083
- type: nauc_mrr_at_5_std
value: -24.5211
- type: nauc_mrr_at_5_diff1
value: 37.406299999999995
- type: nauc_mrr_at_10_max
value: 29.3792
- type: nauc_mrr_at_10_std
value: -26.3575
- type: nauc_mrr_at_10_diff1
value: 35.702
- type: nauc_mrr_at_20_max
value: 27.6229
- type: nauc_mrr_at_20_std
value: -26.238699999999998
- type: nauc_mrr_at_20_diff1
value: 34.6871
- type: nauc_mrr_at_100_max
value: 24.1785
- type: nauc_mrr_at_100_std
value: -24.1922
- type: nauc_mrr_at_100_diff1
value: 31.005399999999998
- type: nauc_mrr_at_1000_max
value: 23.3615
- type: nauc_mrr_at_1000_std
value: -23.3509
- type: nauc_mrr_at_1000_diff1
value: 29.904700000000002
- type: main_score
value: 1.3719999999999999
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (deu-ara)
type: facebook/mlqa
config: deu-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 4.002
- type: ndcg_at_3
value: 5.329
- type: ndcg_at_5
value: 6.068
- type: ndcg_at_10
value: 7.2090000000000005
- type: ndcg_at_20
value: 8.128
- type: ndcg_at_100
value: 11.172
- type: ndcg_at_1000
value: 18.029999999999998
- type: map_at_1
value: 4.002
- type: map_at_3
value: 4.993
- type: map_at_5
value: 5.396
- type: map_at_10
value: 5.869
- type: map_at_20
value: 6.121
- type: map_at_100
value: 6.485
- type: map_at_1000
value: 6.678000000000001
- type: recall_at_1
value: 4.002
- type: recall_at_3
value: 6.307
- type: recall_at_5
value: 8.126
- type: recall_at_10
value: 11.643
- type: recall_at_20
value: 15.282000000000002
- type: recall_at_100
value: 32.565
- type: recall_at_1000
value: 90.29700000000001
- type: precision_at_1
value: 4.002
- type: precision_at_3
value: 2.102
- type: precision_at_5
value: 1.625
- type: precision_at_10
value: 1.164
- type: precision_at_20
value: 0.764
- type: precision_at_100
value: 0.326
- type: precision_at_1000
value: 0.09
- type: mrr_at_1
value: 4.0024
- type: mrr_at_3
value: 4.992900000000001
- type: mrr_at_5
value: 5.3962
- type: mrr_at_10
value: 5.869400000000001
- type: mrr_at_20
value: 6.1213999999999995
- type: mrr_at_100
value: 6.4847
- type: mrr_at_1000
value: 6.677700000000001
- type: nauc_ndcg_at_1_max
value: 29.866300000000003
- type: nauc_ndcg_at_1_std
value: 28.7551
- type: nauc_ndcg_at_1_diff1
value: 35.9379
- type: nauc_ndcg_at_3_max
value: 30.2298
- type: nauc_ndcg_at_3_std
value: 26.9338
- type: nauc_ndcg_at_3_diff1
value: 31.617299999999997
- type: nauc_ndcg_at_5_max
value: 30.8693
- type: nauc_ndcg_at_5_std
value: 25.6915
- type: nauc_ndcg_at_5_diff1
value: 31.159799999999997
- type: nauc_ndcg_at_10_max
value: 27.778599999999997
- type: nauc_ndcg_at_10_std
value: 26.418599999999998
- type: nauc_ndcg_at_10_diff1
value: 28.4012
- type: nauc_ndcg_at_20_max
value: 26.2104
- type: nauc_ndcg_at_20_std
value: 25.141599999999997
- type: nauc_ndcg_at_20_diff1
value: 26.9839
- type: nauc_ndcg_at_100_max
value: 26.0935
- type: nauc_ndcg_at_100_std
value: 25.050299999999996
- type: nauc_ndcg_at_100_diff1
value: 23.3752
- type: nauc_ndcg_at_1000_max
value: 26.9319
- type: nauc_ndcg_at_1000_std
value: 24.7647
- type: nauc_ndcg_at_1000_diff1
value: 24.8456
- type: nauc_map_at_1_max
value: 29.866300000000003
- type: nauc_map_at_1_std
value: 28.7551
- type: nauc_map_at_1_diff1
value: 35.9379
- type: nauc_map_at_3_max
value: 30.1102
- type: nauc_map_at_3_std
value: 27.1845
- type: nauc_map_at_3_diff1
value: 32.466499999999996
- type: nauc_map_at_5_max
value: 30.497200000000003
- type: nauc_map_at_5_std
value: 26.3919
- type: nauc_map_at_5_diff1
value: 32.1354
- type: nauc_map_at_10_max
value: 28.938599999999997
- type: nauc_map_at_10_std
value: 26.647100000000002
- type: nauc_map_at_10_diff1
value: 30.680200000000003
- type: nauc_map_at_20_max
value: 28.3236
- type: nauc_map_at_20_std
value: 26.2003
- type: nauc_map_at_20_diff1
value: 30.104599999999998
- type: nauc_map_at_100_max
value: 28.203699999999998
- type: nauc_map_at_100_std
value: 26.063
- type: nauc_map_at_100_diff1
value: 29.361900000000002
- type: nauc_map_at_1000_max
value: 28.2009
- type: nauc_map_at_1000_std
value: 26.002399999999998
- type: nauc_map_at_1000_diff1
value: 29.400100000000002
- type: nauc_recall_at_1_max
value: 29.866300000000003
- type: nauc_recall_at_1_std
value: 28.7551
- type: nauc_recall_at_1_diff1
value: 35.9379
- type: nauc_recall_at_3_max
value: 30.5192
- type: nauc_recall_at_3_std
value: 26.394299999999998
- type: nauc_recall_at_3_diff1
value: 29.672900000000002
- type: nauc_recall_at_5_max
value: 31.6714
- type: nauc_recall_at_5_std
value: 24.2596
- type: nauc_recall_at_5_diff1
value: 29.2296
- type: nauc_recall_at_10_max
value: 25.4894
- type: nauc_recall_at_10_std
value: 26.235999999999997
- type: nauc_recall_at_10_diff1
value: 24.346400000000003
- type: nauc_recall_at_20_max
value: 22.488
- type: nauc_recall_at_20_std
value: 23.3806
- type: nauc_recall_at_20_diff1
value: 21.9467
- type: nauc_recall_at_100_max
value: 23.635900000000003
- type: nauc_recall_at_100_std
value: 24.1875
- type: nauc_recall_at_100_diff1
value: 14.701
- type: nauc_recall_at_1000_max
value: 29.423500000000004
- type: nauc_recall_at_1000_std
value: 22.7087
- type: nauc_recall_at_1000_diff1
value: 10.8994
- type: nauc_precision_at_1_max
value: 29.866300000000003
- type: nauc_precision_at_1_std
value: 28.7551
- type: nauc_precision_at_1_diff1
value: 35.9379
- type: nauc_precision_at_3_max
value: 30.5192
- type: nauc_precision_at_3_std
value: 26.394299999999998
- type: nauc_precision_at_3_diff1
value: 29.672900000000002
- type: nauc_precision_at_5_max
value: 31.6714
- type: nauc_precision_at_5_std
value: 24.2596
- type: nauc_precision_at_5_diff1
value: 29.2296
- type: nauc_precision_at_10_max
value: 25.4894
- type: nauc_precision_at_10_std
value: 26.235999999999997
- type: nauc_precision_at_10_diff1
value: 24.346400000000003
- type: nauc_precision_at_20_max
value: 22.488
- type: nauc_precision_at_20_std
value: 23.3806
- type: nauc_precision_at_20_diff1
value: 21.9467
- type: nauc_precision_at_100_max
value: 23.635900000000003
- type: nauc_precision_at_100_std
value: 24.1875
- type: nauc_precision_at_100_diff1
value: 14.701
- type: nauc_precision_at_1000_max
value: 29.423500000000004
- type: nauc_precision_at_1000_std
value: 22.7087
- type: nauc_precision_at_1000_diff1
value: 10.8994
- type: nauc_mrr_at_1_max
value: 29.866300000000003
- type: nauc_mrr_at_1_std
value: 28.7551
- type: nauc_mrr_at_1_diff1
value: 35.9379
- type: nauc_mrr_at_3_max
value: 30.1102
- type: nauc_mrr_at_3_std
value: 27.1845
- type: nauc_mrr_at_3_diff1
value: 32.466499999999996
- type: nauc_mrr_at_5_max
value: 30.497200000000003
- type: nauc_mrr_at_5_std
value: 26.3919
- type: nauc_mrr_at_5_diff1
value: 32.1354
- type: nauc_mrr_at_10_max
value: 28.938599999999997
- type: nauc_mrr_at_10_std
value: 26.647100000000002
- type: nauc_mrr_at_10_diff1
value: 30.680200000000003
- type: nauc_mrr_at_20_max
value: 28.3236
- type: nauc_mrr_at_20_std
value: 26.2003
- type: nauc_mrr_at_20_diff1
value: 30.104599999999998
- type: nauc_mrr_at_100_max
value: 28.203699999999998
- type: nauc_mrr_at_100_std
value: 26.063
- type: nauc_mrr_at_100_diff1
value: 29.361900000000002
- type: nauc_mrr_at_1000_max
value: 28.2009
- type: nauc_mrr_at_1000_std
value: 26.002399999999998
- type: nauc_mrr_at_1000_diff1
value: 29.400100000000002
- type: main_score
value: 7.2090000000000005
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (eng-ara)
type: facebook/mlqa
config: eng-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 4.857
- type: ndcg_at_3
value: 7.247000000000001
- type: ndcg_at_5
value: 8.391
- type: ndcg_at_10
value: 9.808
- type: ndcg_at_20
value: 11.392
- type: ndcg_at_100
value: 15.203
- type: ndcg_at_1000
value: 19.99
- type: map_at_1
value: 4.857
- type: map_at_3
value: 6.633
- type: map_at_5
value: 7.269
- type: map_at_10
value: 7.845000000000001
- type: map_at_20
value: 8.28
- type: map_at_100
value: 8.763
- type: map_at_1000
value: 8.911
- type: recall_at_1
value: 4.857
- type: recall_at_3
value: 9.029
- type: recall_at_5
value: 11.804
- type: recall_at_10
value: 16.229
- type: recall_at_20
value: 22.492
- type: recall_at_100
value: 43.69
- type: recall_at_1000
value: 83.19
- type: precision_at_1
value: 4.857
- type: precision_at_3
value: 3.013
- type: precision_at_5
value: 2.363
- type: precision_at_10
value: 1.624
- type: precision_at_20
value: 1.125
- type: precision_at_100
value: 0.437
- type: precision_at_1000
value: 0.083
- type: mrr_at_1
value: 4.8566
- type: mrr_at_3
value: 6.637899999999999
- type: mrr_at_5
value: 7.273599999999999
- type: mrr_at_10
value: 7.8496
- type: mrr_at_20
value: 8.2844
- type: mrr_at_100
value: 8.7671
- type: mrr_at_1000
value: 8.9155
- type: nauc_ndcg_at_1_max
value: 30.738100000000003
- type: nauc_ndcg_at_1_std
value: 23.4738
- type: nauc_ndcg_at_1_diff1
value: 29.6428
- type: nauc_ndcg_at_3_max
value: 25.063299999999998
- type: nauc_ndcg_at_3_std
value: 23.311899999999998
- type: nauc_ndcg_at_3_diff1
value: 20.8211
- type: nauc_ndcg_at_5_max
value: 23.3085
- type: nauc_ndcg_at_5_std
value: 23.5156
- type: nauc_ndcg_at_5_diff1
value: 17.7465
- type: nauc_ndcg_at_10_max
value: 21.992
- type: nauc_ndcg_at_10_std
value: 23.742
- type: nauc_ndcg_at_10_diff1
value: 16.4182
- type: nauc_ndcg_at_20_max
value: 21.343999999999998
- type: nauc_ndcg_at_20_std
value: 23.8546
- type: nauc_ndcg_at_20_diff1
value: 14.791699999999999
- type: nauc_ndcg_at_100_max
value: 20.0127
- type: nauc_ndcg_at_100_std
value: 25.2797
- type: nauc_ndcg_at_100_diff1
value: 14.0799
- type: nauc_ndcg_at_1000_max
value: 21.2727
- type: nauc_ndcg_at_1000_std
value: 25.2949
- type: nauc_ndcg_at_1000_diff1
value: 14.6762
- type: nauc_map_at_1_max
value: 30.738100000000003
- type: nauc_map_at_1_std
value: 23.4738
- type: nauc_map_at_1_diff1
value: 29.6428
- type: nauc_map_at_3_max
value: 26.267200000000003
- type: nauc_map_at_3_std
value: 23.302400000000002
- type: nauc_map_at_3_diff1
value: 22.665499999999998
- type: nauc_map_at_5_max
value: 25.0361
- type: nauc_map_at_5_std
value: 23.4055
- type: nauc_map_at_5_diff1
value: 20.5664
- type: nauc_map_at_10_max
value: 24.3108
- type: nauc_map_at_10_std
value: 23.56
- type: nauc_map_at_10_diff1
value: 19.7728
- type: nauc_map_at_20_max
value: 24.0046
- type: nauc_map_at_20_std
value: 23.6389
- type: nauc_map_at_20_diff1
value: 19.0906
- type: nauc_map_at_100_max
value: 23.7818
- type: nauc_map_at_100_std
value: 23.8873
- type: nauc_map_at_100_diff1
value: 18.9038
- type: nauc_map_at_1000_max
value: 23.846700000000002
- type: nauc_map_at_1000_std
value: 23.8945
- type: nauc_map_at_1000_diff1
value: 18.955
- type: nauc_recall_at_1_max
value: 30.738100000000003
- type: nauc_recall_at_1_std
value: 23.4738
- type: nauc_recall_at_1_diff1
value: 29.6428
- type: nauc_recall_at_3_max
value: 22.4695
- type: nauc_recall_at_3_std
value: 23.352
- type: nauc_recall_at_3_diff1
value: 16.8167
- type: nauc_recall_at_5_max
value: 19.9589
- type: nauc_recall_at_5_std
value: 23.7703
- type: nauc_recall_at_5_diff1
value: 12.213000000000001
- type: nauc_recall_at_10_max
value: 17.985300000000002
- type: nauc_recall_at_10_std
value: 24.0633
- type: nauc_recall_at_10_diff1
value: 10.6866
- type: nauc_recall_at_20_max
value: 17.3067
- type: nauc_recall_at_20_std
value: 24.1389
- type: nauc_recall_at_20_diff1
value: 8.123800000000001
- type: nauc_recall_at_100_max
value: 13.9575
- type: nauc_recall_at_100_std
value: 28.151300000000003
- type: nauc_recall_at_100_diff1
value: 7.1502
- type: nauc_recall_at_1000_max
value: 16.669800000000002
- type: nauc_recall_at_1000_std
value: 31.237
- type: nauc_recall_at_1000_diff1
value: 3.0153
- type: nauc_precision_at_1_max
value: 30.738100000000003
- type: nauc_precision_at_1_std
value: 23.4738
- type: nauc_precision_at_1_diff1
value: 29.6428
- type: nauc_precision_at_3_max
value: 22.4388
- type: nauc_precision_at_3_std
value: 23.338
- type: nauc_precision_at_3_diff1
value: 16.8328
- type: nauc_precision_at_5_max
value: 19.9419
- type: nauc_precision_at_5_std
value: 23.7654
- type: nauc_precision_at_5_diff1
value: 12.2334
- type: nauc_precision_at_10_max
value: 17.9727
- type: nauc_precision_at_10_std
value: 24.0593
- type: nauc_precision_at_10_diff1
value: 10.7034
- type: nauc_precision_at_20_max
value: 17.2999
- type: nauc_precision_at_20_std
value: 24.14
- type: nauc_precision_at_20_diff1
value: 8.1398
- type: nauc_precision_at_100_max
value: 13.938400000000001
- type: nauc_precision_at_100_std
value: 28.134700000000002
- type: nauc_precision_at_100_diff1
value: 7.1732000000000005
- type: nauc_precision_at_1000_max
value: 16.622600000000002
- type: nauc_precision_at_1000_std
value: 31.1766
- type: nauc_precision_at_1000_diff1
value: 3.087
- type: nauc_mrr_at_1_max
value: 30.738100000000003
- type: nauc_mrr_at_1_std
value: 23.4738
- type: nauc_mrr_at_1_diff1
value: 29.6428
- type: nauc_mrr_at_3_max
value: 26.243699999999997
- type: nauc_mrr_at_3_std
value: 23.2929
- type: nauc_mrr_at_3_diff1
value: 22.6723
- type: nauc_mrr_at_5_max
value: 25.0151
- type: nauc_mrr_at_5_std
value: 23.3966
- type: nauc_mrr_at_5_diff1
value: 20.5742
- type: nauc_mrr_at_10_max
value: 24.2912
- type: nauc_mrr_at_10_std
value: 23.5515
- type: nauc_mrr_at_10_diff1
value: 19.7807
- type: nauc_mrr_at_20_max
value: 23.985899999999997
- type: nauc_mrr_at_20_std
value: 23.630599999999998
- type: nauc_mrr_at_20_diff1
value: 19.098599999999998
- type: nauc_mrr_at_100_max
value: 23.7648
- type: nauc_mrr_at_100_std
value: 23.8796
- type: nauc_mrr_at_100_diff1
value: 18.9113
- type: nauc_mrr_at_1000_max
value: 23.8295
- type: nauc_mrr_at_1000_std
value: 23.8864
- type: nauc_mrr_at_1000_diff1
value: 18.9626
- type: main_score
value: 9.808
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (spa-ara)
type: facebook/mlqa
config: spa-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 2.275
- type: ndcg_at_3
value: 3.961
- type: ndcg_at_5
value: 4.55
- type: ndcg_at_10
value: 5.316
- type: ndcg_at_20
value: 6.457
- type: ndcg_at_100
value: 9.857000000000001
- type: ndcg_at_1000
value: 16.057
- type: map_at_1
value: 2.275
- type: map_at_3
value: 3.547
- type: map_at_5
value: 3.866
- type: map_at_10
value: 4.170999999999999
- type: map_at_20
value: 4.486
- type: map_at_100
value: 4.907
- type: map_at_1000
value: 5.086
- type: recall_at_1
value: 2.275
- type: recall_at_3
value: 5.157
- type: recall_at_5
value: 6.622999999999999
- type: recall_at_10
value: 9.049999999999999
- type: recall_at_20
value: 13.549
- type: recall_at_100
value: 32.609
- type: recall_at_1000
value: 84.631
- type: precision_at_1
value: 2.275
- type: precision_at_3
value: 1.719
- type: precision_at_5
value: 1.325
- type: precision_at_10
value: 0.905
- type: precision_at_20
value: 0.677
- type: precision_at_100
value: 0.326
- type: precision_at_1000
value: 0.08499999999999999
- type: mrr_at_1
value: 2.275
- type: mrr_at_3
value: 3.5473999999999997
- type: mrr_at_5
value: 3.8659
- type: mrr_at_10
value: 4.1711
- type: mrr_at_20
value: 4.4859
- type: mrr_at_100
value: 4.9069
- type: mrr_at_1000
value: 5.0863
- type: nauc_ndcg_at_1_max
value: 42.763
- type: nauc_ndcg_at_1_std
value: 26.793400000000002
- type: nauc_ndcg_at_1_diff1
value: 32.359100000000005
- type: nauc_ndcg_at_3_max
value: 32.7598
- type: nauc_ndcg_at_3_std
value: 31.3869
- type: nauc_ndcg_at_3_diff1
value: 22.9771
- type: nauc_ndcg_at_5_max
value: 29.557899999999997
- type: nauc_ndcg_at_5_std
value: 29.2269
- type: nauc_ndcg_at_5_diff1
value: 20.508499999999998
- type: nauc_ndcg_at_10_max
value: 25.771699999999996
- type: nauc_ndcg_at_10_std
value: 27.260099999999998
- type: nauc_ndcg_at_10_diff1
value: 18.2208
- type: nauc_ndcg_at_20_max
value: 24.7409
- type: nauc_ndcg_at_20_std
value: 26.6067
- type: nauc_ndcg_at_20_diff1
value: 17.3434
- type: nauc_ndcg_at_100_max
value: 23.070899999999998
- type: nauc_ndcg_at_100_std
value: 27.9696
- type: nauc_ndcg_at_100_diff1
value: 13.425500000000001
- type: nauc_ndcg_at_1000_max
value: 23.4468
- type: nauc_ndcg_at_1000_std
value: 27.359
- type: nauc_ndcg_at_1000_diff1
value: 15.1178
- type: nauc_map_at_1_max
value: 42.763
- type: nauc_map_at_1_std
value: 26.793400000000002
- type: nauc_map_at_1_diff1
value: 32.359100000000005
- type: nauc_map_at_3_max
value: 34.5133
- type: nauc_map_at_3_std
value: 30.6626
- type: nauc_map_at_3_diff1
value: 24.3931
- type: nauc_map_at_5_max
value: 32.303
- type: nauc_map_at_5_std
value: 29.4094
- type: nauc_map_at_5_diff1
value: 22.6904
- type: nauc_map_at_10_max
value: 30.213600000000003
- type: nauc_map_at_10_std
value: 28.3638
- type: nauc_map_at_10_diff1
value: 21.277099999999997
- type: nauc_map_at_20_max
value: 29.530299999999997
- type: nauc_map_at_20_std
value: 28.016999999999996
- type: nauc_map_at_20_diff1
value: 20.758
- type: nauc_map_at_100_max
value: 28.8051
- type: nauc_map_at_100_std
value: 28.262700000000002
- type: nauc_map_at_100_diff1
value: 19.7487
- type: nauc_map_at_1000_max
value: 28.7919
- type: nauc_map_at_1000_std
value: 28.2294
- type: nauc_map_at_1000_diff1
value: 19.7847
- type: nauc_recall_at_1_max
value: 42.763
- type: nauc_recall_at_1_std
value: 26.793400000000002
- type: nauc_recall_at_1_diff1
value: 32.359100000000005
- type: nauc_recall_at_3_max
value: 29.2199
- type: nauc_recall_at_3_std
value: 32.8289
- type: nauc_recall_at_3_diff1
value: 20.176099999999998
- type: nauc_recall_at_5_max
value: 24.6016
- type: nauc_recall_at_5_std
value: 28.669800000000002
- type: nauc_recall_at_5_diff1
value: 16.615
- type: nauc_recall_at_10_max
value: 18.805
- type: nauc_recall_at_10_std
value: 25.247700000000002
- type: nauc_recall_at_10_diff1
value: 13.631699999999999
- type: nauc_recall_at_20_max
value: 18.753
- type: nauc_recall_at_20_std
value: 24.5916
- type: nauc_recall_at_20_diff1
value: 13.1638
- type: nauc_recall_at_100_max
value: 18.0435
- type: nauc_recall_at_100_std
value: 28.1351
- type: nauc_recall_at_100_diff1
value: 6.680400000000001
- type: nauc_recall_at_1000_max
value: 15.244
- type: nauc_recall_at_1000_std
value: 24.7548
- type: nauc_recall_at_1000_diff1
value: 9.8426
- type: nauc_precision_at_1_max
value: 42.763
- type: nauc_precision_at_1_std
value: 26.793400000000002
- type: nauc_precision_at_1_diff1
value: 32.359100000000005
- type: nauc_precision_at_3_max
value: 29.2199
- type: nauc_precision_at_3_std
value: 32.8289
- type: nauc_precision_at_3_diff1
value: 20.176099999999998
- type: nauc_precision_at_5_max
value: 24.6016
- type: nauc_precision_at_5_std
value: 28.669800000000002
- type: nauc_precision_at_5_diff1
value: 16.615
- type: nauc_precision_at_10_max
value: 18.805
- type: nauc_precision_at_10_std
value: 25.247700000000002
- type: nauc_precision_at_10_diff1
value: 13.631699999999999
- type: nauc_precision_at_20_max
value: 18.753
- type: nauc_precision_at_20_std
value: 24.5916
- type: nauc_precision_at_20_diff1
value: 13.1638
- type: nauc_precision_at_100_max
value: 18.0435
- type: nauc_precision_at_100_std
value: 28.1351
- type: nauc_precision_at_100_diff1
value: 6.680400000000001
- type: nauc_precision_at_1000_max
value: 15.244
- type: nauc_precision_at_1000_std
value: 24.7548
- type: nauc_precision_at_1000_diff1
value: 9.8426
- type: nauc_mrr_at_1_max
value: 42.763
- type: nauc_mrr_at_1_std
value: 26.793400000000002
- type: nauc_mrr_at_1_diff1
value: 32.359100000000005
- type: nauc_mrr_at_3_max
value: 34.5133
- type: nauc_mrr_at_3_std
value: 30.6626
- type: nauc_mrr_at_3_diff1
value: 24.3931
- type: nauc_mrr_at_5_max
value: 32.303
- type: nauc_mrr_at_5_std
value: 29.4094
- type: nauc_mrr_at_5_diff1
value: 22.6904
- type: nauc_mrr_at_10_max
value: 30.213600000000003
- type: nauc_mrr_at_10_std
value: 28.3638
- type: nauc_mrr_at_10_diff1
value: 21.277099999999997
- type: nauc_mrr_at_20_max
value: 29.530299999999997
- type: nauc_mrr_at_20_std
value: 28.016999999999996
- type: nauc_mrr_at_20_diff1
value: 20.758
- type: nauc_mrr_at_100_max
value: 28.8051
- type: nauc_mrr_at_100_std
value: 28.262700000000002
- type: nauc_mrr_at_100_diff1
value: 19.7487
- type: nauc_mrr_at_1000_max
value: 28.7919
- type: nauc_mrr_at_1000_std
value: 28.2294
- type: nauc_mrr_at_1000_diff1
value: 19.7847
- type: main_score
value: 5.316
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (hin-ara)
type: facebook/mlqa
config: hin-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 3.113
- type: ndcg_at_3
value: 4.199
- type: ndcg_at_5
value: 4.622
- type: ndcg_at_10
value: 5.2780000000000005
- type: ndcg_at_20
value: 5.6259999999999994
- type: ndcg_at_100
value: 7.430000000000001
- type: ndcg_at_1000
value: 14.321
- type: map_at_1
value: 3.113
- type: map_at_3
value: 3.932
- type: map_at_5
value: 4.164000000000001
- type: map_at_10
value: 4.437
- type: map_at_20
value: 4.534
- type: map_at_100
value: 4.756
- type: map_at_1000
value: 4.925
- type: recall_at_1
value: 3.113
- type: recall_at_3
value: 4.97
- type: recall_at_5
value: 6.008
- type: recall_at_10
value: 8.028
- type: recall_at_20
value: 9.394
- type: recall_at_100
value: 19.552
- type: recall_at_1000
value: 79.35600000000001
- type: precision_at_1
value: 3.113
- type: precision_at_3
value: 1.657
- type: precision_at_5
value: 1.202
- type: precision_at_10
value: 0.803
- type: precision_at_20
value: 0.47000000000000003
- type: precision_at_100
value: 0.196
- type: precision_at_1000
value: 0.079
- type: mrr_at_1
value: 3.1130999999999998
- type: mrr_at_3
value: 3.9322999999999997
- type: mrr_at_5
value: 4.1644
- type: mrr_at_10
value: 4.4371
- type: mrr_at_20
value: 4.5343
- type: mrr_at_100
value: 4.7557
- type: mrr_at_1000
value: 4.9247
- type: nauc_ndcg_at_1_max
value: 38.3461
- type: nauc_ndcg_at_1_std
value: 42.357099999999996
- type: nauc_ndcg_at_1_diff1
value: 45.6064
- type: nauc_ndcg_at_3_max
value: 31.1164
- type: nauc_ndcg_at_3_std
value: 36.978
- type: nauc_ndcg_at_3_diff1
value: 33.0373
- type: nauc_ndcg_at_5_max
value: 27.4854
- type: nauc_ndcg_at_5_std
value: 36.381
- type: nauc_ndcg_at_5_diff1
value: 28.9872
- type: nauc_ndcg_at_10_max
value: 25.1205
- type: nauc_ndcg_at_10_std
value: 36.1055
- type: nauc_ndcg_at_10_diff1
value: 27.8873
- type: nauc_ndcg_at_20_max
value: 24.1398
- type: nauc_ndcg_at_20_std
value: 34.0479
- type: nauc_ndcg_at_20_diff1
value: 25.171
- type: nauc_ndcg_at_100_max
value: 19.453
- type: nauc_ndcg_at_100_std
value: 29.2945
- type: nauc_ndcg_at_100_diff1
value: 19.8794
- type: nauc_ndcg_at_1000_max
value: 18.9865
- type: nauc_ndcg_at_1000_std
value: 27.2695
- type: nauc_ndcg_at_1000_diff1
value: 19.7427
- type: nauc_map_at_1_max
value: 38.3461
- type: nauc_map_at_1_std
value: 42.357099999999996
- type: nauc_map_at_1_diff1
value: 45.6064
- type: nauc_map_at_3_max
value: 32.466699999999996
- type: nauc_map_at_3_std
value: 38.0248
- type: nauc_map_at_3_diff1
value: 35.416399999999996
- type: nauc_map_at_5_max
value: 30.189
- type: nauc_map_at_5_std
value: 37.5654
- type: nauc_map_at_5_diff1
value: 32.8839
- type: nauc_map_at_10_max
value: 28.842200000000002
- type: nauc_map_at_10_std
value: 37.3428
- type: nauc_map_at_10_diff1
value: 32.066
- type: nauc_map_at_20_max
value: 28.4441
- type: nauc_map_at_20_std
value: 36.6104
- type: nauc_map_at_20_diff1
value: 31.069000000000003
- type: nauc_map_at_100_max
value: 27.4914
- type: nauc_map_at_100_std
value: 35.6224
- type: nauc_map_at_100_diff1
value: 29.9003
- type: nauc_map_at_1000_max
value: 27.268700000000003
- type: nauc_map_at_1000_std
value: 35.438199999999995
- type: nauc_map_at_1000_diff1
value: 29.7381
- type: nauc_recall_at_1_max
value: 38.3461
- type: nauc_recall_at_1_std
value: 42.357099999999996
- type: nauc_recall_at_1_diff1
value: 45.6064
- type: nauc_recall_at_3_max
value: 28.0433
- type: nauc_recall_at_3_std
value: 34.5815
- type: nauc_recall_at_3_diff1
value: 27.6117
- type: nauc_recall_at_5_max
value: 21.695
- type: nauc_recall_at_5_std
value: 33.976099999999995
- type: nauc_recall_at_5_diff1
value: 20.7131
- type: nauc_recall_at_10_max
value: 18.3982
- type: nauc_recall_at_10_std
value: 34.071
- type: nauc_recall_at_10_diff1
value: 20.6696
- type: nauc_recall_at_20_max
value: 16.9984
- type: nauc_recall_at_20_std
value: 29.505
- type: nauc_recall_at_20_diff1
value: 15.207999999999998
- type: nauc_recall_at_100_max
value: 8.7388
- type: nauc_recall_at_100_std
value: 20.3546
- type: nauc_recall_at_100_diff1
value: 7.0043999999999995
- type: nauc_recall_at_1000_max
value: 6.571000000000001
- type: nauc_recall_at_1000_std
value: 8.7357
- type: nauc_recall_at_1000_diff1
value: 3.8280000000000003
- type: nauc_precision_at_1_max
value: 38.3461
- type: nauc_precision_at_1_std
value: 42.357099999999996
- type: nauc_precision_at_1_diff1
value: 45.6064
- type: nauc_precision_at_3_max
value: 28.0433
- type: nauc_precision_at_3_std
value: 34.5815
- type: nauc_precision_at_3_diff1
value: 27.6117
- type: nauc_precision_at_5_max
value: 21.695
- type: nauc_precision_at_5_std
value: 33.976099999999995
- type: nauc_precision_at_5_diff1
value: 20.7131
- type: nauc_precision_at_10_max
value: 18.3982
- type: nauc_precision_at_10_std
value: 34.071
- type: nauc_precision_at_10_diff1
value: 20.6696
- type: nauc_precision_at_20_max
value: 16.9984
- type: nauc_precision_at_20_std
value: 29.505
- type: nauc_precision_at_20_diff1
value: 15.207999999999998
- type: nauc_precision_at_100_max
value: 8.7388
- type: nauc_precision_at_100_std
value: 20.3546
- type: nauc_precision_at_100_diff1
value: 7.0043999999999995
- type: nauc_precision_at_1000_max
value: 6.571000000000001
- type: nauc_precision_at_1000_std
value: 8.7357
- type: nauc_precision_at_1000_diff1
value: 3.8280000000000003
- type: nauc_mrr_at_1_max
value: 38.3461
- type: nauc_mrr_at_1_std
value: 42.357099999999996
- type: nauc_mrr_at_1_diff1
value: 45.6064
- type: nauc_mrr_at_3_max
value: 32.466699999999996
- type: nauc_mrr_at_3_std
value: 38.0248
- type: nauc_mrr_at_3_diff1
value: 35.416399999999996
- type: nauc_mrr_at_5_max
value: 30.189
- type: nauc_mrr_at_5_std
value: 37.5654
- type: nauc_mrr_at_5_diff1
value: 32.8839
- type: nauc_mrr_at_10_max
value: 28.842200000000002
- type: nauc_mrr_at_10_std
value: 37.3428
- type: nauc_mrr_at_10_diff1
value: 32.066
- type: nauc_mrr_at_20_max
value: 28.4441
- type: nauc_mrr_at_20_std
value: 36.6104
- type: nauc_mrr_at_20_diff1
value: 31.069000000000003
- type: nauc_mrr_at_100_max
value: 27.4914
- type: nauc_mrr_at_100_std
value: 35.6224
- type: nauc_mrr_at_100_diff1
value: 29.9003
- type: nauc_mrr_at_1000_max
value: 27.268700000000003
- type: nauc_mrr_at_1000_std
value: 35.438199999999995
- type: nauc_mrr_at_1000_diff1
value: 29.7381
- type: main_score
value: 5.2780000000000005
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (vie-ara)
type: facebook/mlqa
config: vie-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 2.785
- type: ndcg_at_3
value: 4.376
- type: ndcg_at_5
value: 5.116
- type: ndcg_at_10
value: 6.275
- type: ndcg_at_20
value: 7.585
- type: ndcg_at_100
value: 10.374
- type: ndcg_at_1000
value: 16.346
- type: map_at_1
value: 2.785
- type: map_at_3
value: 3.981
- type: map_at_5
value: 4.389
- type: map_at_10
value: 4.871
- type: map_at_20
value: 5.224
- type: map_at_100
value: 5.561
- type: map_at_1000
value: 5.723000000000001
- type: recall_at_1
value: 2.785
- type: recall_at_3
value: 5.52
- type: recall_at_5
value: 7.327999999999999
- type: recall_at_10
value: 10.894
- type: recall_at_20
value: 16.121
- type: recall_at_100
value: 31.900000000000002
- type: recall_at_1000
value: 82.609
- type: precision_at_1
value: 2.785
- type: precision_at_3
value: 1.8399999999999999
- type: precision_at_5
value: 1.466
- type: precision_at_10
value: 1.089
- type: precision_at_20
value: 0.8059999999999999
- type: precision_at_100
value: 0.319
- type: precision_at_1000
value: 0.083
- type: mrr_at_1
value: 2.7845999999999997
- type: mrr_at_3
value: 3.9814000000000003
- type: mrr_at_5
value: 4.3894
- type: mrr_at_10
value: 4.8708
- type: mrr_at_20
value: 5.2244
- type: mrr_at_100
value: 5.5607999999999995
- type: mrr_at_1000
value: 5.7233
- type: nauc_ndcg_at_1_max
value: 49.0499
- type: nauc_ndcg_at_1_std
value: 38.6812
- type: nauc_ndcg_at_1_diff1
value: 52.2489
- type: nauc_ndcg_at_3_max
value: 40.9962
- type: nauc_ndcg_at_3_std
value: 33.514500000000005
- type: nauc_ndcg_at_3_diff1
value: 34.2081
- type: nauc_ndcg_at_5_max
value: 38.2688
- type: nauc_ndcg_at_5_std
value: 32.745000000000005
- type: nauc_ndcg_at_5_diff1
value: 30.5589
- type: nauc_ndcg_at_10_max
value: 34.7962
- type: nauc_ndcg_at_10_std
value: 30.3547
- type: nauc_ndcg_at_10_diff1
value: 26.0212
- type: nauc_ndcg_at_20_max
value: 32.932
- type: nauc_ndcg_at_20_std
value: 29.4971
- type: nauc_ndcg_at_20_diff1
value: 22.8512
- type: nauc_ndcg_at_100_max
value: 30.3474
- type: nauc_ndcg_at_100_std
value: 28.380499999999998
- type: nauc_ndcg_at_100_diff1
value: 20.9232
- type: nauc_ndcg_at_1000_max
value: 32.407399999999996
- type: nauc_ndcg_at_1000_std
value: 31.176199999999998
- type: nauc_ndcg_at_1000_diff1
value: 22.3578
- type: nauc_map_at_1_max
value: 49.0499
- type: nauc_map_at_1_std
value: 38.6812
- type: nauc_map_at_1_diff1
value: 52.2489
- type: nauc_map_at_3_max
value: 42.479499999999994
- type: nauc_map_at_3_std
value: 34.5065
- type: nauc_map_at_3_diff1
value: 37.5021
- type: nauc_map_at_5_max
value: 40.6623
- type: nauc_map_at_5_std
value: 34.0191
- type: nauc_map_at_5_diff1
value: 34.8592
- type: nauc_map_at_10_max
value: 38.600899999999996
- type: nauc_map_at_10_std
value: 32.5849
- type: nauc_map_at_10_diff1
value: 32.1012
- type: nauc_map_at_20_max
value: 37.6983
- type: nauc_map_at_20_std
value: 32.2239
- type: nauc_map_at_20_diff1
value: 30.6472
- type: nauc_map_at_100_max
value: 37.0514
- type: nauc_map_at_100_std
value: 31.941000000000003
- type: nauc_map_at_100_diff1
value: 29.9615
- type: nauc_map_at_1000_max
value: 37.1014
- type: nauc_map_at_1000_std
value: 32.0581
- type: nauc_map_at_1000_diff1
value: 30.025000000000002
- type: nauc_recall_at_1_max
value: 49.0499
- type: nauc_recall_at_1_std
value: 38.6812
- type: nauc_recall_at_1_diff1
value: 52.2489
- type: nauc_recall_at_3_max
value: 37.8719
- type: nauc_recall_at_3_std
value: 31.4138
- type: nauc_recall_at_3_diff1
value: 27.2774
- type: nauc_recall_at_5_max
value: 33.8087
- type: nauc_recall_at_5_std
value: 30.3732
- type: nauc_recall_at_5_diff1
value: 22.7426
- type: nauc_recall_at_10_max
value: 28.926299999999998
- type: nauc_recall_at_10_std
value: 26.916600000000003
- type: nauc_recall_at_10_diff1
value: 16.872300000000003
- type: nauc_recall_at_20_max
value: 26.705499999999997
- type: nauc_recall_at_20_std
value: 25.8692
- type: nauc_recall_at_20_diff1
value: 12.734599999999999
- type: nauc_recall_at_100_max
value: 22.6795
- type: nauc_recall_at_100_std
value: 24.3181
- type: nauc_recall_at_100_diff1
value: 11.6484
- type: nauc_recall_at_1000_max
value: 28.498800000000003
- type: nauc_recall_at_1000_std
value: 36.8172
- type: nauc_recall_at_1000_diff1
value: 11.0337
- type: nauc_precision_at_1_max
value: 49.0499
- type: nauc_precision_at_1_std
value: 38.6812
- type: nauc_precision_at_1_diff1
value: 52.2489
- type: nauc_precision_at_3_max
value: 37.8719
- type: nauc_precision_at_3_std
value: 31.4138
- type: nauc_precision_at_3_diff1
value: 27.2774
- type: nauc_precision_at_5_max
value: 33.8087
- type: nauc_precision_at_5_std
value: 30.3732
- type: nauc_precision_at_5_diff1
value: 22.7426
- type: nauc_precision_at_10_max
value: 28.926299999999998
- type: nauc_precision_at_10_std
value: 26.916600000000003
- type: nauc_precision_at_10_diff1
value: 16.872300000000003
- type: nauc_precision_at_20_max
value: 26.705499999999997
- type: nauc_precision_at_20_std
value: 25.8692
- type: nauc_precision_at_20_diff1
value: 12.734599999999999
- type: nauc_precision_at_100_max
value: 22.6795
- type: nauc_precision_at_100_std
value: 24.3181
- type: nauc_precision_at_100_diff1
value: 11.6484
- type: nauc_precision_at_1000_max
value: 28.498800000000003
- type: nauc_precision_at_1000_std
value: 36.8172
- type: nauc_precision_at_1000_diff1
value: 11.0337
- type: nauc_mrr_at_1_max
value: 49.0499
- type: nauc_mrr_at_1_std
value: 38.6812
- type: nauc_mrr_at_1_diff1
value: 52.2489
- type: nauc_mrr_at_3_max
value: 42.479499999999994
- type: nauc_mrr_at_3_std
value: 34.5065
- type: nauc_mrr_at_3_diff1
value: 37.5021
- type: nauc_mrr_at_5_max
value: 40.6623
- type: nauc_mrr_at_5_std
value: 34.0191
- type: nauc_mrr_at_5_diff1
value: 34.8592
- type: nauc_mrr_at_10_max
value: 38.600899999999996
- type: nauc_mrr_at_10_std
value: 32.5849
- type: nauc_mrr_at_10_diff1
value: 32.1012
- type: nauc_mrr_at_20_max
value: 37.6983
- type: nauc_mrr_at_20_std
value: 32.2239
- type: nauc_mrr_at_20_diff1
value: 30.6472
- type: nauc_mrr_at_100_max
value: 37.0514
- type: nauc_mrr_at_100_std
value: 31.941000000000003
- type: nauc_mrr_at_100_diff1
value: 29.9615
- type: nauc_mrr_at_1000_max
value: 37.1014
- type: nauc_mrr_at_1000_std
value: 32.0581
- type: nauc_mrr_at_1000_diff1
value: 30.025000000000002
- type: main_score
value: 6.275
- task:
type: Retrieval
dataset:
name: MTEB MLQARetrieval (zho-ara)
type: facebook/mlqa
config: zho-ara
split: test
revision: 397ed406c1a7902140303e7faf60fff35b58d285
metrics:
- type: ndcg_at_1
value: 3.1399999999999997
- type: ndcg_at_3
value: 4.377000000000001
- type: ndcg_at_5
value: 4.825
- type: ndcg_at_10
value: 5.487
- type: ndcg_at_20
value: 6.002
- type: ndcg_at_100
value: 7.968
- type: ndcg_at_1000
value: 14.102999999999998
- type: map_at_1
value: 3.1399999999999997
- type: map_at_3
value: 4.064
- type: map_at_5
value: 4.31
- type: map_at_10
value: 4.585
- type: map_at_20
value: 4.718
- type: map_at_100
value: 4.972
- type: map_at_1000
value: 5.132
- type: recall_at_1
value: 3.1399999999999997
- type: recall_at_3
value: 5.285
- type: recall_at_5
value: 6.3839999999999995
- type: recall_at_10
value: 8.425
- type: recall_at_20
value: 10.517999999999999
- type: recall_at_100
value: 21.401999999999997
- type: recall_at_1000
value: 74.09700000000001
- type: precision_at_1
value: 3.1399999999999997
- type: precision_at_3
value: 1.762
- type: precision_at_5
value: 1.277
- type: precision_at_10
value: 0.8420000000000001
- type: precision_at_20
value: 0.526
- type: precision_at_100
value: 0.214
- type: precision_at_1000
value: 0.074
- type: mrr_at_1
value: 3.1397
- type: mrr_at_3
value: 4.0642
- type: mrr_at_5
value: 4.3101
- type: mrr_at_10
value: 4.584499999999999
- type: mrr_at_20
value: 4.7184
- type: mrr_at_100
value: 4.9722
- type: mrr_at_1000
value: 5.1322
- type: nauc_ndcg_at_1_max
value: 53.1102
- type: nauc_ndcg_at_1_std
value: 41.6914
- type: nauc_ndcg_at_1_diff1
value: 60.5043
- type: nauc_ndcg_at_3_max
value: 49.2169
- type: nauc_ndcg_at_3_std
value: 46.7961
- type: nauc_ndcg_at_3_diff1
value: 43.0363
- type: nauc_ndcg_at_5_max
value: 46.6068
- type: nauc_ndcg_at_5_std
value: 44.6031
- type: nauc_ndcg_at_5_diff1
value: 39.915
- type: nauc_ndcg_at_10_max
value: 43.007400000000004
- type: nauc_ndcg_at_10_std
value: 41.646300000000004
- type: nauc_ndcg_at_10_diff1
value: 36.1524
- type: nauc_ndcg_at_20_max
value: 40.2
- type: nauc_ndcg_at_20_std
value: 40.2874
- type: nauc_ndcg_at_20_diff1
value: 33.4982
- type: nauc_ndcg_at_100_max
value: 32.7883
- type: nauc_ndcg_at_100_std
value: 37.7631
- type: nauc_ndcg_at_100_diff1
value: 25.5545
- type: nauc_ndcg_at_1000_max
value: 31.622600000000002
- type: nauc_ndcg_at_1000_std
value: 34.7798
- type: nauc_ndcg_at_1000_diff1
value: 26.189
- type: nauc_map_at_1_max
value: 53.1102
- type: nauc_map_at_1_std
value: 41.6914
- type: nauc_map_at_1_diff1
value: 60.5043
- type: nauc_map_at_3_max
value: 50.2741
- type: nauc_map_at_3_std
value: 45.9366
- type: nauc_map_at_3_diff1
value: 46.476800000000004
- type: nauc_map_at_5_max
value: 48.6312
- type: nauc_map_at_5_std
value: 44.6575
- type: nauc_map_at_5_diff1
value: 44.4099
- type: nauc_map_at_10_max
value: 46.7695
- type: nauc_map_at_10_std
value: 43.1466
- type: nauc_map_at_10_diff1
value: 42.2738
- type: nauc_map_at_20_max
value: 45.7776
- type: nauc_map_at_20_std
value: 42.6586
- type: nauc_map_at_20_diff1
value: 41.2568
- type: nauc_map_at_100_max
value: 44.1608
- type: nauc_map_at_100_std
value: 42.1323
- type: nauc_map_at_100_diff1
value: 39.4298
- type: nauc_map_at_1000_max
value: 43.9725
- type: nauc_map_at_1000_std
value: 41.9294
- type: nauc_map_at_1000_diff1
value: 39.3602
- type: nauc_recall_at_1_max
value: 53.1102
- type: nauc_recall_at_1_std
value: 41.6914
- type: nauc_recall_at_1_diff1
value: 60.5043
- type: nauc_recall_at_3_max
value: 46.7656
- type: nauc_recall_at_3_std
value: 48.6744
- type: nauc_recall_at_3_diff1
value: 35.342400000000005
- type: nauc_recall_at_5_max
value: 42.2896
- type: nauc_recall_at_5_std
value: 44.2316
- type: nauc_recall_at_5_diff1
value: 30.748399999999997
- type: nauc_recall_at_10_max
value: 35.9736
- type: nauc_recall_at_10_std
value: 38.500099999999996
- type: nauc_recall_at_10_diff1
value: 25.4139
- type: nauc_recall_at_20_max
value: 30.5874
- type: nauc_recall_at_20_std
value: 35.9068
- type: nauc_recall_at_20_diff1
value: 21.124000000000002
- type: nauc_recall_at_100_max
value: 17.197699999999998
- type: nauc_recall_at_100_std
value: 31.5631
- type: nauc_recall_at_100_diff1
value: 7.7295
- type: nauc_recall_at_1000_max
value: 10.2237
- type: nauc_recall_at_1000_std
value: 18.3387
- type: nauc_recall_at_1000_diff1
value: 6.905200000000001
- type: nauc_precision_at_1_max
value: 53.1102
- type: nauc_precision_at_1_std
value: 41.6914
- type: nauc_precision_at_1_diff1
value: 60.5043
- type: nauc_precision_at_3_max
value: 46.7656
- type: nauc_precision_at_3_std
value: 48.6744
- type: nauc_precision_at_3_diff1
value: 35.342400000000005
- type: nauc_precision_at_5_max
value: 42.2896
- type: nauc_precision_at_5_std
value: 44.2316
- type: nauc_precision_at_5_diff1
value: 30.748399999999997
- type: nauc_precision_at_10_max
value: 35.9736
- type: nauc_precision_at_10_std
value: 38.500099999999996
- type: nauc_precision_at_10_diff1
value: 25.4139
- type: nauc_precision_at_20_max
value: 30.5874
- type: nauc_precision_at_20_std
value: 35.9068
- type: nauc_precision_at_20_diff1
value: 21.124000000000002
- type: nauc_precision_at_100_max
value: 17.197699999999998
- type: nauc_precision_at_100_std
value: 31.5631
- type: nauc_precision_at_100_diff1
value: 7.7295
- type: nauc_precision_at_1000_max
value: 10.0574
- type: nauc_precision_at_1000_std
value: 18.2383
- type: nauc_precision_at_1000_diff1
value: 6.6805
- type: nauc_mrr_at_1_max
value: 53.1102
- type: nauc_mrr_at_1_std
value: 41.6914
- type: nauc_mrr_at_1_diff1
value: 60.5043
- type: nauc_mrr_at_3_max
value: 50.2741
- type: nauc_mrr_at_3_std
value: 45.9366
- type: nauc_mrr_at_3_diff1
value: 46.476800000000004
- type: nauc_mrr_at_5_max
value: 48.6312
- type: nauc_mrr_at_5_std
value: 44.6575
- type: nauc_mrr_at_5_diff1
value: 44.4099
- type: nauc_mrr_at_10_max
value: 46.7695
- type: nauc_mrr_at_10_std
value: 43.1466
- type: nauc_mrr_at_10_diff1
value: 42.2738
- type: nauc_mrr_at_20_max
value: 45.7776
- type: nauc_mrr_at_20_std
value: 42.6586
- type: nauc_mrr_at_20_diff1
value: 41.2568
- type: nauc_mrr_at_100_max
value: 44.1609
- type: nauc_mrr_at_100_std
value: 42.1322
- type: nauc_mrr_at_100_diff1
value: 39.4299
- type: nauc_mrr_at_1000_max
value: 43.973099999999995
- type: nauc_mrr_at_1000_std
value: 41.9295
- type: nauc_mrr_at_1000_diff1
value: 39.361000000000004
- type: main_score
value: 5.487
- task:
type: Retrieval
dataset:
name: MTEB MintakaRetrieval (ar)
type: jinaai/mintakaqa
config: ar
split: test
revision: efa78cc2f74bbcd21eff2261f9e13aebe40b814e
metrics:
- type: ndcg_at_1
value: 9.940999999999999
- type: ndcg_at_3
value: 14.41
- type: ndcg_at_5
value: 16.303
- type: ndcg_at_10
value: 18.23
- type: ndcg_at_20
value: 19.891000000000002
- type: ndcg_at_100
value: 22.578
- type: ndcg_at_1000
value: 27.236
- type: map_at_1
value: 9.940999999999999
- type: map_at_3
value: 13.277
- type: map_at_5
value: 14.330000000000002
- type: map_at_10
value: 15.120000000000001
- type: map_at_20
value: 15.573
- type: map_at_100
value: 15.925
- type: map_at_1000
value: 16.056
- type: recall_at_1
value: 9.940999999999999
- type: recall_at_3
value: 17.703
- type: recall_at_5
value: 22.288
- type: recall_at_10
value: 28.28
- type: recall_at_20
value: 34.862
- type: recall_at_100
value: 49.66
- type: recall_at_1000
value: 88.97
- type: precision_at_1
value: 9.940999999999999
- type: precision_at_3
value: 5.901
- type: precision_at_5
value: 4.458
- type: precision_at_10
value: 2.828
- type: precision_at_20
value: 1.743
- type: precision_at_100
value: 0.49699999999999994
- type: precision_at_1000
value: 0.089
- type: mrr_at_1
value: 9.940999999999999
- type: mrr_at_3
value: 13.2773
- type: mrr_at_5
value: 14.330499999999999
- type: mrr_at_10
value: 15.1196
- type: mrr_at_20
value: 15.5731
- type: mrr_at_100
value: 15.9247
- type: mrr_at_1000
value: 16.0563
- type: nauc_ndcg_at_1_max
value: 29.738799999999998
- type: nauc_ndcg_at_1_std
value: 3.3945999999999996
- type: nauc_ndcg_at_1_diff1
value: 27.060000000000002
- type: nauc_ndcg_at_3_max
value: 27.002399999999998
- type: nauc_ndcg_at_3_std
value: 6.1634
- type: nauc_ndcg_at_3_diff1
value: 19.4654
- type: nauc_ndcg_at_5_max
value: 26.9374
- type: nauc_ndcg_at_5_std
value: 8.087
- type: nauc_ndcg_at_5_diff1
value: 17.641399999999997
- type: nauc_ndcg_at_10_max
value: 26.239
- type: nauc_ndcg_at_10_std
value: 9.7034
- type: nauc_ndcg_at_10_diff1
value: 16.309199999999997
- type: nauc_ndcg_at_20_max
value: 25.8932
- type: nauc_ndcg_at_20_std
value: 10.4576
- type: nauc_ndcg_at_20_diff1
value: 16.0602
- type: nauc_ndcg_at_100_max
value: 25.400299999999998
- type: nauc_ndcg_at_100_std
value: 11.3135
- type: nauc_ndcg_at_100_diff1
value: 16.2558
- type: nauc_ndcg_at_1000_max
value: 25.879
- type: nauc_ndcg_at_1000_std
value: 10.5304
- type: nauc_ndcg_at_1000_diff1
value: 16.8128
- type: nauc_map_at_1_max
value: 29.738799999999998
- type: nauc_map_at_1_std
value: 3.3945999999999996
- type: nauc_map_at_1_diff1
value: 27.060000000000002
- type: nauc_map_at_3_max
value: 27.478599999999997
- type: nauc_map_at_3_std
value: 5.5567
- type: nauc_map_at_3_diff1
value: 20.8918
- type: nauc_map_at_5_max
value: 27.447300000000002
- type: nauc_map_at_5_std
value: 6.7867999999999995
- type: nauc_map_at_5_diff1
value: 19.7197
- type: nauc_map_at_10_max
value: 27.095599999999997
- type: nauc_map_at_10_std
value: 7.552499999999999
- type: nauc_map_at_10_diff1
value: 19.05
- type: nauc_map_at_20_max
value: 26.9449
- type: nauc_map_at_20_std
value: 7.807500000000001
- type: nauc_map_at_20_diff1
value: 18.9194
- type: nauc_map_at_100_max
value: 26.8807
- type: nauc_map_at_100_std
value: 7.9676
- type: nauc_map_at_100_diff1
value: 18.9621
- type: nauc_map_at_1000_max
value: 26.8887
- type: nauc_map_at_1000_std
value: 7.9346
- type: nauc_map_at_1000_diff1
value: 18.9753
- type: nauc_recall_at_1_max
value: 29.738799999999998
- type: nauc_recall_at_1_std
value: 3.3945999999999996
- type: nauc_recall_at_1_diff1
value: 27.060000000000002
- type: nauc_recall_at_3_max
value: 25.9167
- type: nauc_recall_at_3_std
value: 7.593999999999999
- type: nauc_recall_at_3_diff1
value: 16.1735
- type: nauc_recall_at_5_max
value: 25.8469
- type: nauc_recall_at_5_std
value: 11.0169
- type: nauc_recall_at_5_diff1
value: 13.0884
- type: nauc_recall_at_10_max
value: 24.4113
- type: nauc_recall_at_10_std
value: 14.496999999999998
- type: nauc_recall_at_10_diff1
value: 10.5047
- type: nauc_recall_at_20_max
value: 23.6952
- type: nauc_recall_at_20_std
value: 16.3849
- type: nauc_recall_at_20_diff1
value: 10.2638
- type: nauc_recall_at_100_max
value: 21.5628
- type: nauc_recall_at_100_std
value: 19.586100000000002
- type: nauc_recall_at_100_diff1
value: 10.7761
- type: nauc_recall_at_1000_max
value: 22.493199999999998
- type: nauc_recall_at_1000_std
value: 23.7462
- type: nauc_recall_at_1000_diff1
value: 9.5045
- type: nauc_precision_at_1_max
value: 29.738799999999998
- type: nauc_precision_at_1_std
value: 3.3945999999999996
- type: nauc_precision_at_1_diff1
value: 27.060000000000002
- type: nauc_precision_at_3_max
value: 25.9167
- type: nauc_precision_at_3_std
value: 7.593999999999999
- type: nauc_precision_at_3_diff1
value: 16.1735
- type: nauc_precision_at_5_max
value: 25.8469
- type: nauc_precision_at_5_std
value: 11.0169
- type: nauc_precision_at_5_diff1
value: 13.0884
- type: nauc_precision_at_10_max
value: 24.4113
- type: nauc_precision_at_10_std
value: 14.496999999999998
- type: nauc_precision_at_10_diff1
value: 10.5047
- type: nauc_precision_at_20_max
value: 23.6952
- type: nauc_precision_at_20_std
value: 16.3849
- type: nauc_precision_at_20_diff1
value: 10.2638
- type: nauc_precision_at_100_max
value: 21.5628
- type: nauc_precision_at_100_std
value: 19.586100000000002
- type: nauc_precision_at_100_diff1
value: 10.7761
- type: nauc_precision_at_1000_max
value: 22.493199999999998
- type: nauc_precision_at_1000_std
value: 23.7462
- type: nauc_precision_at_1000_diff1
value: 9.5045
- type: nauc_mrr_at_1_max
value: 29.738799999999998
- type: nauc_mrr_at_1_std
value: 3.3945999999999996
- type: nauc_mrr_at_1_diff1
value: 27.060000000000002
- type: nauc_mrr_at_3_max
value: 27.478599999999997
- type: nauc_mrr_at_3_std
value: 5.5567
- type: nauc_mrr_at_3_diff1
value: 20.8918
- type: nauc_mrr_at_5_max
value: 27.447300000000002
- type: nauc_mrr_at_5_std
value: 6.7867999999999995
- type: nauc_mrr_at_5_diff1
value: 19.7197
- type: nauc_mrr_at_10_max
value: 27.095599999999997
- type: nauc_mrr_at_10_std
value: 7.552499999999999
- type: nauc_mrr_at_10_diff1
value: 19.05
- type: nauc_mrr_at_20_max
value: 26.9449
- type: nauc_mrr_at_20_std
value: 7.807500000000001
- type: nauc_mrr_at_20_diff1
value: 18.9194
- type: nauc_mrr_at_100_max
value: 26.8807
- type: nauc_mrr_at_100_std
value: 7.9676
- type: nauc_mrr_at_100_diff1
value: 18.9621
- type: nauc_mrr_at_1000_max
value: 26.8887
- type: nauc_mrr_at_1000_std
value: 7.9346
- type: nauc_mrr_at_1000_diff1
value: 18.9753
- type: main_score
value: 18.23
- task:
type: Retrieval
dataset:
name: MTEB MrTidyRetrieval (arabic)
type: mteb/mrtidy
config: arabic
split: test
revision: fc24a3ce8f09746410daee3d5cd823ff7a0675b7
metrics:
- type: ndcg_at_1
value: 7.401000000000001
- type: ndcg_at_3
value: 11.512
- type: ndcg_at_5
value: 14.002999999999998
- type: ndcg_at_10
value: 17.378
- type: ndcg_at_20
value: 20.241
- type: ndcg_at_100
value: 24.549000000000003
- type: ndcg_at_1000
value: 27.012000000000004
- type: map_at_1
value: 6.984
- type: map_at_3
value: 10.213999999999999
- type: map_at_5
value: 11.603
- type: map_at_10
value: 13.025
- type: map_at_20
value: 13.816999999999998
- type: map_at_100
value: 14.447
- type: map_at_1000
value: 14.536999999999999
- type: recall_at_1
value: 6.984
- type: recall_at_3
value: 14.462
- type: recall_at_5
value: 20.321
- type: recall_at_10
value: 30.342000000000002
- type: recall_at_20
value: 41.243
- type: recall_at_100
value: 63.599000000000004
- type: recall_at_1000
value: 82.609
- type: precision_at_1
value: 7.401000000000001
- type: precision_at_3
value: 5.365
- type: precision_at_5
value: 4.569999999999999
- type: precision_at_10
value: 3.4410000000000003
- type: precision_at_20
value: 2.3539999999999996
- type: precision_at_100
value: 0.731
- type: precision_at_1000
value: 0.096
- type: mrr_at_1
value: 7.4006
- type: mrr_at_3
value: 10.9929
- type: mrr_at_5
value: 12.417499999999999
- type: mrr_at_10
value: 13.8602
- type: mrr_at_20
value: 14.682500000000001
- type: mrr_at_100
value: 15.25
- type: mrr_at_1000
value: 15.3278
- type: nauc_ndcg_at_1_max
value: 4.4628
- type: nauc_ndcg_at_1_std
value: 0.0991
- type: nauc_ndcg_at_1_diff1
value: 7.2256
- type: nauc_ndcg_at_3_max
value: 5.8659
- type: nauc_ndcg_at_3_std
value: 4.412599999999999
- type: nauc_ndcg_at_3_diff1
value: 5.5699
- type: nauc_ndcg_at_5_max
value: 7.5637
- type: nauc_ndcg_at_5_std
value: 5.2681
- type: nauc_ndcg_at_5_diff1
value: 6.2124
- type: nauc_ndcg_at_10_max
value: 10.6347
- type: nauc_ndcg_at_10_std
value: 6.1522
- type: nauc_ndcg_at_10_diff1
value: 6.2313
- type: nauc_ndcg_at_20_max
value: 11.1052
- type: nauc_ndcg_at_20_std
value: 8.0997
- type: nauc_ndcg_at_20_diff1
value: 6.259099999999999
- type: nauc_ndcg_at_100_max
value: 12.1237
- type: nauc_ndcg_at_100_std
value: 11.128300000000001
- type: nauc_ndcg_at_100_diff1
value: 6.855
- type: nauc_ndcg_at_1000_max
value: 12.0395
- type: nauc_ndcg_at_1000_std
value: 11.9957
- type: nauc_ndcg_at_1000_diff1
value: 7.0405999999999995
- type: nauc_map_at_1_max
value: 4.0845
- type: nauc_map_at_1_std
value: -0.6178
- type: nauc_map_at_1_diff1
value: 6.468400000000001
- type: nauc_map_at_3_max
value: 5.214499999999999
- type: nauc_map_at_3_std
value: 3.3358
- type: nauc_map_at_3_diff1
value: 5.5802
- type: nauc_map_at_5_max
value: 6.3618999999999994
- type: nauc_map_at_5_std
value: 4.0575
- type: nauc_map_at_5_diff1
value: 6.0938
- type: nauc_map_at_10_max
value: 7.9055
- type: nauc_map_at_10_std
value: 4.4857000000000005
- type: nauc_map_at_10_diff1
value: 5.9283
- type: nauc_map_at_20_max
value: 8.0925
- type: nauc_map_at_20_std
value: 5.194
- type: nauc_map_at_20_diff1
value: 5.9140999999999995
- type: nauc_map_at_100_max
value: 8.315100000000001
- type: nauc_map_at_100_std
value: 5.7394
- type: nauc_map_at_100_diff1
value: 6.0712
- type: nauc_map_at_1000_max
value: 8.3048
- type: nauc_map_at_1000_std
value: 5.7991
- type: nauc_map_at_1000_diff1
value: 6.0765
- type: nauc_recall_at_1_max
value: 4.0845
- type: nauc_recall_at_1_std
value: -0.6178
- type: nauc_recall_at_1_diff1
value: 6.468400000000001
- type: nauc_recall_at_3_max
value: 7.1412
- type: nauc_recall_at_3_std
value: 6.5206
- type: nauc_recall_at_3_diff1
value: 5.220000000000001
- type: nauc_recall_at_5_max
value: 9.8023
- type: nauc_recall_at_5_std
value: 7.240099999999999
- type: nauc_recall_at_5_diff1
value: 6.4299
- type: nauc_recall_at_10_max
value: 15.7093
- type: nauc_recall_at_10_std
value: 8.549800000000001
- type: nauc_recall_at_10_diff1
value: 6.7775
- type: nauc_recall_at_20_max
value: 16.723
- type: nauc_recall_at_20_std
value: 13.177
- type: nauc_recall_at_20_diff1
value: 6.816
- type: nauc_recall_at_100_max
value: 21.105999999999998
- type: nauc_recall_at_100_std
value: 25.0098
- type: nauc_recall_at_100_diff1
value: 8.9565
- type: nauc_recall_at_1000_max
value: 26.9686
- type: nauc_recall_at_1000_std
value: 41.6479
- type: nauc_recall_at_1000_diff1
value: 12.691099999999999
- type: nauc_precision_at_1_max
value: 4.4628
- type: nauc_precision_at_1_std
value: 0.0991
- type: nauc_precision_at_1_diff1
value: 7.2256
- type: nauc_precision_at_3_max
value: 8.185
- type: nauc_precision_at_3_std
value: 7.5577000000000005
- type: nauc_precision_at_3_diff1
value: 6.4395999999999995
- type: nauc_precision_at_5_max
value: 10.7
- type: nauc_precision_at_5_std
value: 9.5349
- type: nauc_precision_at_5_diff1
value: 6.7633
- type: nauc_precision_at_10_max
value: 15.4529
- type: nauc_precision_at_10_std
value: 10.758700000000001
- type: nauc_precision_at_10_diff1
value: 5.9852
- type: nauc_precision_at_20_max
value: 16.1342
- type: nauc_precision_at_20_std
value: 15.7733
- type: nauc_precision_at_20_diff1
value: 5.9866
- type: nauc_precision_at_100_max
value: 18.0199
- type: nauc_precision_at_100_std
value: 25.7156
- type: nauc_precision_at_100_diff1
value: 6.7398
- type: nauc_precision_at_1000_max
value: 16.2
- type: nauc_precision_at_1000_std
value: 30.476599999999998
- type: nauc_precision_at_1000_diff1
value: 4.853
- type: nauc_mrr_at_1_max
value: 4.4628
- type: nauc_mrr_at_1_std
value: 0.0991
- type: nauc_mrr_at_1_diff1
value: 7.2256
- type: nauc_mrr_at_3_max
value: 5.3888
- type: nauc_mrr_at_3_std
value: 3.6304000000000003
- type: nauc_mrr_at_3_diff1
value: 5.9391
- type: nauc_mrr_at_5_max
value: 6.442399999999999
- type: nauc_mrr_at_5_std
value: 4.1495999999999995
- type: nauc_mrr_at_5_diff1
value: 6.15
- type: nauc_mrr_at_10_max
value: 8.031
- type: nauc_mrr_at_10_std
value: 4.7912
- type: nauc_mrr_at_10_diff1
value: 6.269900000000001
- type: nauc_mrr_at_20_max
value: 8.0549
- type: nauc_mrr_at_20_std
value: 5.2743
- type: nauc_mrr_at_20_diff1
value: 6.2928999999999995
- type: nauc_mrr_at_100_max
value: 8.2201
- type: nauc_mrr_at_100_std
value: 5.7367
- type: nauc_mrr_at_100_diff1
value: 6.3441
- type: nauc_mrr_at_1000_max
value: 8.211
- type: nauc_mrr_at_1000_std
value: 5.7768
- type: nauc_mrr_at_1000_diff1
value: 6.366199999999999
- type: main_score
value: 17.378
- task:
type: Retrieval
dataset:
name: MTEB SadeemQuestionRetrieval (default)
type: sadeem-ai/sadeem-ar-eval-retrieval-questions
config: default
split: test
revision: 3cb0752b182e5d5d740df547748b06663c8e0bd9
metrics:
- type: ndcg_at_1
value: 28.147
- type: ndcg_at_3
value: 59.156
- type: ndcg_at_5
value: 61.065999999999995
- type: ndcg_at_10
value: 62.241
- type: ndcg_at_20
value: 62.873000000000005
- type: ndcg_at_100
value: 63.676
- type: ndcg_at_1000
value: 63.904
- type: map_at_1
value: 28.147
- type: map_at_3
value: 50.989
- type: map_at_5
value: 52.059
- type: map_at_10
value: 52.553000000000004
- type: map_at_20
value: 52.727999999999994
- type: map_at_100
value: 52.842999999999996
- type: map_at_1000
value: 52.852
- type: recall_at_1
value: 28.147
- type: recall_at_3
value: 83.006
- type: recall_at_5
value: 87.602
- type: recall_at_10
value: 91.192
- type: recall_at_20
value: 93.681
- type: recall_at_100
value: 97.942
- type: recall_at_1000
value: 99.713
- type: precision_at_1
value: 28.147
- type: precision_at_3
value: 27.669
- type: precision_at_5
value: 17.52
- type: precision_at_10
value: 9.119
- type: precision_at_20
value: 4.684
- type: precision_at_100
value: 0.979
- type: precision_at_1000
value: 0.1
- type: mrr_at_1
value: 26.9507
- type: mrr_at_3
value: 50.1675
- type: mrr_at_5
value: 51.220699999999994
- type: mrr_at_10
value: 51.739599999999996
- type: mrr_at_20
value: 51.9078
- type: mrr_at_100
value: 52.019000000000005
- type: mrr_at_1000
value: 52.027699999999996
- type: nauc_ndcg_at_1_max
value: 12.1091
- type: nauc_ndcg_at_1_std
value: -0.2641
- type: nauc_ndcg_at_1_diff1
value: -0.0456
- type: nauc_ndcg_at_3_max
value: 32.2194
- type: nauc_ndcg_at_3_std
value: 6.8115
- type: nauc_ndcg_at_3_diff1
value: -44.6169
- type: nauc_ndcg_at_5_max
value: 30.223499999999998
- type: nauc_ndcg_at_5_std
value: 6.616
- type: nauc_ndcg_at_5_diff1
value: -37.8131
- type: nauc_ndcg_at_10_max
value: 28.215
- type: nauc_ndcg_at_10_std
value: 6.638199999999999
- type: nauc_ndcg_at_10_diff1
value: -34.1462
- type: nauc_ndcg_at_20_max
value: 27.520699999999998
- type: nauc_ndcg_at_20_std
value: 6.793
- type: nauc_ndcg_at_20_diff1
value: -31.5702
- type: nauc_ndcg_at_100_max
value: 25.8959
- type: nauc_ndcg_at_100_std
value: 6.0431
- type: nauc_ndcg_at_100_diff1
value: -27.7369
- type: nauc_ndcg_at_1000_max
value: 25.263999999999996
- type: nauc_ndcg_at_1000_std
value: 5.544099999999999
- type: nauc_ndcg_at_1000_diff1
value: -26.8195
- type: nauc_map_at_1_max
value: 12.1091
- type: nauc_map_at_1_std
value: -0.2641
- type: nauc_map_at_1_diff1
value: -0.0456
- type: nauc_map_at_3_max
value: 25.443500000000004
- type: nauc_map_at_3_std
value: 4.6888
- type: nauc_map_at_3_diff1
value: -28.6402
- type: nauc_map_at_5_max
value: 24.252000000000002
- type: nauc_map_at_5_std
value: 4.518599999999999
- type: nauc_map_at_5_diff1
value: -24.7719
- type: nauc_map_at_10_max
value: 23.4405
- type: nauc_map_at_10_std
value: 4.5044
- type: nauc_map_at_10_diff1
value: -23.2632
- type: nauc_map_at_20_max
value: 23.2572
- type: nauc_map_at_20_std
value: 4.539499999999999
- type: nauc_map_at_20_diff1
value: -22.6096
- type: nauc_map_at_100_max
value: 23.055
- type: nauc_map_at_100_std
value: 4.4593
- type: nauc_map_at_100_diff1
value: -22.1369
- type: nauc_map_at_1000_max
value: 23.035600000000002
- type: nauc_map_at_1000_std
value: 4.4453
- type: nauc_map_at_1000_diff1
value: -22.1081
- type: nauc_recall_at_1_max
value: 12.1091
- type: nauc_recall_at_1_std
value: -0.2641
- type: nauc_recall_at_1_diff1
value: -0.0456
- type: nauc_recall_at_3_max
value: 66.4442
- type: nauc_recall_at_3_std
value: 17.372799999999998
- type: nauc_recall_at_3_diff1
value: -125.90520000000001
- type: nauc_recall_at_5_max
value: 68.48689999999999
- type: nauc_recall_at_5_std
value: 19.979
- type: nauc_recall_at_5_diff1
value: -121.6742
- type: nauc_recall_at_10_max
value: 67.44839999999999
- type: nauc_recall_at_10_std
value: 24.8948
- type: nauc_recall_at_10_diff1
value: -124.82839999999999
- type: nauc_recall_at_20_max
value: 73.4407
- type: nauc_recall_at_20_std
value: 33.7021
- type: nauc_recall_at_20_diff1
value: -126.0851
- type: nauc_recall_at_100_max
value: 81.9264
- type: nauc_recall_at_100_std
value: 46.7656
- type: nauc_recall_at_100_diff1
value: -117.83879999999999
- type: nauc_recall_at_1000_max
value: 76.4994
- type: nauc_recall_at_1000_std
value: 16.3124
- type: nauc_recall_at_1000_diff1
value: -164.1088
- type: nauc_precision_at_1_max
value: 12.1091
- type: nauc_precision_at_1_std
value: -0.2641
- type: nauc_precision_at_1_diff1
value: -0.0456
- type: nauc_precision_at_3_max
value: 66.4442
- type: nauc_precision_at_3_std
value: 17.372799999999998
- type: nauc_precision_at_3_diff1
value: -125.90520000000001
- type: nauc_precision_at_5_max
value: 68.48689999999999
- type: nauc_precision_at_5_std
value: 19.979
- type: nauc_precision_at_5_diff1
value: -121.6742
- type: nauc_precision_at_10_max
value: 67.44839999999999
- type: nauc_precision_at_10_std
value: 24.8948
- type: nauc_precision_at_10_diff1
value: -124.82839999999999
- type: nauc_precision_at_20_max
value: 73.4407
- type: nauc_precision_at_20_std
value: 33.7021
- type: nauc_precision_at_20_diff1
value: -126.0851
- type: nauc_precision_at_100_max
value: 81.9264
- type: nauc_precision_at_100_std
value: 46.7656
- type: nauc_precision_at_100_diff1
value: -117.83879999999999
- type: nauc_precision_at_1000_max
value: 76.4994
- type: nauc_precision_at_1000_std
value: 16.3124
- type: nauc_precision_at_1000_diff1
value: -164.1088
- type: nauc_mrr_at_1_max
value: 12.9902
- type: nauc_mrr_at_1_std
value: 4.414499999999999
- type: nauc_mrr_at_1_diff1
value: -24.3025
- type: nauc_mrr_at_3_max
value: 26.009500000000003
- type: nauc_mrr_at_3_std
value: 7.7266
- type: nauc_mrr_at_3_diff1
value: -47.2008
- type: nauc_mrr_at_5_max
value: 24.5728
- type: nauc_mrr_at_5_std
value: 7.8084
- type: nauc_mrr_at_5_diff1
value: -44.370599999999996
- type: nauc_mrr_at_10_max
value: 23.688000000000002
- type: nauc_mrr_at_10_std
value: 7.656300000000001
- type: nauc_mrr_at_10_diff1
value: -42.9363
- type: nauc_mrr_at_20_max
value: 23.5016
- type: nauc_mrr_at_20_std
value: 7.7171
- type: nauc_mrr_at_20_diff1
value: -42.4626
- type: nauc_mrr_at_100_max
value: 23.304
- type: nauc_mrr_at_100_std
value: 7.6429
- type: nauc_mrr_at_100_diff1
value: -42.094
- type: nauc_mrr_at_1000_max
value: 23.2846
- type: nauc_mrr_at_1000_std
value: 7.6298
- type: nauc_mrr_at_1000_diff1
value: -42.0719
- type: main_score
value: 62.241
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (ara-ara)
type: jinaai/xpqa
config: ara-ara
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 26.0
- type: ndcg_at_3
value: 27.519
- type: ndcg_at_5
value: 29.212
- type: ndcg_at_10
value: 33.564
- type: ndcg_at_20
value: 36.436
- type: ndcg_at_100
value: 40.905
- type: ndcg_at_1000
value: 44.172
- type: map_at_1
value: 13.862
- type: map_at_3
value: 22.226000000000003
- type: map_at_5
value: 24.876
- type: map_at_10
value: 27.217000000000002
- type: map_at_20
value: 28.259
- type: map_at_100
value: 29.076999999999998
- type: map_at_1000
value: 29.232000000000003
- type: recall_at_1
value: 13.862
- type: recall_at_3
value: 26.700000000000003
- type: recall_at_5
value: 33.42
- type: recall_at_10
value: 44.393
- type: recall_at_20
value: 54.08
- type: recall_at_100
value: 74.53999999999999
- type: recall_at_1000
value: 97.251
- type: precision_at_1
value: 26.0
- type: precision_at_3
value: 19.022
- type: precision_at_5
value: 14.613000000000001
- type: precision_at_10
value: 9.68
- type: precision_at_20
value: 5.779999999999999
- type: precision_at_100
value: 1.5650000000000002
- type: precision_at_1000
value: 0.196
- type: mrr_at_1
value: 26.0
- type: mrr_at_3
value: 31.2222
- type: mrr_at_5
value: 32.8089
- type: mrr_at_10
value: 34.2539
- type: mrr_at_20
value: 34.8057
- type: mrr_at_100
value: 35.2117
- type: mrr_at_1000
value: 35.2937
- type: nauc_ndcg_at_1_max
value: 37.0131
- type: nauc_ndcg_at_1_std
value: 1.23
- type: nauc_ndcg_at_1_diff1
value: 34.386
- type: nauc_ndcg_at_3_max
value: 30.478300000000004
- type: nauc_ndcg_at_3_std
value: -0.42189999999999994
- type: nauc_ndcg_at_3_diff1
value: 28.220699999999997
- type: nauc_ndcg_at_5_max
value: 28.219699999999996
- type: nauc_ndcg_at_5_std
value: -1.0019
- type: nauc_ndcg_at_5_diff1
value: 27.2105
- type: nauc_ndcg_at_10_max
value: 30.467100000000002
- type: nauc_ndcg_at_10_std
value: 0.0898
- type: nauc_ndcg_at_10_diff1
value: 27.1735
- type: nauc_ndcg_at_20_max
value: 31.635400000000004
- type: nauc_ndcg_at_20_std
value: 1.0711
- type: nauc_ndcg_at_20_diff1
value: 27.1711
- type: nauc_ndcg_at_100_max
value: 31.730000000000004
- type: nauc_ndcg_at_100_std
value: 2.5065
- type: nauc_ndcg_at_100_diff1
value: 26.785700000000002
- type: nauc_ndcg_at_1000_max
value: 32.5146
- type: nauc_ndcg_at_1000_std
value: 2.1953
- type: nauc_ndcg_at_1000_diff1
value: 27.626299999999997
- type: nauc_map_at_1_max
value: 20.5785
- type: nauc_map_at_1_std
value: 0.1734
- type: nauc_map_at_1_diff1
value: 33.5835
- type: nauc_map_at_3_max
value: 27.1963
- type: nauc_map_at_3_std
value: -1.038
- type: nauc_map_at_3_diff1
value: 29.028399999999998
- type: nauc_map_at_5_max
value: 28.5489
- type: nauc_map_at_5_std
value: -1.4671999999999998
- type: nauc_map_at_5_diff1
value: 28.2777
- type: nauc_map_at_10_max
value: 30.2132
- type: nauc_map_at_10_std
value: -0.984
- type: nauc_map_at_10_diff1
value: 28.527
- type: nauc_map_at_20_max
value: 30.8029
- type: nauc_map_at_20_std
value: -0.6748
- type: nauc_map_at_20_diff1
value: 28.4974
- type: nauc_map_at_100_max
value: 30.868000000000002
- type: nauc_map_at_100_std
value: -0.4051
- type: nauc_map_at_100_diff1
value: 28.348000000000003
- type: nauc_map_at_1000_max
value: 30.9483
- type: nauc_map_at_1000_std
value: -0.3498
- type: nauc_map_at_1000_diff1
value: 28.407799999999998
- type: nauc_recall_at_1_max
value: 20.5785
- type: nauc_recall_at_1_std
value: 0.1734
- type: nauc_recall_at_1_diff1
value: 33.5835
- type: nauc_recall_at_3_max
value: 22.6433
- type: nauc_recall_at_3_std
value: -1.0766
- type: nauc_recall_at_3_diff1
value: 24.5419
- type: nauc_recall_at_5_max
value: 21.1675
- type: nauc_recall_at_5_std
value: -1.6594000000000002
- type: nauc_recall_at_5_diff1
value: 20.7746
- type: nauc_recall_at_10_max
value: 25.8163
- type: nauc_recall_at_10_std
value: 1.4134
- type: nauc_recall_at_10_diff1
value: 20.0466
- type: nauc_recall_at_20_max
value: 28.211000000000002
- type: nauc_recall_at_20_std
value: 4.3018
- type: nauc_recall_at_20_diff1
value: 19.7529
- type: nauc_recall_at_100_max
value: 28.4752
- type: nauc_recall_at_100_std
value: 13.855300000000002
- type: nauc_recall_at_100_diff1
value: 15.8335
- type: nauc_recall_at_1000_max
value: 56.1762
- type: nauc_recall_at_1000_std
value: 40.7642
- type: nauc_recall_at_1000_diff1
value: 7.8241000000000005
- type: nauc_precision_at_1_max
value: 37.0131
- type: nauc_precision_at_1_std
value: 1.23
- type: nauc_precision_at_1_diff1
value: 34.386
- type: nauc_precision_at_3_max
value: 37.2799
- type: nauc_precision_at_3_std
value: 0.3125
- type: nauc_precision_at_3_diff1
value: 22.5924
- type: nauc_precision_at_5_max
value: 36.275200000000005
- type: nauc_precision_at_5_std
value: -0.4414
- type: nauc_precision_at_5_diff1
value: 20.1792
- type: nauc_precision_at_10_max
value: 36.3329
- type: nauc_precision_at_10_std
value: 0.7673
- type: nauc_precision_at_10_diff1
value: 18.4001
- type: nauc_precision_at_20_max
value: 36.1432
- type: nauc_precision_at_20_std
value: 2.7744
- type: nauc_precision_at_20_diff1
value: 15.949399999999999
- type: nauc_precision_at_100_max
value: 29.2087
- type: nauc_precision_at_100_std
value: 5.795
- type: nauc_precision_at_100_diff1
value: 9.8339
- type: nauc_precision_at_1000_max
value: 25.1923
- type: nauc_precision_at_1000_std
value: 4.9289
- type: nauc_precision_at_1000_diff1
value: 5.8301
- type: nauc_mrr_at_1_max
value: 37.0131
- type: nauc_mrr_at_1_std
value: 1.23
- type: nauc_mrr_at_1_diff1
value: 34.386
- type: nauc_mrr_at_3_max
value: 32.9506
- type: nauc_mrr_at_3_std
value: 1.0282
- type: nauc_mrr_at_3_diff1
value: 31.368000000000002
- type: nauc_mrr_at_5_max
value: 32.4437
- type: nauc_mrr_at_5_std
value: 0.8541
- type: nauc_mrr_at_5_diff1
value: 30.3286
- type: nauc_mrr_at_10_max
value: 32.9949
- type: nauc_mrr_at_10_std
value: 1.1716
- type: nauc_mrr_at_10_diff1
value: 30.272900000000003
- type: nauc_mrr_at_20_max
value: 33.1598
- type: nauc_mrr_at_20_std
value: 1.4285
- type: nauc_mrr_at_20_diff1
value: 30.3452
- type: nauc_mrr_at_100_max
value: 33.1941
- type: nauc_mrr_at_100_std
value: 1.5522
- type: nauc_mrr_at_100_diff1
value: 30.411899999999996
- type: nauc_mrr_at_1000_max
value: 33.218599999999995
- type: nauc_mrr_at_1000_std
value: 1.5448
- type: nauc_mrr_at_1000_diff1
value: 30.4433
- type: main_score
value: 33.564
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (eng-ara)
type: jinaai/xpqa
config: eng-ara
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 5.6000000000000005
- type: ndcg_at_3
value: 6.115
- type: ndcg_at_5
value: 6.412
- type: ndcg_at_10
value: 8.06
- type: ndcg_at_20
value: 9.904
- type: ndcg_at_100
value: 13.441
- type: ndcg_at_1000
value: 21.157999999999998
- type: map_at_1
value: 2.858
- type: map_at_3
value: 4.5760000000000005
- type: map_at_5
value: 5.008
- type: map_at_10
value: 5.769
- type: map_at_20
value: 6.32
- type: map_at_100
value: 6.84
- type: map_at_1000
value: 7.114
- type: recall_at_1
value: 2.858
- type: recall_at_3
value: 6.262
- type: recall_at_5
value: 7.558
- type: recall_at_10
value: 11.600000000000001
- type: recall_at_20
value: 17.843999999999998
- type: recall_at_100
value: 33.924
- type: recall_at_1000
value: 88.14
- type: precision_at_1
value: 5.6000000000000005
- type: precision_at_3
value: 4.133
- type: precision_at_5
value: 3.2
- type: precision_at_10
value: 2.547
- type: precision_at_20
value: 1.867
- type: precision_at_100
value: 0.716
- type: precision_at_1000
value: 0.182
- type: mrr_at_1
value: 5.6000000000000005
- type: mrr_at_3
value: 7.6667
- type: mrr_at_5
value: 8.093300000000001
- type: mrr_at_10
value: 8.8209
- type: mrr_at_20
value: 9.3654
- type: mrr_at_100
value: 9.8288
- type: mrr_at_1000
value: 10.009500000000001
- type: nauc_ndcg_at_1_max
value: 32.838899999999995
- type: nauc_ndcg_at_1_std
value: 20.5796
- type: nauc_ndcg_at_1_diff1
value: 22.6813
- type: nauc_ndcg_at_3_max
value: 35.1866
- type: nauc_ndcg_at_3_std
value: 24.829
- type: nauc_ndcg_at_3_diff1
value: 20.6032
- type: nauc_ndcg_at_5_max
value: 36.8889
- type: nauc_ndcg_at_5_std
value: 27.8175
- type: nauc_ndcg_at_5_diff1
value: 18.686
- type: nauc_ndcg_at_10_max
value: 37.3493
- type: nauc_ndcg_at_10_std
value: 31.882700000000003
- type: nauc_ndcg_at_10_diff1
value: 18.4922
- type: nauc_ndcg_at_20_max
value: 37.1177
- type: nauc_ndcg_at_20_std
value: 33.9735
- type: nauc_ndcg_at_20_diff1
value: 17.1864
- type: nauc_ndcg_at_100_max
value: 34.8607
- type: nauc_ndcg_at_100_std
value: 32.9944
- type: nauc_ndcg_at_100_diff1
value: 18.2682
- type: nauc_ndcg_at_1000_max
value: 32.228899999999996
- type: nauc_ndcg_at_1000_std
value: 31.282500000000002
- type: nauc_ndcg_at_1000_diff1
value: 18.4402
- type: nauc_map_at_1_max
value: 28.424300000000002
- type: nauc_map_at_1_std
value: 18.1568
- type: nauc_map_at_1_diff1
value: 27.4362
- type: nauc_map_at_3_max
value: 34.8293
- type: nauc_map_at_3_std
value: 23.643
- type: nauc_map_at_3_diff1
value: 21.8558
- type: nauc_map_at_5_max
value: 36.3296
- type: nauc_map_at_5_std
value: 25.9859
- type: nauc_map_at_5_diff1
value: 20.552999999999997
- type: nauc_map_at_10_max
value: 37.282199999999996
- type: nauc_map_at_10_std
value: 28.8291
- type: nauc_map_at_10_diff1
value: 20.2188
- type: nauc_map_at_20_max
value: 37.366
- type: nauc_map_at_20_std
value: 30.12
- type: nauc_map_at_20_diff1
value: 19.4849
- type: nauc_map_at_100_max
value: 37.0376
- type: nauc_map_at_100_std
value: 30.318800000000003
- type: nauc_map_at_100_diff1
value: 19.8468
- type: nauc_map_at_1000_max
value: 36.9108
- type: nauc_map_at_1000_std
value: 30.303600000000003
- type: nauc_map_at_1000_diff1
value: 19.8765
- type: nauc_recall_at_1_max
value: 28.424300000000002
- type: nauc_recall_at_1_std
value: 18.1568
- type: nauc_recall_at_1_diff1
value: 27.4362
- type: nauc_recall_at_3_max
value: 35.3652
- type: nauc_recall_at_3_std
value: 26.3617
- type: nauc_recall_at_3_diff1
value: 18.121499999999997
- type: nauc_recall_at_5_max
value: 37.9415
- type: nauc_recall_at_5_std
value: 31.6361
- type: nauc_recall_at_5_diff1
value: 14.7091
- type: nauc_recall_at_10_max
value: 36.7605
- type: nauc_recall_at_10_std
value: 36.6161
- type: nauc_recall_at_10_diff1
value: 14.8281
- type: nauc_recall_at_20_max
value: 35.1301
- type: nauc_recall_at_20_std
value: 38.683800000000005
- type: nauc_recall_at_20_diff1
value: 13.0095
- type: nauc_recall_at_100_max
value: 29.624
- type: nauc_recall_at_100_std
value: 34.0362
- type: nauc_recall_at_100_diff1
value: 15.9544
- type: nauc_recall_at_1000_max
value: 13.4196
- type: nauc_recall_at_1000_std
value: 34.4493
- type: nauc_recall_at_1000_diff1
value: 13.950899999999999
- type: nauc_precision_at_1_max
value: 32.838899999999995
- type: nauc_precision_at_1_std
value: 20.5796
- type: nauc_precision_at_1_diff1
value: 22.6813
- type: nauc_precision_at_3_max
value: 40.4435
- type: nauc_precision_at_3_std
value: 27.6221
- type: nauc_precision_at_3_diff1
value: 19.8144
- type: nauc_precision_at_5_max
value: 41.9666
- type: nauc_precision_at_5_std
value: 31.5946
- type: nauc_precision_at_5_diff1
value: 16.1282
- type: nauc_precision_at_10_max
value: 39.9322
- type: nauc_precision_at_10_std
value: 36.756499999999996
- type: nauc_precision_at_10_diff1
value: 16.2153
- type: nauc_precision_at_20_max
value: 38.3678
- type: nauc_precision_at_20_std
value: 38.7305
- type: nauc_precision_at_20_diff1
value: 12.822700000000001
- type: nauc_precision_at_100_max
value: 28.3971
- type: nauc_precision_at_100_std
value: 30.848100000000002
- type: nauc_precision_at_100_diff1
value: 12.8062
- type: nauc_precision_at_1000_max
value: 2.3346999999999998
- type: nauc_precision_at_1000_std
value: 5.900799999999999
- type: nauc_precision_at_1000_diff1
value: 5.9445
- type: nauc_mrr_at_1_max
value: 32.838899999999995
- type: nauc_mrr_at_1_std
value: 20.5796
- type: nauc_mrr_at_1_diff1
value: 22.6813
- type: nauc_mrr_at_3_max
value: 34.682
- type: nauc_mrr_at_3_std
value: 22.7573
- type: nauc_mrr_at_3_diff1
value: 21.3031
- type: nauc_mrr_at_5_max
value: 35.1101
- type: nauc_mrr_at_5_std
value: 24.595200000000002
- type: nauc_mrr_at_5_diff1
value: 19.8655
- type: nauc_mrr_at_10_max
value: 34.9324
- type: nauc_mrr_at_10_std
value: 26.1953
- type: nauc_mrr_at_10_diff1
value: 19.862199999999998
- type: nauc_mrr_at_20_max
value: 34.7806
- type: nauc_mrr_at_20_std
value: 26.606999999999996
- type: nauc_mrr_at_20_diff1
value: 19.4267
- type: nauc_mrr_at_100_max
value: 34.3513
- type: nauc_mrr_at_100_std
value: 26.3405
- type: nauc_mrr_at_100_diff1
value: 19.5093
- type: nauc_mrr_at_1000_max
value: 34.3621
- type: nauc_mrr_at_1000_std
value: 26.3118
- type: nauc_mrr_at_1000_diff1
value: 19.557
- type: main_score
value: 8.06
- task:
type: Retrieval
dataset:
name: MTEB XPQARetrieval (ara-eng)
type: jinaai/xpqa
config: ara-eng
split: test
revision: c99d599f0a6ab9b85b065da6f9d94f9cf731679f
metrics:
- type: ndcg_at_1
value: 4.717
- type: ndcg_at_3
value: 6.136
- type: ndcg_at_5
value: 6.796
- type: ndcg_at_10
value: 8.417
- type: ndcg_at_20
value: 10.041
- type: ndcg_at_100
value: 13.668
- type: ndcg_at_1000
value: 21.077
- type: map_at_1
value: 2.3810000000000002
- type: map_at_3
value: 4.62
- type: map_at_5
value: 5.285
- type: map_at_10
value: 6.115
- type: map_at_20
value: 6.605999999999999
- type: map_at_100
value: 7.173
- type: map_at_1000
value: 7.424
- type: recall_at_1
value: 2.3810000000000002
- type: recall_at_3
value: 6.611000000000001
- type: recall_at_5
value: 8.643
- type: recall_at_10
value: 12.873000000000001
- type: recall_at_20
value: 18.358
- type: recall_at_100
value: 35.274
- type: recall_at_1000
value: 87.25699999999999
- type: precision_at_1
value: 4.717
- type: precision_at_3
value: 4.717
- type: precision_at_5
value: 3.7740000000000005
- type: precision_at_10
value: 2.709
- type: precision_at_20
value: 1.8800000000000001
- type: precision_at_100
value: 0.697
- type: precision_at_1000
value: 0.17700000000000002
- type: mrr_at_1
value: 4.717
- type: mrr_at_3
value: 6.9407
- type: mrr_at_5
value: 7.5066999999999995
- type: mrr_at_10
value: 8.0793
- type: mrr_at_20
value: 8.5387
- type: mrr_at_100
value: 8.9732
- type: mrr_at_1000
value: 9.1562
- type: nauc_ndcg_at_1_max
value: 54.243300000000005
- type: nauc_ndcg_at_1_std
value: 25.9453
- type: nauc_ndcg_at_1_diff1
value: 39.2959
- type: nauc_ndcg_at_3_max
value: 42.9191
- type: nauc_ndcg_at_3_std
value: 20.4861
- type: nauc_ndcg_at_3_diff1
value: 25.1422
- type: nauc_ndcg_at_5_max
value: 38.6922
- type: nauc_ndcg_at_5_std
value: 20.5677
- type: nauc_ndcg_at_5_diff1
value: 21.3885
- type: nauc_ndcg_at_10_max
value: 36.5826
- type: nauc_ndcg_at_10_std
value: 20.7746
- type: nauc_ndcg_at_10_diff1
value: 18.6611
- type: nauc_ndcg_at_20_max
value: 35.204299999999996
- type: nauc_ndcg_at_20_std
value: 21.1932
- type: nauc_ndcg_at_20_diff1
value: 17.1578
- type: nauc_ndcg_at_100_max
value: 32.2066
- type: nauc_ndcg_at_100_std
value: 22.0766
- type: nauc_ndcg_at_100_diff1
value: 13.971
- type: nauc_ndcg_at_1000_max
value: 33.6484
- type: nauc_ndcg_at_1000_std
value: 22.9162
- type: nauc_ndcg_at_1000_diff1
value: 14.0986
- type: nauc_map_at_1_max
value: 40.3701
- type: nauc_map_at_1_std
value: 16.161900000000003
- type: nauc_map_at_1_diff1
value: 39.9372
- type: nauc_map_at_3_max
value: 41.3994
- type: nauc_map_at_3_std
value: 19.808400000000002
- type: nauc_map_at_3_diff1
value: 27.0159
- type: nauc_map_at_5_max
value: 39.7394
- type: nauc_map_at_5_std
value: 19.3577
- type: nauc_map_at_5_diff1
value: 25.1608
- type: nauc_map_at_10_max
value: 39.2515
- type: nauc_map_at_10_std
value: 20.1689
- type: nauc_map_at_10_diff1
value: 22.7535
- type: nauc_map_at_20_max
value: 38.8313
- type: nauc_map_at_20_std
value: 20.5593
- type: nauc_map_at_20_diff1
value: 21.933600000000002
- type: nauc_map_at_100_max
value: 38.0329
- type: nauc_map_at_100_std
value: 20.7943
- type: nauc_map_at_100_diff1
value: 20.9206
- type: nauc_map_at_1000_max
value: 38.0858
- type: nauc_map_at_1000_std
value: 20.8558
- type: nauc_map_at_1000_diff1
value: 20.887700000000002
- type: nauc_recall_at_1_max
value: 40.3701
- type: nauc_recall_at_1_std
value: 16.161900000000003
- type: nauc_recall_at_1_diff1
value: 39.9372
- type: nauc_recall_at_3_max
value: 36.5375
- type: nauc_recall_at_3_std
value: 18.166
- type: nauc_recall_at_3_diff1
value: 18.7422
- type: nauc_recall_at_5_max
value: 32.6016
- type: nauc_recall_at_5_std
value: 18.378700000000002
- type: nauc_recall_at_5_diff1
value: 15.2924
- type: nauc_recall_at_10_max
value: 28.719299999999997
- type: nauc_recall_at_10_std
value: 18.121499999999997
- type: nauc_recall_at_10_diff1
value: 12.0404
- type: nauc_recall_at_20_max
value: 27.1826
- type: nauc_recall_at_20_std
value: 19.482499999999998
- type: nauc_recall_at_20_diff1
value: 11.1159
- type: nauc_recall_at_100_max
value: 21.4272
- type: nauc_recall_at_100_std
value: 21.723200000000002
- type: nauc_recall_at_100_diff1
value: 4.9525
- type: nauc_recall_at_1000_max
value: 24.616699999999998
- type: nauc_recall_at_1000_std
value: 36.6124
- type: nauc_recall_at_1000_diff1
value: -1.4559
- type: nauc_precision_at_1_max
value: 54.243300000000005
- type: nauc_precision_at_1_std
value: 25.9453
- type: nauc_precision_at_1_diff1
value: 39.2959
- type: nauc_precision_at_3_max
value: 48.6299
- type: nauc_precision_at_3_std
value: 24.9782
- type: nauc_precision_at_3_diff1
value: 23.6147
- type: nauc_precision_at_5_max
value: 43.9644
- type: nauc_precision_at_5_std
value: 23.6441
- type: nauc_precision_at_5_diff1
value: 20.3201
- type: nauc_precision_at_10_max
value: 41.4126
- type: nauc_precision_at_10_std
value: 24.6059
- type: nauc_precision_at_10_diff1
value: 16.0803
- type: nauc_precision_at_20_max
value: 37.7543
- type: nauc_precision_at_20_std
value: 23.7518
- type: nauc_precision_at_20_diff1
value: 11.8993
- type: nauc_precision_at_100_max
value: 28.8901
- type: nauc_precision_at_100_std
value: 21.9506
- type: nauc_precision_at_100_diff1
value: 7.3769
- type: nauc_precision_at_1000_max
value: 12.132900000000001
- type: nauc_precision_at_1000_std
value: 8.134
- type: nauc_precision_at_1000_diff1
value: -1.0386
- type: nauc_mrr_at_1_max
value: 54.243300000000005
- type: nauc_mrr_at_1_std
value: 25.9453
- type: nauc_mrr_at_1_diff1
value: 39.2959
- type: nauc_mrr_at_3_max
value: 45.3324
- type: nauc_mrr_at_3_std
value: 23.9364
- type: nauc_mrr_at_3_diff1
value: 25.5843
- type: nauc_mrr_at_5_max
value: 43.5379
- type: nauc_mrr_at_5_std
value: 23.9876
- type: nauc_mrr_at_5_diff1
value: 24.0945
- type: nauc_mrr_at_10_max
value: 41.2615
- type: nauc_mrr_at_10_std
value: 23.1665
- type: nauc_mrr_at_10_diff1
value: 22.6914
- type: nauc_mrr_at_20_max
value: 40.3956
- type: nauc_mrr_at_20_std
value: 22.9236
- type: nauc_mrr_at_20_diff1
value: 22.037399999999998
- type: nauc_mrr_at_100_max
value: 39.8172
- type: nauc_mrr_at_100_std
value: 23.0539
- type: nauc_mrr_at_100_diff1
value: 21.4238
- type: nauc_mrr_at_1000_max
value: 39.9549
- type: nauc_mrr_at_1000_std
value: 23.125999999999998
- type: nauc_mrr_at_1000_diff1
value: 21.4921
- type: main_score
value: 8.417
- task:
type: STS
dataset:
name: MTEB BIOSSES (default)
type: mteb/biosses-sts
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cosine_pearson
value: 67.88078975738149
- type: cosine_spearman
value: 67.36900492799694
- type: euclidean_pearson
value: 66.00402957388015
- type: euclidean_spearman
value: 65.70270189991112
- type: main_score
value: 67.36900492799694
- type: manhattan_pearson
value: 66.54937895501651
- type: manhattan_spearman
value: 66.12198856207587
- task:
type: STS
dataset:
name: MTEB SICK-R (default)
type: mteb/sickr-sts
config: default
split: test
revision: 20a6d6f312dd54037fe07a32d58e5e168867909d
metrics:
- type: cosine_pearson
value: 62.931439439697044
- type: cosine_spearman
value: 57.64441663261227
- type: euclidean_pearson
value: 61.119408834167835
- type: euclidean_spearman
value: 57.42332323654558
- type: main_score
value: 57.64441663261227
- type: manhattan_pearson
value: 60.692516462749204
- type: manhattan_spearman
value: 56.99349446063643
- task:
type: STS
dataset:
name: MTEB STS12 (default)
type: mteb/sts12-sts
config: default
split: test
revision: a0d554a64d88156834ff5ae9920b964011b16384
metrics:
- type: cosine_pearson
value: 70.42631404785132
- type: cosine_spearman
value: 69.67060431422327
- type: euclidean_pearson
value: 68.70261457119209
- type: euclidean_spearman
value: 68.99597672902992
- type: main_score
value: 69.67060431422327
- type: manhattan_pearson
value: 67.99048393745159
- type: manhattan_spearman
value: 68.1853179140009
- task:
type: STS
dataset:
name: MTEB STS13 (default)
type: mteb/sts13-sts
config: default
split: test
revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
metrics:
- type: cosine_pearson
value: 49.46916157874787
- type: cosine_spearman
value: 51.95037157769884
- type: euclidean_pearson
value: 55.17336596392549
- type: euclidean_spearman
value: 54.312304378478835
- type: main_score
value: 51.95037157769884
- type: manhattan_pearson
value: 55.09060773902408
- type: manhattan_spearman
value: 53.96813218977611
- task:
type: STS
dataset:
name: MTEB STS14 (default)
type: mteb/sts14-sts
config: default
split: test
revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
metrics:
- type: cosine_pearson
value: 54.37699141667456
- type: cosine_spearman
value: 57.36607721958864
- type: euclidean_pearson
value: 57.98000825695592
- type: euclidean_spearman
value: 59.08844527739818
- type: main_score
value: 57.36607721958864
- type: manhattan_pearson
value: 57.588062173142106
- type: manhattan_spearman
value: 58.35590953779109
- task:
type: STS
dataset:
name: MTEB STS15 (default)
type: mteb/sts15-sts
config: default
split: test
revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
metrics:
- type: cosine_pearson
value: 67.37948361289261
- type: cosine_spearman
value: 70.0994395240558
- type: euclidean_pearson
value: 70.28341277052768
- type: euclidean_spearman
value: 70.11050982217422
- type: main_score
value: 70.0994395240558
- type: manhattan_pearson
value: 70.66000566140171
- type: manhattan_spearman
value: 70.41742785288693
- task:
type: STS
dataset:
name: MTEB STS16 (default)
type: mteb/sts16-sts
config: default
split: test
revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
metrics:
- type: cosine_pearson
value: 61.559501698409434
- type: cosine_spearman
value: 65.04903130808405
- type: euclidean_pearson
value: 63.92021058086694
- type: euclidean_spearman
value: 64.22673046991633
- type: main_score
value: 65.04903130808405
- type: manhattan_pearson
value: 63.958100692077956
- type: manhattan_spearman
value: 64.15057001708075
- task:
type: STS
dataset:
name: MTEB STS17 (ar-ar)
type: mteb/sts17-crosslingual-sts
config: ar-ar
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 82.35377320218275
- type: cosine_spearman
value: 83.15514468203664
- type: euclidean_pearson
value: 80.56116685008965
- type: euclidean_spearman
value: 82.38252301503367
- type: main_score
value: 83.15514468203664
- type: manhattan_pearson
value: 80.74794586574093
- type: manhattan_spearman
value: 82.54224799581789
- task:
type: STS
dataset:
name: MTEB STS22 (ar)
type: mteb/sts22-crosslingual-sts
config: ar
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 48.22154847597003
- type: cosine_spearman
value: 58.29235719729918
- type: euclidean_pearson
value: 51.54481297728728
- type: euclidean_spearman
value: 58.990627664376674
- type: main_score
value: 58.29235719729918
- type: manhattan_pearson
value: 52.195039627338126
- type: manhattan_spearman
value: 59.12018922641005
- task:
type: STS
dataset:
name: MTEB STSBenchmark (default)
type: mteb/stsbenchmark-sts
config: default
split: test
revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831
metrics:
- type: cosine_pearson
value: 59.50286436994106
- type: cosine_spearman
value: 61.592426810014366
- type: euclidean_pearson
value: 63.268627193788916
- type: euclidean_spearman
value: 63.16239630067321
- type: main_score
value: 61.592426810014366
- type: manhattan_pearson
value: 62.95949714767757
- type: manhattan_spearman
value: 62.687737378385364
- task:
type: Summarization
dataset:
name: MTEB SummEval (default)
type: mteb/summeval
config: default
split: test
revision: cda12ad7615edc362dbf25a00fdd61d3b1eaf93c
metrics:
- type: cosine_pearson
value: 31.1427099547469
- type: cosine_spearman
value: 31.32880594576111
- type: dot_pearson
value: 25.98395652985614
- type: dot_spearman
value: 25.30831374828529
- type: main_score
value: 31.32880594576111
- type: pearson
value: 31.1427099547469
- type: spearman
value: 31.32880594576111
- task:
type: semantic-similarity
name: Semantic Similarity
dataset:
name: sts test 768
type: sts-test-768
metrics:
- type: pearson_cosine
value: 0.5949906740977448
name: Pearson Cosine
- type: spearman_cosine
value: 0.6159750250469712
name: Spearman Cosine
- type: pearson_manhattan
value: 0.6295622269205102
name: Pearson Manhattan
- type: spearman_manhattan
value: 0.6269654283099967
name: Spearman Manhattan
- type: pearson_euclidean
value: 0.6326526932327604
name: Pearson Euclidean
- type: spearman_euclidean
value: 0.6317081341785673
name: Spearman Euclidean
- type: pearson_dot
value: 0.42816790752358297
name: Pearson Dot
- type: spearman_dot
value: 0.4295282086669423
name: Spearman Dot
- type: pearson_max
value: 0.6326526932327604
name: Pearson Max
- type: spearman_max
value: 0.6317081341785673
name: Spearman Max
- task:
type: semantic-similarity
name: Semantic Similarity
dataset:
name: sts test 512
type: sts-test-512
metrics:
- type: pearson_cosine
value: 0.5846223235167534
name: Pearson Cosine
- type: spearman_cosine
value: 0.6064092420664184
name: Spearman Cosine
- type: pearson_manhattan
value: 0.6287774004727389
name: Pearson Manhattan
- type: spearman_manhattan
value: 0.6263546541183983
name: Spearman Manhattan
- type: pearson_euclidean
value: 0.631267664308041
name: Pearson Euclidean
- type: spearman_euclidean
value: 0.6301778108727977
name: Spearman Euclidean
- type: pearson_dot
value: 0.3788565672017437
name: Pearson Dot
- type: spearman_dot
value: 0.37680551461721923
name: Spearman Dot
- type: pearson_max
value: 0.631267664308041
name: Pearson Max
- type: spearman_max
value: 0.6301778108727977
name: Spearman Max
- task:
type: semantic-similarity
name: Semantic Similarity
dataset:
name: sts test 256
type: sts-test-256
metrics:
- type: pearson_cosine
value: 0.5778623383989389
name: Pearson Cosine
- type: spearman_cosine
value: 0.5959667709300495
name: Spearman Cosine
- type: pearson_manhattan
value: 0.6242980982402613
name: Pearson Manhattan
- type: spearman_manhattan
value: 0.6217473192873829
name: Spearman Manhattan
- type: pearson_euclidean
value: 0.6237908608463304
name: Pearson Euclidean
- type: spearman_euclidean
value: 0.6215304658549996
name: Spearman Euclidean
- type: pearson_dot
value: 0.35968442092444003
name: Pearson Dot
- type: spearman_dot
value: 0.35304547874806785
name: Spearman Dot
- type: pearson_max
value: 0.6242980982402613
name: Pearson Max
- type: spearman_max
value: 0.6217473192873829
name: Spearman Max
- task:
type: semantic-similarity
name: Semantic Similarity
dataset:
name: sts test 128
type: sts-test-128
metrics:
- type: pearson_cosine
value: 0.5830782075122916
name: Pearson Cosine
- type: spearman_cosine
value: 0.6022044167653756
name: Spearman Cosine
- type: pearson_manhattan
value: 0.6151866925343435
name: Pearson Manhattan
- type: spearman_manhattan
value: 0.6121950064533626
name: Spearman Manhattan
- type: pearson_euclidean
value: 0.6162225316000448
name: Pearson Euclidean
- type: spearman_euclidean
value: 0.615301209345362
name: Spearman Euclidean
- type: pearson_dot
value: 0.40438461342780957
name: Pearson Dot
- type: spearman_dot
value: 0.40153111017443666
name: Spearman Dot
- type: pearson_max
value: 0.6162225316000448
name: Pearson Max
- type: spearman_max
value: 0.615301209345362
name: Spearman Max
- task:
type: semantic-similarity
name: Semantic Similarity
dataset:
name: sts test 64
type: sts-test-64
metrics:
- type: pearson_cosine
value: 0.5724838823862283
name: Pearson Cosine
- type: spearman_cosine
value: 0.5914127847098
name: Spearman Cosine
- type: pearson_manhattan
value: 0.6023812283389073
name: Pearson Manhattan
- type: spearman_manhattan
value: 0.5967205030284914
name: Spearman Manhattan
- type: pearson_euclidean
value: 0.6069294574719372
name: Pearson Euclidean
- type: spearman_euclidean
value: 0.6041440553344074
name: Spearman Euclidean
- type: pearson_dot
value: 0.36315938245739166
name: Pearson Dot
- type: spearman_dot
value: 0.358512645020771
name: Spearman Dot
- type: pearson_max
value: 0.6069294574719372
name: Pearson Max
- type: spearman_max
value: 0.6041440553344074
name: Spearman Max
---
# Arabert All NLI Triplet Matryoshka Model
This is a [sentence-transformers](https://www.SBERT.net) model finetuned from [aubmindlab/bert-base-arabertv02](https://huggingface.co/aubmindlab/bert-base-arabertv02) on the Omartificial-Intelligence-Space/arabic-n_li-triplet dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.
## Model Details
### Model Description
- **Model Type:** Sentence Transformer
- **Base model:** [aubmindlab/bert-base-arabertv02](https://huggingface.co/aubmindlab/bert-base-arabertv02) <!-- at revision 016fb9d6768f522a59c6e0d2d5d5d43a4e1bff60 -->
- **Maximum Sequence Length:** 512 tokens
- **Output Dimensionality:** 768 tokens
- **Similarity Function:** Cosine Similarity
- **Training Dataset:**
- Omartificial-Intelligence-Space/arabic-n_li-triplet
<!-- - **Language:** Unknown -->
<!-- - **License:** Unknown -->
### Model Sources
- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
- **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers)
- **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers)
### Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)
```
## Usage
### Direct Usage (Sentence Transformers)
First install the Sentence Transformers library:
```bash
pip install -U sentence-transformers
```
Then you can load this model and run inference.
```python
from sentence_transformers import SentenceTransformer
# Download from the 🤗 Hub
model = SentenceTransformer("Omartificial-Intelligence-Space/Arabic-arabert-all-nli-triplet")
# Run inference
sentences = [
'يجلس شاب ذو شعر أشقر على الحائط يقرأ جريدة بينما تمر امرأة وفتاة شابة.',
'ذكر شاب ينظر إلى جريدة بينما تمر إمرأتان بجانبه',
'الشاب نائم بينما الأم تقود ابنتها إلى الحديقة',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]
# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]
```
<!--
### Direct Usage (Transformers)
<details><summary>Click to see the direct usage in Transformers</summary>
</details>
-->
<!--
### Downstream Usage (Sentence Transformers)
You can finetune this model on your own dataset.
<details><summary>Click to expand</summary>
</details>
-->
<!--
### Out-of-Scope Use
*List how the model may foreseeably be misused and address what users ought not to do with the model.*
-->
## Evaluation
### Metrics
#### Semantic Similarity
* Dataset: `sts-test-768`
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)
| Metric | Value |
|:--------------------|:----------|
| pearson_cosine | 0.595 |
| **spearman_cosine** | **0.616** |
| pearson_manhattan | 0.6296 |
| spearman_manhattan | 0.627 |
| pearson_euclidean | 0.6327 |
| spearman_euclidean | 0.6317 |
| pearson_dot | 0.4282 |
| spearman_dot | 0.4295 |
| pearson_max | 0.6327 |
| spearman_max | 0.6317 |
#### Semantic Similarity
* Dataset: `sts-test-512`
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)
| Metric | Value |
|:--------------------|:-----------|
| pearson_cosine | 0.5846 |
| **spearman_cosine** | **0.6064** |
| pearson_manhattan | 0.6288 |
| spearman_manhattan | 0.6264 |
| pearson_euclidean | 0.6313 |
| spearman_euclidean | 0.6302 |
| pearson_dot | 0.3789 |
| spearman_dot | 0.3768 |
| pearson_max | 0.6313 |
| spearman_max | 0.6302 |
#### Semantic Similarity
* Dataset: `sts-test-256`
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)
| Metric | Value |
|:--------------------|:----------|
| pearson_cosine | 0.5779 |
| **spearman_cosine** | **0.596** |
| pearson_manhattan | 0.6243 |
| spearman_manhattan | 0.6217 |
| pearson_euclidean | 0.6238 |
| spearman_euclidean | 0.6215 |
| pearson_dot | 0.3597 |
| spearman_dot | 0.353 |
| pearson_max | 0.6243 |
| spearman_max | 0.6217 |
#### Semantic Similarity
* Dataset: `sts-test-128`
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)
| Metric | Value |
|:--------------------|:-----------|
| pearson_cosine | 0.5831 |
| **spearman_cosine** | **0.6022** |
| pearson_manhattan | 0.6152 |
| spearman_manhattan | 0.6122 |
| pearson_euclidean | 0.6162 |
| spearman_euclidean | 0.6153 |
| pearson_dot | 0.4044 |
| spearman_dot | 0.4015 |
| pearson_max | 0.6162 |
| spearman_max | 0.6153 |
#### Semantic Similarity
* Dataset: `sts-test-64`
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)
| Metric | Value |
|:--------------------|:-----------|
| pearson_cosine | 0.5725 |
| **spearman_cosine** | **0.5914** |
| pearson_manhattan | 0.6024 |
| spearman_manhattan | 0.5967 |
| pearson_euclidean | 0.6069 |
| spearman_euclidean | 0.6041 |
| pearson_dot | 0.3632 |
| spearman_dot | 0.3585 |
| pearson_max | 0.6069 |
| spearman_max | 0.6041 |
<!--
## Bias, Risks and Limitations
*What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.*
-->
<!--
### Recommendations
*What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.*
-->
## Training Details
### Training Dataset
#### Omartificial-Intelligence-Space/arabic-n_li-triplet
* Dataset: Omartificial-Intelligence-Space/arabic-n_li-triplet
* Size: 557,850 training samples
* Columns: <code>anchor</code>, <code>positive</code>, and <code>negative</code>
* Approximate statistics based on the first 1000 samples:
| | anchor | positive | negative |
|:--------|:---------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|
| type | string | string | string |
| details | <ul><li>min: 4 tokens</li><li>mean: 8.02 tokens</li><li>max: 41 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 10.03 tokens</li><li>max: 34 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 10.72 tokens</li><li>max: 38 tokens</li></ul> |
* Samples:
| anchor | positive | negative |
|:------------------------------------------------------------|:--------------------------------------------|:------------------------------------|
| <code>شخص على حصان يقفز فوق طائرة معطلة</code> | <code>شخص في الهواء الطلق، على حصان.</code> | <code>شخص في مطعم، يطلب عجة.</code> |
| <code>أطفال يبتسمون و يلوحون للكاميرا</code> | <code>هناك أطفال حاضرون</code> | <code>الاطفال يتجهمون</code> |
| <code>صبي يقفز على لوح التزلج في منتصف الجسر الأحمر.</code> | <code>الفتى يقوم بخدعة التزلج</code> | <code>الصبي يتزلج على الرصيف</code> |
* Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:
```json
{
"loss": "MultipleNegativesRankingLoss",
"matryoshka_dims": [
768,
512,
256,
128,
64
],
"matryoshka_weights": [
1,
1,
1,
1,
1
],
"n_dims_per_step": -1
}
```
### Evaluation Dataset
#### Omartificial-Intelligence-Space/arabic-n_li-triplet
* Dataset: Omartificial-Intelligence-Space/arabic-n_li-triplet
* Size: 6,584 evaluation samples
* Columns: <code>anchor</code>, <code>positive</code>, and <code>negative</code>
* Approximate statistics based on the first 1000 samples:
| | anchor | positive | negative |
|:--------|:----------------------------------------------------------------------------------|:---------------------------------------------------------------------------------|:---------------------------------------------------------------------------------|
| type | string | string | string |
| details | <ul><li>min: 4 tokens</li><li>mean: 14.87 tokens</li><li>max: 70 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 7.54 tokens</li><li>max: 26 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 8.14 tokens</li><li>max: 23 tokens</li></ul> |
* Samples:
| anchor | positive | negative |
|:-----------------------------------------------------------------------------------------------------------------------------------------------------|:-------------------------------------------------------|:---------------------------------------------------|
| <code>امرأتان يتعانقان بينما يحملان حزمة</code> | <code>إمرأتان يحملان حزمة</code> | <code>الرجال يتشاجرون خارج مطعم</code> |
| <code>طفلين صغيرين يرتديان قميصاً أزرق، أحدهما يرتدي الرقم 9 والآخر يرتدي الرقم 2 يقفان على خطوات خشبية في الحمام ويغسلان أيديهما في المغسلة.</code> | <code>طفلين يرتديان قميصاً مرقماً يغسلون أيديهم</code> | <code>طفلين يرتديان سترة يذهبان إلى المدرسة</code> |
| <code>رجل يبيع الدونات لعميل خلال معرض عالمي أقيم في مدينة أنجليس</code> | <code>رجل يبيع الدونات لعميل</code> | <code>امرأة تشرب قهوتها في مقهى صغير</code> |
* Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:
```json
{
"loss": "MultipleNegativesRankingLoss",
"matryoshka_dims": [
768,
512,
256,
128,
64
],
"matryoshka_weights": [
1,
1,
1,
1,
1
],
"n_dims_per_step": -1
}
```
### Training Hyperparameters
#### Non-Default Hyperparameters
- `per_device_train_batch_size`: 64
- `per_device_eval_batch_size`: 64
- `num_train_epochs`: 1
- `warmup_ratio`: 0.1
- `fp16`: True
- `batch_sampler`: no_duplicates
#### All Hyperparameters
<details><summary>Click to expand</summary>
- `overwrite_output_dir`: False
- `do_predict`: False
- `prediction_loss_only`: True
- `per_device_train_batch_size`: 64
- `per_device_eval_batch_size`: 64
- `per_gpu_train_batch_size`: None
- `per_gpu_eval_batch_size`: None
- `gradient_accumulation_steps`: 1
- `eval_accumulation_steps`: None
- `learning_rate`: 5e-05
- `weight_decay`: 0.0
- `adam_beta1`: 0.9
- `adam_beta2`: 0.999
- `adam_epsilon`: 1e-08
- `max_grad_norm`: 1.0
- `num_train_epochs`: 1
- `max_steps`: -1
- `lr_scheduler_type`: linear
- `lr_scheduler_kwargs`: {}
- `warmup_ratio`: 0.1
- `warmup_steps`: 0
- `log_level`: passive
- `log_level_replica`: warning
- `log_on_each_node`: True
- `logging_nan_inf_filter`: True
- `save_safetensors`: True
- `save_on_each_node`: False
- `save_only_model`: False
- `no_cuda`: False
- `use_cpu`: False
- `use_mps_device`: False
- `seed`: 42
- `data_seed`: None
- `jit_mode_eval`: False
- `use_ipex`: False
- `bf16`: False
- `fp16`: True
- `fp16_opt_level`: O1
- `half_precision_backend`: auto
- `bf16_full_eval`: False
- `fp16_full_eval`: False
- `tf32`: None
- `local_rank`: 0
- `ddp_backend`: None
- `tpu_num_cores`: None
- `tpu_metrics_debug`: False
- `debug`: []
- `dataloader_drop_last`: False
- `dataloader_num_workers`: 0
- `dataloader_prefetch_factor`: None
- `past_index`: -1
- `disable_tqdm`: False
- `remove_unused_columns`: True
- `label_names`: None
- `load_best_model_at_end`: False
- `ignore_data_skip`: False
- `fsdp`: []
- `fsdp_min_num_params`: 0
- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
- `fsdp_transformer_layer_cls_to_wrap`: None
- `accelerator_config`: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'gradient_accumulation_kwargs': None}
- `deepspeed`: None
- `label_smoothing_factor`: 0.0
- `optim`: adamw_torch
- `optim_args`: None
- `adafactor`: False
- `group_by_length`: False
- `length_column_name`: length
- `ddp_find_unused_parameters`: None
- `ddp_bucket_cap_mb`: None
- `ddp_broadcast_buffers`: False
- `dataloader_pin_memory`: True
- `dataloader_persistent_workers`: False
- `skip_memory_metrics`: True
- `use_legacy_prediction_loop`: False
- `push_to_hub`: False
- `resume_from_checkpoint`: None
- `hub_model_id`: None
- `hub_strategy`: every_save
- `hub_private_repo`: False
- `hub_always_push`: False
- `gradient_checkpointing`: False
- `gradient_checkpointing_kwargs`: None
- `include_inputs_for_metrics`: False
- `eval_do_concat_batches`: True
- `fp16_backend`: auto
- `push_to_hub_model_id`: None
- `push_to_hub_organization`: None
- `mp_parameters`:
- `auto_find_batch_size`: False
- `full_determinism`: False
- `torchdynamo`: None
- `ray_scope`: last
- `ddp_timeout`: 1800
- `torch_compile`: False
- `torch_compile_backend`: None
- `torch_compile_mode`: None
- `dispatch_batches`: None
- `split_batches`: None
- `include_tokens_per_second`: False
- `include_num_input_tokens_seen`: False
- `neftune_noise_alpha`: None
- `optim_target_modules`: None
- `batch_sampler`: no_duplicates
- `multi_dataset_batch_sampler`: proportional
</details>
### Training Logs
| Epoch | Step | Training Loss | sts-test-128_spearman_cosine | sts-test-256_spearman_cosine | sts-test-512_spearman_cosine | sts-test-64_spearman_cosine | sts-test-768_spearman_cosine |
|:------:|:----:|:-------------:|:----------------------------:|:----------------------------:|:----------------------------:|:---------------------------:|:----------------------------:|
| 0.0229 | 200 | 14.4811 | - | - | - | - | - |
| 0.0459 | 400 | 9.0389 | - | - | - | - | - |
| 0.0688 | 600 | 8.1478 | - | - | - | - | - |
| 0.0918 | 800 | 7.168 | - | - | - | - | - |
| 0.1147 | 1000 | 7.1998 | - | - | - | - | - |
| 0.1377 | 1200 | 6.7985 | - | - | - | - | - |
| 0.1606 | 1400 | 6.3754 | - | - | - | - | - |
| 0.1835 | 1600 | 6.3202 | - | - | - | - | - |
| 0.2065 | 1800 | 5.9186 | - | - | - | - | - |
| 0.2294 | 2000 | 5.9594 | - | - | - | - | - |
| 0.2524 | 2200 | 6.0211 | - | - | - | - | - |
| 0.2753 | 2400 | 5.9984 | - | - | - | - | - |
| 0.2983 | 2600 | 5.8321 | - | - | - | - | - |
| 0.3212 | 2800 | 5.621 | - | - | - | - | - |
| 0.3442 | 3000 | 5.9004 | - | - | - | - | - |
| 0.3671 | 3200 | 5.562 | - | - | - | - | - |
| 0.3900 | 3400 | 5.5125 | - | - | - | - | - |
| 0.4130 | 3600 | 5.4922 | - | - | - | - | - |
| 0.4359 | 3800 | 5.3023 | - | - | - | - | - |
| 0.4589 | 4000 | 5.4376 | - | - | - | - | - |
| 0.4818 | 4200 | 5.1048 | - | - | - | - | - |
| 0.5048 | 4400 | 5.0605 | - | - | - | - | - |
| 0.5277 | 4600 | 4.9985 | - | - | - | - | - |
| 0.5506 | 4800 | 5.2594 | - | - | - | - | - |
| 0.5736 | 5000 | 5.2183 | - | - | - | - | - |
| 0.5965 | 5200 | 5.1621 | - | - | - | - | - |
| 0.6195 | 5400 | 5.166 | - | - | - | - | - |
| 0.6424 | 5600 | 5.2241 | - | - | - | - | - |
| 0.6654 | 5800 | 5.1342 | - | - | - | - | - |
| 0.6883 | 6000 | 5.2267 | - | - | - | - | - |
| 0.7113 | 6200 | 5.1083 | - | - | - | - | - |
| 0.7342 | 6400 | 5.0119 | - | - | - | - | - |
| 0.7571 | 6600 | 4.6471 | - | - | - | - | - |
| 0.7801 | 6800 | 3.6699 | - | - | - | - | - |
| 0.8030 | 7000 | 3.2954 | - | - | - | - | - |
| 0.8260 | 7200 | 3.1039 | - | - | - | - | - |
| 0.8489 | 7400 | 3.001 | - | - | - | - | - |
| 0.8719 | 7600 | 2.8992 | - | - | - | - | - |
| 0.8948 | 7800 | 2.7504 | - | - | - | - | - |
| 0.9177 | 8000 | 2.7891 | - | - | - | - | - |
| 0.9407 | 8200 | 2.7157 | - | - | - | - | - |
| 0.9636 | 8400 | 2.6795 | - | - | - | - | - |
| 0.9866 | 8600 | 2.6278 | - | - | - | - | - |
| 1.0 | 8717 | - | 0.6022 | 0.5960 | 0.6064 | 0.5914 | 0.6160 |
### Framework Versions
- Python: 3.9.18
- Sentence Transformers: 3.0.1
- Transformers: 4.40.0
- PyTorch: 2.2.2+cu121
- Accelerate: 0.26.1
- Datasets: 2.19.0
- Tokenizers: 0.19.1
## Citation
### BibTeX
#### Sentence Transformers
```bibtex
@inproceedings{reimers-2019-sentence-bert,
title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
author = "Reimers, Nils and Gurevych, Iryna",
booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
month = "11",
year = "2019",
publisher = "Association for Computational Linguistics",
url = "https://arxiv.org/abs/1908.10084",
}
```
#### MatryoshkaLoss
```bibtex
@misc{kusupati2024matryoshka,
title={Matryoshka Representation Learning},
author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
year={2024},
eprint={2205.13147},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
#### MultipleNegativesRankingLoss
```bibtex
@misc{henderson2017efficient,
title={Efficient Natural Language Response Suggestion for Smart Reply},
author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
year={2017},
eprint={1705.00652},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## <span style="color:blue">Acknowledgments</span>
The author would like to thank Prince Sultan University for their invaluable support in this project. Their contributions and resources have been instrumental in the development and fine-tuning of these models.
```markdown
## Citation
If you use the Arabic Matryoshka Embeddings Model, please cite it as follows:
@misc{nacar2024enhancingsemanticsimilarityunderstanding,
title={Enhancing Semantic Similarity Understanding in Arabic NLP with Nested Embedding Learning},
author={Omer Nacar and Anis Koubaa},
year={2024},
eprint={2407.21139},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2407.21139},
} | [
"BIOSSES"
] |
RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf | RichardErkhov | null | [
"gguf",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-08-23T02:02:12Z" | 2024-08-23T05:24:33+00:00 | 446 | 0 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
Bio-Medical-Llama-3-8B - GGUF
- Model creator: https://huggingface.co/ContactDoctor/
- Original model: https://huggingface.co/ContactDoctor/Bio-Medical-Llama-3-8B/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [Bio-Medical-Llama-3-8B.Q2_K.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q2_K.gguf) | Q2_K | 2.96GB |
| [Bio-Medical-Llama-3-8B.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.IQ3_XS.gguf) | IQ3_XS | 3.28GB |
| [Bio-Medical-Llama-3-8B.IQ3_S.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.IQ3_S.gguf) | IQ3_S | 3.43GB |
| [Bio-Medical-Llama-3-8B.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q3_K_S.gguf) | Q3_K_S | 3.41GB |
| [Bio-Medical-Llama-3-8B.IQ3_M.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.IQ3_M.gguf) | IQ3_M | 3.52GB |
| [Bio-Medical-Llama-3-8B.Q3_K.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q3_K.gguf) | Q3_K | 3.74GB |
| [Bio-Medical-Llama-3-8B.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q3_K_M.gguf) | Q3_K_M | 3.74GB |
| [Bio-Medical-Llama-3-8B.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q3_K_L.gguf) | Q3_K_L | 4.03GB |
| [Bio-Medical-Llama-3-8B.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.IQ4_XS.gguf) | IQ4_XS | 4.18GB |
| [Bio-Medical-Llama-3-8B.Q4_0.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q4_0.gguf) | Q4_0 | 4.34GB |
| [Bio-Medical-Llama-3-8B.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.IQ4_NL.gguf) | IQ4_NL | 4.38GB |
| [Bio-Medical-Llama-3-8B.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q4_K_S.gguf) | Q4_K_S | 4.37GB |
| [Bio-Medical-Llama-3-8B.Q4_K.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q4_K.gguf) | Q4_K | 4.58GB |
| [Bio-Medical-Llama-3-8B.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q4_K_M.gguf) | Q4_K_M | 4.58GB |
| [Bio-Medical-Llama-3-8B.Q4_1.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q4_1.gguf) | Q4_1 | 4.78GB |
| [Bio-Medical-Llama-3-8B.Q5_0.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q5_0.gguf) | Q5_0 | 5.21GB |
| [Bio-Medical-Llama-3-8B.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q5_K_S.gguf) | Q5_K_S | 5.21GB |
| [Bio-Medical-Llama-3-8B.Q5_K.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q5_K.gguf) | Q5_K | 5.34GB |
| [Bio-Medical-Llama-3-8B.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q5_K_M.gguf) | Q5_K_M | 5.34GB |
| [Bio-Medical-Llama-3-8B.Q5_1.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q5_1.gguf) | Q5_1 | 5.65GB |
| [Bio-Medical-Llama-3-8B.Q6_K.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q6_K.gguf) | Q6_K | 6.14GB |
| [Bio-Medical-Llama-3-8B.Q8_0.gguf](https://huggingface.co/RichardErkhov/ContactDoctor_-_Bio-Medical-Llama-3-8B-gguf/blob/main/Bio-Medical-Llama-3-8B.Q8_0.gguf) | Q8_0 | 7.95GB |
Original model description:
---
license: llama3
library_name: transformers
tags:
- generated_from_trainer
- medical
- Healthcare & Lifesciences
- BioMed
base_model: meta-llama/Meta-Llama-3-8B-Instruct
thumbnail: https://collaiborate.com/logo/logo-blue-bg-1.png
model-index:
- name: Bio-Medical-Llama-3-8B
results: []
datasets:
- collaiborateorg/BioMedData
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# Bio-Medical
This model is a fine-tuned version of https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct on our custom "BioMedData" dataset.
## Model details
Model Name: Bio-Medical-Llama-3-8B
Base Model: Llama-3-8B-Instruct
Parameter Count: 8 billion
Training Data: Custom high-quality biomedical dataset
Number of Entries in Dataset: 500,000+
Dataset Composition: The dataset comprises both synthetic and manually curated samples, ensuring a diverse and comprehensive coverage of biomedical knowledge.
## Model description
Bio-Medical-Llama-3-8B model is a specialized large language model designed for biomedical applications. It is finetuned from the meta-llama/Meta-Llama-3.1-8B-Instruct model using a custom dataset containing over 500,000 diverse entries. These entries include a mix of synthetic and manually curated data, ensuring high quality and broad coverage of biomedical topics.
The model is trained to understand and generate text related to various biomedical fields, making it a valuable tool for researchers, clinicians, and other professionals in the biomedical domain.
## Evaluation Metrics
Bio-Medical-Llama-3-8B model outperforms many of the leading LLMs and find below its metrics evaluated using the Eleuther AI Language Model Evaluation Harness framework against the tasks medmcqa, medqa_4options, mmlu_anatomy, mmlu_clinical_knowledge, mmlu_college_biology, mmlu_college_medicine, mmlu_medical_genetics, mmlu_professional_medicine and pubmedqa.
## Intended uses & limitations
Bio-Medical-Llama-3-8B model is intended for a wide range of applications within the biomedical field, including:
1. Research Support: Assisting researchers in literature review and data extraction from biomedical texts.
2. Clinical Decision Support: Providing information to support clinical decision-making processes.
3. Educational Tool: Serving as a resource for medical students and professionals seeking to expand their knowledge base.
## Limitations and Ethical Considerations
While Bio-Medical-Llama-3-8B model performs well in various biomedical NLP tasks, users should be aware of the following limitations:
> Biases: The model may inherit biases present in the training data. Efforts have been made to curate a balanced dataset, but some biases may persist.
> Accuracy: The model's responses are based on patterns in the data it has seen and may not always be accurate or up-to-date. Users should verify critical information from reliable sources.
> Ethical Use: The model should be used responsibly, particularly in clinical settings where the stakes are high. It should complement, not replace, professional judgment and expertise.
## How to use
import transformers
import torch
model_id = "ContactDoctor/Bio-Medical-Llama-3-8B"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
messages = [
{"role": "system", "content": "You are an expert trained on healthcare and biomedical domain!"},
{"role": "user", "content": "I'm a 35-year-old male and for the past few months, I've been experiencing fatigue, increased sensitivity to cold, and dry, itchy skin. What is the diagnosis here?"},
]
prompt = pipeline.tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
terminators = [
pipeline.tokenizer.eos_token_id,
pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = pipeline(
prompt,
max_new_tokens=256,
eos_token_id=terminators,
do_sample=True,
temperature=0.6,
top_p=0.9,
)
print(outputs[0]["generated_text"][len(prompt):])
### Contact Information
For further information, inquiries, or issues related to Biomed-LLM, please contact:
Email: info@contactdoctor.in
Website: https://www.contactdoctor.in
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0002
- train_batch_size: 12
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.03
- training_steps: 2000
- mixed_precision_training: Native AMP
### Framework versions
- PEFT 0.11.0
- Transformers 4.40.2
- Pytorch 2.1.2
- Datasets 2.19.1
- Tokenizers 0.19.1
### Citation
If you use Bio-Medical LLM in your research or applications, please cite it as follows:
@misc{ContactDoctor_Bio-Medical-Llama-3-8B,
author = ContactDoctor,
title = {Bio-Medical: A High-Performance Biomedical Language Model},
year = {2024},
howpublished = {https://huggingface.co/ContactDoctor/Bio-Medical-Llama-3-8B},
}
| [
"MEDQA",
"PUBMEDQA"
] |
AITeamVN/Vi-Qwen2-3B-RAG | AITeamVN | text-generation | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"retrieval-augmented-generation",
"text-generation-inference",
"conversational",
"vi",
"base_model:Qwen/Qwen2-7B-Instruct",
"base_model:finetune:Qwen/Qwen2-7B-Instruct",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-10-01T03:11:34Z" | 2025-02-28T09:05:55+00:00 | 442 | 0 | ---
base_model: Qwen/Qwen2-7B-Instruct
language:
- vi
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
tags:
- retrieval-augmented-generation
- text-generation-inference
---
## Model Card: Vi-Qwen2-7B-RAG
**Mô tả mô hình:**
Vi-Qwen2-7B-RAG là một mô hình ngôn ngữ lớn được tinh chỉnh từ mô hình cơ sở Qwen2-7B-Instruct (https://huggingface.co/Qwen/Qwen2-7B-Instruct) phục vụ cho RAG tasks. Mô hình được đào tạo trên tập dữ liệu tiếng Việt với mục tiêu cải thiện khả năng xử lý ngôn ngữ tiếng Việt và nâng cao hiệu suất cho các tác vụ Retrieval Augmented Generation (RAG).
**Mục đích sử dụng:**
Mô hình Vi-Qwen2-7B-RAG được thiết kế chuyên biệt cho RAG (ngữ cảnh chấp nhận lên đến 8192 tokens), vì vậy nó có thể giải quyết các trường hợp sau:
* Khả năng chống nhiều: Mô hình trích xuất thông tin hữu ích từ các tài liệu nhiễu. ( 1 positive + 4 negative hoặc 1 negative)
* Loại bỏ negative: Mô hình từ chối trả lời câu hỏi khi kiến thức cần thiết không có trong bất kỳ tài liệu nào được truy xuất. (1-6 negative)
* Tích hợp thông tin: Mô hình trả lời các câu hỏi phức tạp đòi hỏi phải tích hợp thông tin từ nhiều tài liệu. ( 2 part positive + 3 negative hoặc 3 part positive + 2 negative)
* Xác đinh positive/negative: Mô hình xác định xem ngữ cảnh có chứa câu trả lời cho câu hỏi hay không. (độ chính xác xấp xỉ 99%)
Ngoài ra, chúng tôi cũng triển khai các mô hình nhỏ hơn phù hợp với mục đích sử dụng khác nhau như Vi-Qwen2-1.5B-RAG (https://huggingface.co/AITeamVN/Vi-Qwen2-1.5B-RAG)
và Vi-Qwen2.5-3B-RAG (https://huggingface.co/AITeamVN/Vi-Qwen2-3B-RAG)
* Ngoài RAG task, bạn vẫn có thể chatbot bình thường với model của chúng tôi. Thậm chí có thể hỏi các câu hỏi liên tục với ngữ cảnh đầu vào.
**Hạn chế:**
Vì mô hình chỉ được thiết kế chuyên biệt cho RAG task, nên có thể gặp 1 số hạn chế sau:
* Không đảm bảo độ chính xác về các câu hỏi liên quan đến chính trị, xã hội,...
* Có thể thể hiện thành kiến hoặc quan điểm không phù hợp.
**Benchmarks**
Chúng tôi đã đánh giá một số mô hình LLMs cho RAG task sử dụng tập dữ liệu được tạo thủ công bởi chúng tôi [EvalRAGData](https://huggingface.co/datasets/AITeamVN/EvalRAGData).
Đánh giá được thực hiện bởi con người chấm điểm trên thang điểm 10. Dưới đây là kết quả chi tiết:
**Các cách sử dụng:**
#### 1. Sử dụng cơ bản
Ngữ cảnh đầu vào chỉ chứa 1 ngữ cảnh (1 postive hoặc 1 negative).
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
path = 'AITeamVN/Vi-Qwen2-7B-RAG'
model = AutoModelForCausalLM.from_pretrained(
path,
torch_dtype=torch.bfloat16,
device_map="auto",
use_cache=True
)
tokenizer = AutoTokenizer.from_pretrained(path)
system_prompt = "Bạn là một trợ lí Tiếng Việt nhiệt tình và trung thực. Hãy luôn trả lời một cách hữu ích nhất có thể."
template = '''Chú ý các yêu cầu sau:
- Câu trả lời phải chính xác và đầy đủ nếu ngữ cảnh có câu trả lời.
- Chỉ sử dụng các thông tin có trong ngữ cảnh được cung cấp.
- Chỉ cần từ chối trả lời và không suy luận gì thêm nếu ngữ cảnh không có câu trả lời.
Hãy trả lời câu hỏi dựa trên ngữ cảnh:
### Ngữ cảnh :
{context}
### Câu hỏi :
{question}
### Trả lời :'''
# Ví dụ
context = '''Thuốc Insuact 10 trị bệnh gì? Thuốc Insuact 10mg có thành phần chính là Atorvastatin. Thuốc Insuact 10 có tác dụng làm giảm cholesterol và triglycerid trong máu ở bệnh nhân tăng cholesterol máu nguyên phát, rối loạn lipid máu hỗn hợp. 1. Thuốc Insuact 10 trị bệnh gì? Thuốc Insuact 10 thuộc nhóm thuốc điều trị rối loạn lipid máu, có thành phần chính là Atorvastatin 10mg. Atorvastatin có tác dụng làm giảm cholesterol, ức chế enzym tạo cholesterol ở gan. Atorvastatin làm giảm cholesterol chung bao gồm cholesterol LDL , triglycerid trong máu. Thuốc Insuact 10mg được bào chế dưới dạng viên nén bao phim, được chỉ định dùng trong những trường hợp sau: Ðiều trị hỗ trợ tăng cholesterol máu nguyên phát và rối loạn lipid máu hỗn hợp trên bệnh nhân đang áp dụng chế độ ăn kiêng để làm giảm cholesterol toàn phần , cholesterol LDL , apolipoprotein B, triglycerid và tăng cholesterol HDL . Insuact 10 cũng được dùng để điều trị rối loạn betalipoprotein trong máu nguyên phát. Ðiều trị hỗ trợ tăng cholesterol trong máu có tính gia đình đồng hợp tử trên bệnh nhân đang áp dụng các biện pháp làm giảm lipid khác để làm giảm cholesterol toàn phần và cholesterol LDL. 2. Liều dùng và cách dùng thuốc Insuact 10 Cách dùng thuốc Insuact 10: Thuốc được dùng theo đường uống, uống khi bụng đói hoặc no đều được, có thể uống vào bất cứ lúc nào trong ngày. Liều dùng thuốc Insuact 10mg khởi đầu là 10mg/lần/ngày, tối đa là 80mg/lần/ngày. Liều dùng thuốc Insuact 10 tùy vào mục đích điều trị cụ thể như sau: Tăng cholesterol máu nguyên phát và rối loạn lipid máu phối hợp: 10mg/lần/ngày, sau 2 - 4 tuần sẽ thấy hiệu quả của thuốc. Thuốc cần được được sử dụng duy trì trong thời gian dài để có hiệu quả. Tăng cholesterol trong máu có tính gia đình đồng hợp tử: Liều thường dùng là thuốc Insuact 10mg /lần/ngày và tối đa là 80mg/lần/ngày. Rối loạn lipid máu nghiêm trọng ở trẻ từ 10 - 17 tuổi: 10mg/lần/ngày, sau đó tăng lên 20mg/lần/ngày tùy vào cơ địa, tiến triển bệnh và khả năng dung nạp thuốc của người bệnh. Thời gian điều chỉnh liều thuốc tối thiểu là 4 tuần. 3. Tác dụng phụ của thuốc Insuact 10mg Thuốc Insuact 10 có thể gây một số tác dụng phụ không mong muốn với tần suất như sau: Thường gặp: Viêm mũi - họng, phản ứng dị ứng, tăng đường huyết, nhức đầu, đau thanh quản, chảy máu cam , đau cơ, co thắt cơ, đau khớp, sưng khớp, đau các chi, đau lưng, xét nghiệm gan bất thường, tăng creatine kinase trong máu, buồn nôn, khó tiêu, đầy hơi, táo bón, tiêu chảy. Ít gặp: Insuact 10 ít gây hạ đường huyết, tăng cân, chán ăn, mất ngủ, gặp ác mộng, choáng váng, dị cảm, mất trí nhớ, giảm cảm giác, loạn vị giác , nôn, đau bụng, ợ hơi, viêm tụy, viêm gan, nổi mày đay , phát ban, ngứa, rụng tóc, đau cổ, mỏi cơ, mệt mỏi, suy nhược, đau ngực, phù ngoại biên, sốt, xuất hiện bạch cầu trong nước tiểu, nhìn mờ, ù tai. Hiếm gặp: Insuact 10 hiếm khi làm giảm tiểu cầu, bệnh lý thần kinh ngoại biên, hoa mắt, ứ mật, phù thần kinh, nổi hồng ban, hội chứng hoại tử da nhiễm độc , hội chứng Stevens-Johnson , bệnh cơ, viêm cơ, tiêu cơ vân, bệnh gân, đôi khi nghiêm trọng hơn có thể đứt gân. Rất hiếm gặp: Insuact 10 rất hiếm khi gây sốc phản vệ , mất thính giác , suy gan , hội chứng to vú ở nam giới. Không rõ tần suất: Hoại tử cơ tự miễn trung gian. 4. Một số lưu ý khi dùng thuốc Insuact 10mg Không dùng thuốc Insuact với người bị quá mẫn với thành phần của thuốc, người có bệnh gan hoạt động hoặc tăng transaminase huyết thanh vô căn kéo dài, phụ nữ đang mang thai hoặc nuôi con cho bú, phụ nữ đang có ý định mang thai. Thuốc Insuact 10mg chỉ được dùng ở bệnh nhân có nguy cơ xơ vữa mạch máu cao do tăng cholesterol trong máu và phải kết hợp với chế độ ăn kiêng ít chất béo bão hòa , ít cholesterol và người bệnh đang áp dụng các biện pháp điều trị không dùng thuốc khác. Trước khi điều trị với Insuact 10 , người bệnh cần được loại trừ các nguyên nhân thứ phát gây tăng cholesterol bao gồm suy tuyến giáp , tiểu đường khó kiểm soát, hội chứng thận hư, nghiện rượu, bệnh gan tắc nghẽn, rối loạn protein trong máu, .... Ngoài ra, người bệnh cũng cần được kiểm tra, đo lường nồng độ lipid máu. Trước khi điều trị với thuốc Insuact 10mg , cần kiểm tra chức năng gan và trong quá trình dùng thuốc, người bệnh cần theo dõi, kiểm tra chức năng gan thường xuyên. Người có tiền sử mắc bệnh gan do rượu, bị nghiện rượu cần thận trọng khi dùng Insuact 10 . Trước khi dùng thuốc, người bệnh cần được cảnh báo nguy cơ gặp phải các vấn đề về cơ như căng cơ , đau cơ, yếu cơ. Thuốc Insuact 10mg có thể tương tác với các thuốc: Tăng nguy cơ tổn thương cơ (đặc biệt cơ vân) khi dùng đồng thời với thuốc điều trị viêm gan C và HIV, hạ cholesterol máu nhóm fibrat khác, thuốc ức chế miễn dịch; tăng thời gian chảy máu hoặc thời gian prothrombin khi dùng phối hợp với thuốc chống đông, indandione; tăng nhẹ nồng độ digoxin trong huyết thanh khi dùng phối hợp với Digoxin'''
question = '''Insuact 10mg được chỉ định dùng trong những trường hợp nào và liều dùng cụ thể ra sao?'''
conversation = [{"role": "system", "content": system_prompt }]
conversation.append({"role": "user", "content": template.format(context = context, question = question)})
text = tokenizer.apply_chat_template(
conversation,
tokenize=False,
add_generation_prompt=True)
model_inputs = tokenizer(text,return_tensors="pt").to(model.device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=2048,
temperature = 0.1,
#top_p=0.95,
#top_k=40,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
'''Trả lời:
Dựa trên ngữ cảnh cung cấp, Insuact 10mg được chỉ định dùng trong các trường hợp sau:
1. **Điều trị hỗ trợ tăng cholesterol máu nguyên phát và rối loạn lipid máu hỗn hợp**:
- **Liều dùng**: 10mg/lần/ngày.
- **Hiệu quả**: Sau 2 - 4 tuần sử dụng, bệnh nhân sẽ thấy hiệu quả của thuốc.
- **Lưu ý**: Thuốc cần được sử dụng duy trì trong thời gian dài để có hiệu quả.
2. **Điều trị hỗ trợ tăng cholesterol trong máu có tính gia đình đồng hợp tử**:
- **Liều dùng**: 10mg/lần/ngày (liều thường dùng) và tối đa 80mg/lần/ngày.
- **Lưu ý**: Liều lượng có thể được điều chỉnh tùy theo cơ địa, tiến triển bệnh và khả năng dung nạp thuốc của người bệnh.
3. **Rối loạn lipid máu nghiêm trọng ở trẻ từ 10 - 17 tuổi**:
- **Liều dùng**: 10mg/lần/ngày, sau đó có thể tăng lên 20mg/lần/ngày tùy theo tình trạng bệnh.
- **Thời gian điều chỉnh liều**: Thời gian điều chỉnh liều tối thiểu là 4 tuần.
**Lưu ý chung**:
- Thuốc được dùng theo đường uống, uống khi bụng đói hoặc no đều được, có thể uống vào bất cứ lúc nào trong ngày.
- Không dùng thuốc Insuact 10mg với người bị quá mẫn với thành phần của thuốc, người có bệnh gan hoạt động hoặc tăng transaminase huyết thanh vô căn kéo dài, phụ nữ đang mang thai hoặc nuôi con cho bú, phụ nữ đang có ý định mang thai.
- Cần kiểm tra chức năng gan trước và trong quá trình điều trị.
'''
```
#### 2. Khả năng chống nhiều
```python
# ví dụ
context = '''- Ngữ cảnh 1: . Và dù nguyên nhân thế nào, thì việc đánh người khác đều không thể chấp nhận, không ai có quyền xâm phạm thân thể của người khác, nhất là đánh những đứa trẻ là bạn của con cái mình. Lựa chọn kể với cha mẹ, người lớn về các mâu thuẫn học đường, là quyết định rất đúng của các em. Nhưng người lớn, đặc biệt những ông bố, bà mẹ cần ứng xử ra sao trước xung đột này của các con, thưa bà? Đứng ngoài mâu thuẫn bằng sự khách quan và trách nhiệm nhất có thể. Điều này giúp chúng ta đủ bình tĩnh để làm việc với tất cả các bên liên quan, từ giáo viên, bạn của con, ban giám hiệu để tìm hiểu câu chuyện và tìm kiếm cách giải quyết tích cực, trên cơ sở phối hợp nhà trường. Người lớn không thể chỉ nghe một tai và đặc biệt không nên tự xử. Phụ huynh, kể cả học sinh tự xử các vấn đề học đường là điều rất nguy hiểm và cho thấy sự coi thường pháp luật . Vụ việc ở Tuyên Quang vừa rồi là ví dụ. Các em hoàn toàn có thể phản ứng bằng cách trình bày, gửi yêu cầu, kiến nghị lên nhà trường, nhờ phụ huynh làm việc với ban giám hiệu để có biện pháp giải quyết nếu cô giáo sai, không nên đồng loạt dồn cô giáo vào tường một cách bạo lực và trái đạo đức, tôn ti trật tự như vậy. Ngoài ra, chúng ta cũng có rất nhiều cơ quan chức năng bảo vệ phụ huynh và con em, với những quyền về khiếu nại, tố cáo. Chúng ta nói nhiều về trường học an toàn. Trong những câu chuyện học sinh bị hành hung thế này, có lẽ cũng cần làm rõ vai trò, trách nhiệm của nhà trường? TPHCM và nhiều địa phương đang xây dựng môi trường trường học hạnh phúc, tiêu chí là yêu thương, an toàn, tôn trọng. Không chỉ phòng ngừa, nhà trường còn phải tích cực vào cuộc xử lý các mâu thuẫn học đường, hạn chế tối đa nguy cơ mất an toàn cho học sinh, giáo viên. Đặc biệt, giải quyết câu chuyện bạo lực học đường phải triệt để, tuyệt đối không nửa vời vì nửa vời sẽ tiềm ẩn nguy cơ rất lớn dẫn đến những vụ việc tương tự, với mức độ nghiêm trọng hơn. Vụ việc em M. ở Nha Trang tự vẫn với lá thư tuyệt mệnh bị đổ oan đầu tháng 10 vừa qua là một ví dụ về giải quyết không triệt để. Việc xây dựng trường học hạnh phúc nếu triển khai “đến nơi đến chốn”, sẽ góp phần rất lớn cải thiện tình trạng bạo lực học đường, tạo môi trường sống và học tập bình an cho các con. Từ nhiều sự vụ học sinh bạo hành lẫn nhau, giáo viên bạo hành học sinh, phụ huynh hành hung giáo viên và bạn của con. Tam giác phối hợp bảo vệ học sinh là nhà trường - gia đình - xã hội phải chăng đang có một lỗ hổng lớn, thưa bà? Câu chuyện này có liên quan đến niềm tin của phụ huynh với nhà trường. Tại sao phụ huynh lại chọn cách tự xử? Chúng ta cũng cần phải xem lại cách giải quyết vấn đề của nhà trường đã rốt ráo chưa, coi trọng lợi ích của tất cả các bên liên quan chưa hay chỉ đang xoa dịu? Người ta chỉ tìm đến nhau khi có niềm tin vào nhau. Thực trạng phụ huynh chọn cách chuyển trường cho con cũng nói lên điều này. Đây là cách chạy trốn của phụ huynh với mong muốn con được an toàn, hạnh phúc hơn ở môi trường mới. Xây dựng niềm tin cho phụ huynh, xã hội cần được chú trọng và với mỗi một trường hợp phụ huynh yêu cầu chuyển trường cho con - đang rất phổ biến - nhà trường cần xét kỹ các nguyên nhân và hóa giải. Xin bà cho biết đâu là giải pháp căn cơ cho tất cả những câu chuyện bạo lực nói trên? Để trẻ không là nạn nhân của bạo lực học đường, phụ huynh cần đồng hành và giúp con có sự hiểu biết, ý thức trước vấn đề này. Dạy con kỹ năng giao tiếp, quản lý cảm xúc rất quan trọng và điều này không thể chỉ dựa vào những khóa học kỹ năng sống, mà là từ cách cư xử của người lớn, cha mẹ, thầy cô. Không có tấm gương nào tốt hơn cho con trẻ bằng ứng xử, hành vi của người lớn. Vì vậy, không thể đòi hỏi trẻ nói không với bạo lực học đường khi trong chính từng gia đình, xã hội, người lớn vẫn đối xử với nhau bằng bạo lực.
- Ngữ cảnh 2: Tại sao triều Thanh có rất nhiều thân vương nhưng chẳng có ai dám tạo phản? Không giống như những triều đại trước đó, triều Thanh dù có sự tranh giành ngai vàng khốc liệt giữa các hoàng tử nhưng lại chẳng bao giờ xảy ra thế cục các thân vương tạo phản. Chính vì 3 lý do lớn này đã khiến cho triều đại nhà Thanh khác hẳn triều đại nhà Đường và nhà Minh. Trong thời cổ đại, các vương công quý tộc để tranh giành vương vị của mình, giữa huynh đệ ruột thịt với nhau dường như đều xảy ra đấu đá, hãm hại lẫn nhau, coi nhau như kẻ thù không đội trời chung, có ta thì không có ngươi, có ngươi thì sẽ chẳng có ta, điều này hoàn toàn không phải là điều gì xa lạ. Vậy thì tại sao ngai vàng lại có sức hút lớn đến thế? Không chỉ là đàn ông khát khao quyền lực, mà quan trọng hơn là hoàng đế có thể có được hậu cung rộng lớn, trong hậu cung còn có vô số các mỹ nữ quốc sắc thiên hương. Nhiều phi tần như vậy, đương nhiên hoàng đế cũng sẽ có rất nhiều con cái, không tính đến con gái, chỉ riêng những vị hoàng tử, để có thể có được hoàng vị, họ tranh giành nhau bằng cả sinh mạng. Vậy thì ai là người được lựa chọn để thừa kế ngai vàng, ai mới có thể gánh được trọng trách trị vì đất nước? Đa phần đều theo tục lệ truyền cho con trai đích tôn (con trai do hoàng hậu sinh ra) hoặc con trai trưởng (con trai đầu tiên của hoàng đế). Cho dù tục lệ này có lịch sử lâu đời nhưng nó cũng có những khuyết điểm rất lớn, đó chính là nếu như năng lực và chí hướng của con trai đích tôn hoặc con trai trưởng không thể gánh vác được ngai vị, nếu để anh ta lên ngôi hoàng đế, vậy thì đất nước sẽ rơi vào cục diện suy vong. Còn có một khuyết điểm nữa đó chính là những người con trai có dã tâm lớn khác sẽ không phục việc con trai đích hoặc con trai trưởng kế thừa ngôi báu, họ sẽ khởi binh tạo phản cũng là chuyện rất dễ xảy ra. Ví dụ như trong thời Đường của Trung Quốc, Đường Cao Tổ Lý Uyên đem binh tiêu diệt nhà Tùy thối nát, đồng thời lập nên nhà Đường, vốn dĩ ông cũng dựa theo tục lệ lập con trai trưởng là Lý Kiến Thành làm Thái tử nhưng con trai thứ là Lý Thế Dân lại không phục với sự sắp xếp này. Vì năng lực của ông xuất chúng, văn võ song toàn, còn lập được không ít công lao to lớn trong cuộc chiến tranh tiêu diệt nhà Tùy cùng cha mình, đương nhiên không chịu thấp hơn kẻ khác một bậc. Thế nên đã phát động binh biến Huyền Vũ Môn, trong cuộc binh biến tạo phản này, đích thân ông đã giết chết huynh trưởng của mình, đồng thời ép cha mình là Lý Uyên phải truyền ngôi cho mình. Hay như trong thời nhà Minh của Trung Quốc, trước khi Chu Nguyên Chương chọn người lập làm Thái tử, con trai trưởng Chu Tiêu đã qua đời vì bệnh nặng, thế nên Chu Nguyên Chương đã lập cháu đích tôn của mình làm Thái tử kế thừa vương vị, nhưng em trai của Chu Tiêu là Chu Đệ lại không phục lựa chọn này của Chu Nguyên Chương. Theo lý mà nói thì sau khi anh trai Chu Tiêu qua đời, ông đã có tư cách thừa kế ngai vàng nhưng Chu Nguyên Chương nhất quyết không chọn ông mà lại chọn người cách thế hệ để truyền ngôi. Điều này khiến Chu Đệ với thế lực to lớn không thể nuốt nổi cục tức này, vì thế Chu Tiêu vừa qua đời thì ông đã vội vã khởi binh tạo phản, giết chết cháu trai ruột của mình rồi tự xưng vương. Vậy thì tại sao trong triều Thanh có rất nhiều thân vương như vậy mà lại chẳng có ai đứng ra tạo phản? Đầu tiên phải nói về bối cảnh xã hội trong thời kỳ này. Triều Thanh từ khi thành lập, cũng giống với những triều đại khác, đều có rất nhiều thân vương. Nếu người dân bình thường muốn làm hoàng đế, vậy thì đó là điều hoàn toàn không thể, nhưng đối với những vương công quý tộc trong hoàng thất mà nói, họ đương nhiên sẽ có rất nhiều cơ hội, đặc biệt là những thân vương nắm đại quyền quân sự , họ chính là mối đe dọa lớn nhất đối với nhà vua. Vì thế, các đời hoàng đế đều sẽ nghĩ đủ mọi cách để áp chế, kiểm soát họ, tránh việc họ khởi binh tạo phản. Triều Thanh có lịch sử hơn 300 năm, cũng đã cho ra đời vô số thân vương, đặc biệt là cuối thời Thanh, khi Trung Quốc rơi vào cảnh khốn khó, sau khi Từ Hy Thái Hậu cầm quyền thì thế cục này càng được thể hiện rõ rệt hơn. Nhưng cho dù là một người phụ nữ cầm quyền thì cũng chẳng có một vị thân vương hoàng tộc nào đứng ra tạo phản. Có 3 nguyên nhân sau: Thứ nhất, thân vương triều Thanh không thể nối ngôi, nếu muốn tiếp tục duy trì danh phận thân vương, vậy thì bắt buộc phải có được sự đồng ý của hoàng đế và phải lập được công lao cho đất nước. Thứ hai, triều đình tiến hành giám sát nghiêm ngặt đối với các thân vương, họ không hề có cơ hội để tạo phản. Thứ ba, các thân vương không thể giao thiệp quá sâu với các đại thần, quan lại khác, điều này cũng khiến các thân vương rơi vào cảnh bị cô lập, thế nên càng không có cơ hội để cấu kết với người khác hòng tạo phản. - Video: Ngắm sự kỳ vĩ và lộng lấy của Tử Cấm Thành từ trên cao. Nguồn: Sky Eye.
- Ngữ cảnh 3: . Cùng điều chỉnh với con là điều rất quan trọng bởi vì trẻ sẽ tự tin để tự đặt những giới hạn cho chính mình khi lớn lên”, TS Nguyễn Thị Thanh đưa ra lời khuyên. “Khi con mắc sai lầm, hãy giúp chúng tìm những cách khác tốt hơn. Đơn cử dùng hậu quả để dạy cho chúng bài học, điều đó tốt hơn rất nhiều việc xử phạt. Nếu cha mẹ chỉ biết trừng phạt, sẽ nhận được lời xin lỗi nhưng không thể giúp trẻ tỉnh ngộ. Bởi chúng chỉ biết được mình đã sai mà không biết sai ở chỗ nào và làm thế nào mới là đúng”
- Ngữ cảnh 4: . “MẤT ĐI CHA MẸ Ở TUỔI ĐẸP NHẤT CỦA NGƯỜI PHỤ NỮ CÀNG KHIẾN TÔI PHẢI MẠNH MẼ” - Làm con của nghệ sĩ Thanh Hiền, Đinh Y Nhung cảm nhận sợi dây liên kết giữa hai mẹ con thế nào? Má Thanh Hiền là người rất tuyệt vời. Hai má con hồi xưa từng làm phim truyền hình với nhau rồi, cho nên khi tái hợp thì không mấy bỡ ngỡ. Khi đối diễn, hai má con rất ăn ý, như người thân ruột thịt vậy đó. - Khi thể hiện những phân cảnh cảm động trong phim, có khi nào chị thấy nhớ mẹ không? Có chứ, nhất là ở những phân đoạn gia đình sum họp, tự nhiên mình bị buồn. Ai cũng muốn có cha, có mẹ, ai cũng muốn Tết được chạy về bên gia đình. Trong mười mấy, hai chục năm qua, Nhung bị chạnh lòng. Tuy nhiên, chỉ trong tích tắc, tôi tự trấn an rằng, mình đang quay phim, đang hóa thân vào nhân vật nên không thể xao lãng được. Mình là con người mà, cũng có lúc tâm trạng vui buồn bất chợt, nhưng Nhung luôn cố gắng lấy lại phong độ liền. - Mất ba mẹ từ sớm, không có chỗ dựa tinh thần, cô gái trẻ Đinh Y Nhung năm đó có nhận những lời mời gọi khiếm nhã không? Trước đây, Nhung không có bạn bè nhiều, chủ yếu chỉ lo đi học, đi làm để lo cho cuộc sống thôi. Nên Nhung không phải đón nhận những lời mời gọi nào hết. - Mất mát từ quá khứ có ảnh hưởng gì đến suy nghĩ về tương lai của chị sau này, ví dụ khi có con thì sẽ bù đắp, chăm sóc cho con nhiều hơn? Năm ba mẹ mất thì mình vẫn còn khá trẻ, thật ra cái tuổi đó là tuổi đẹp của người phụ nữ. Sau đó, tôi đi làm, rồi yêu đương và lập gia đình. Có rất nhiều thứ hối tiếc để nói về Nhung của thời điểm đó. Thứ nhất là mình chưa thành công, thứ hai là mình chưa trả hiếu cho cha mẹ, thứ ba là mình còn bấp bênh. Nhung lúc đó lì lợm lắm, không cho phép mình ngã, bằng mọi giá phải tiến về trước dù có hàng ngàn chông gai ngăn cản. Có lúc tôi bị người này cười, người kia mỉa, nhưng mà mình vẫn cố bước đi. Người ta có cười thì cũng không mang lại cho mình được gì, tôi chỉ biết làm hết khả năng để lo cho bản thân, lo cho em của mình. Hiện, con cái Nhung đã đi nước ngoài rồi. Bé đang học đại học về âm nhạc, còn em mình cũng đã lớn rồi. Đối với Nhung ngay lúc này thì không phải thành công hay hoàn hảo lắm, nhưng ít nhất là tôi đã cố gắng để tự chịu trách nhiệm với cuộc đời mình. Mất cha, mất mẹ, đối với một người hai mươi mấy tuổi thì điều cần nhất lúc đó là có được gia đình ở bên. Nhưng mình không có chỗ dựa tinh thần thì càng phải mạnh mẽ hơn nữa. Tôi tự gặm nhấm nỗi đau mất người thân trong một thời gian dài, có khi đến cả bạn bè cũng không hề biết. Một thời gian sau, bạn bè thời và mọi người mới biết. Còn người hâm mộ, đồng nghiệp trong nghề gần như không biết chuyện ba mẹ Nhung mất sớm, chỉ có vài người chơi thân với nhau biết thôi. Sau này, dần dần tâm lý dần ổn định thì mình mới bắt đầu chia sẻ. “CON ĐI DU HỌC, TÔI DẶN BÉ CÁI GÌ KHÔNG TỐT THÌ MÌNH BỎ QUA” - Đinh Y Nhung từng tiết lộ mình rất thân với con gái. Có vẻ như quyết định để con đi du học là không hề dễ dàng? Thật sự là không có ba mẹ nào muốn con mình đi xa, nhưng việc du học lại là quyết định của bé. Con Nhung bày tỏ muốn học đại học ở nước ngoài và muốn đi sớm để thực hiện ước mơ. Nhưng lúc đó con còn nhỏ quá, phải đợi đến năm con 17 tuổi thì Nhung mới quyết định cho bạn nhỏ đi. Con cái từ nhỏ ở với bố mẹ giờ lại đi xa thì tất nhiên người làm cha làm mẹ cùng phải thấy sốc, thấy buồn. Nhưng Nhung hoàn toàn tôn trọng quyết định của con về việc chọn ngành nghề và tương lai của mình. Ba mẹ sẽ đứng sau và là người đưa cho con những lời khuyên và chỉ có thể đồng hành cùng con tới một mốc thời gian nào đó. Về sau, con phải đi làm và tự có trách nhiệm với cuộc đời của mình. - Có con gái đang ở tuổi lớn lại xa bố mẹ và tiếp xúc một nền văn hóa phương Tây cởi mở, Đinh Y Nhung đã khuyên dạy và đồng hành với con như thế nào? Ngay khi ở Việt Nam, con gái Nhung đã được theo học trường quốc tế. Hai mẹ con cũng có rất nhiều dịp để tâm sự và chia sẻ với nhau. Ngay từ nhỏ, Nhung đã cho bé được tiếp xúc song song giữa hai nền văn hóa để con không bỡ ngỡ. Mình là người Việt nên đương nhiên vẫn dạy con theo văn hóa Á Đông là chủ yếu. Nhung vẫn luôn tạo điều kiện để con cảm nhận những nét đẹp trong nền văn hóa quê hương. Văn hóa phương Tây thì xa lạ hơn nhưng Nhung cũng khuyên con rằng điều gì hay thì mình học hỏi, cái gì không tốt thì mình nên bỏ qua. Tất nhiên mình không thể theo sát con, nhất là khi bé đang ở độ tuổi mới lớn, có nhiều sự hiếu kỳ. Tuy nhiên, Nhung cũng không quá lo lắng vì qua quá trình học tập ở các trường quốc tế, bé cùng đã được làm quen dần với văn hóa phương Tây. Bé muốn làm bạn với mẹ nên có nhiều thứ bé muốn hỏi, muốn tiếp thu thì hai mẹ con lại ngồi xuống chia sẻ, tâm sự với nhau. Nhung tin, con luôn tỉnh táo để đưa ra những quyết định cho bản thân mình. Nhung không dám nói trước, nhưng hiện tại con vẫn luôn biết nói cảm ơn, xin phép trước khi làm bất cứ điều gì nên mình vẫn rất tin tưởng con. - Chị nhận xét thế nào về tính cách của con gái? Phải chăng bé là phiên bản nhí của chị? Con gái Nhung có nhiều nét giống mẹ.
- Ngữ cảnh 5: . - Trong quá trình quay phim, má Hiền có tham gia đóng góp ý kiến cho dàn diễn viên trẻ không? Không phải má Hiền muốn làm thầy nhưng có những lúc cũng lên tiếng góp ý một vài điều. Những lúc như thế các bạn diễn viên trẻ cũng nghe theo. Phải giúp đỡ nhau như vậy mới có thể tạo ra được một tác phẩm hay. Chứ có nhiều diễn viên trẻ có lòng tự trọng cao, chỉ khi đạo diễn nói mới nghe thì không bao giờ có thể tiến bộ được. Má Hiền phải cảm thấy hạnh phúc mới đúng, càng ngày xung quanh càng có nhiều bạn bè, các mối quan hệ nhờ đó mà cũng ngày càng đông. Ngày xưa, khi tham gia là Mùi Ngò Gai chưa có ai biết nhưng sau khi phim phát sóng mọi người đã nhớ đến má. Hiện tại, má Hiền đã được đạo diễn, nhà sản xuất mời đóng phim đó đã là một niềm hạnh phúc. Đi ra đường hay đi khám bệnh, người ta cứ vây xung quanh quan tâm hỏi han khiến má cảm thấy rất vui lắm (cười). Cảm ơn nghệ sĩ Thanh Hiền đã nhận lời tham gia phỏng vấn với SAOstar. Chúc má có nhiều sức khỏe để tiếp tục làm nghề!'''
question = '''Tại sao các thân vương triều Thanh không dám tạo phản mặc dù có nhiều quyền lực và cơ hội?'''
'''Trả lời:
Các thân vương triều Thanh không dám tạo phản mặc dù có nhiều quyền lực và cơ hội vì có ba lý do chính:
1. **Không thể nối ngôi**: Để duy trì danh phận thân vương, họ phải có sự đồng ý của hoàng đế và phải lập được công lao cho đất nước. Điều này hạn chế cơ hội của họ để trở thành hoàng đế.
2. **Giám sát nghiêm ngặt**: Triều đình tiến hành giám sát nghiêm ngặt đối với các thân vương, không cho họ cơ hội để tạo phản.
3. **Bị cô lập**: Các thân vương không thể giao thiệp quá sâu với các đại thần, quan lại khác, điều này khiến họ bị cô lập và không có cơ hội cấu kết với người khác để tạo phản.'''
```
#### 3. Loại bỏ negative
```python
context = '''- Ngữ cảnh 1: Thái Bình hướng đến là trung tâm công nghiệp, năng lượng của vùng Với tiềm năng sẵn có, quy hoạch tỉnh Thái Bình thời kỳ 2021-2030, tầm nhìn đến năm 2050 xác định tỉnh sẽ phát triển công nghiệp theo hướng hiện đại, bền vững dựa trên nghiên cứu phát triển điện gió, điện khí, cân bằng lượng phát thải. Sáng 5/3, UBND tỉnh Thái Bình tổ chức Hội nghị công bố quy hoạch của tỉnh thời kỳ 2021-2030, tầm nhìn đến năm 2050 và xúc tiến đầu tư tỉnh Thái Bình. Phát biểu tại hội nghị, Phó Chủ tịch Thường trực UBND tỉnh Nguyễn Quang Hưng cho biết: Mục tiêu của quy hoạch là đến năm 2030, Thái Bình trở thành địa phương thuộc nhóm phát triển khá và là một trong những trung tâm phát triển công nghiệp của vùng Đồng bằng sông Hồng, có cơ cấu kinh tế hiện đại với công nghiệp là động lực chủ yếu cho tăng trưởng để Thái Bình phát triển nhanh, toàn diện và bền vững. Đến năm 2050, Thái Bình là tỉnh phát triển của vùng Đồng bằng sông Hồng, tăng trưởng kinh tế dựa trên nền tảng khoa học công nghệ, đổi mới sáng tạo và các ngành kinh tế trụ cột có sức cạnh tranh cao. Quy hoạch tỉnh đã xác định 4 trụ cột tăng trưởng, 3 khâu đột phá, 4 không gian kinh tế - xã hội, 3 hành lang kinh tế, định hướng phát triển các ngành và lĩnh vực và 6 nhiệm vụ trọng tâm. Quy hoạch tỉnh cũng có nhiều điểm mới, đột phá như mở ra không gian phát triển mới thông qua hoạt động “lấn biển”, tạo quỹ đất cho các hoạt động chức năng, hình thành không gian công nghiệp - đô thị - dịch vụ. Về hạ tầng giao thông, Thái Bình sẽ đầu tư 3 tuyến cao tốc là cao tốc Ninh Bình - Hải Phòng (CT.08), đường vành đai 5 - Hà Nội (CT.39) và tuyến CT.16 kết nối Khu kinh tế với thành phố Thái Bình và vùng kinh tế phía Tây Bắc Thủ đô. Tỉnh cũng sẽ đầu tư 101km đường sắt, khổ đường dự kiến rộng 1.435 mm và sân bay chuyên dụng nằm ở ven biển Thái Bình. Về phát triển kinh tế, quy hoạch tỉnh Thái Bình xác định tỉnh sẽ phát triển công nghiệp theo hướng hiện đại, công nghệ tiên tiến, giá trị gia tăng cao, tham gia sâu, toàn diện vào mạng lưới sản xuất, chuỗi giá trị toàn cầu, phát huy các tiềm năng, thế mạnh để đưa Thái Bình trở thành một trong những trung tâm phát triển công nghiệp, năng lượng của vùng Đồng bằng sông Hồng. Tỉnh khuyến khích đầu tư phát triển các ngành có thế mạnh và có thể tạo đột phá như năng lượng, cơ khí chế biến, chế tạo, công nghiệp công nghệ cao, điện - điện tử, chế biến sản phẩm nông, lâm nghiệp và thủy sản… Đồng thời, tập trung nghiên cứu phát triển điện gió, điện khí để tạo nguồn điện sạch và cân bằng lượng phát thải, nghiên cứu đầu tư xây dựng nhà máy chế biến Condensate, chuẩn bị mọi điều kiện để xây dựng và đưa vào vận hành Nhà máy nhiệt điện LNG Thái Bình. Về nông nghiệp, tỉnh Thái Bình vẫn xác định đây là \"trụ cột quan trọng\" trong phát triển kinh tế của tỉnh, góp phần bảo đảm an ninh lương thực quốc gia, hướng tới trở thành trung tâm sản xuất nông nghiệp hàng đầu của Đồng bằng sông Hồng. Phát biểu tại hội nghị, Phó Thủ tướng Chính phủ Trần Lưu Quang đánh giá Thái Bình có 4 tiềm năng, lợi thế lớn để có thể có sự bứt phá trong thời gian tới như vị trí địa lý và tiếp cận đất đai thuận lợi; từng là địa phương đi đầu trong xây dựng nông thôn mới bài bản và nghiêm túc, nhận được sự quan tâm của nhiều thế hệ lãnh đạo Đảng, Nhà nước và có nhiều doanh nhân người Thái Bình và luôn hướng về quê hương; có sự đoàn kết, thống nhất, trước hết là trong tập thể lãnh đạo. Về vị trí địa lý và tiếp cận đất đai, Phó Thủ tướng cho rằng trong tương lai, khi Luật Đất đai có hiệu lực, Thái Bình sẽ có nhiều điều kiện lấn biển để triển khai các dự án khu đô thị, khu công nghiệp thân thiện với môi trường. Đối với nông nghiệp, Phó Thủ tướng nhấn mạnh về lâu dài Thái Bình có thể ghi điểm từ phát triển công nghiệp nhưng trước mắt, đặc biệt trong lúc khó khăn thì nông nghiệp vẫn là nền tảng rất quý giá. Mặt khác, ứng dụng của công nghệ cao trong sản xuất nông nghiệp sẽ rút ngắn thời gian làm đồng của người nông dân, tạo điều kiện để Thái Bình huy động nguồn nhân lực trong nông nghiệp sang phát triển các ngành công nghiệp và dịch vụ, một lợi thế mà không phải địa phương nào cũng có được như Thái Bình. Bên cạnh những lợi thế trên, lãnh đạo Chính phủ chỉ ra một số khó khăn mà tỉnh phải đối mặt như Thái Bình đã sử dụng hết 1.600 ha chỉ tiêu đất công nghiệp trong giai đoạn này, đòi hỏi phải có phương án giải quyết thấu đáo trong thời gian tới để tỉnh tiếp tục phát triển công nghiệp. Đồng thời, Thái Bình cũng phải cạnh tranh với những địa phương như Hải Phòng, Quảng Ninh trong thu hút FDI trong khi phát triển cơ sở hạ tầng chưa theo kịp mong muốn. Do vậy, khi triển khai quy hoạch tỉnh, Phó Thủ tướng nhắn nhủ tới địa phương 8 chữ: Tuân thủ, linh hoạt, đồng bộ và thấu hiểu. Đồng thời, tỉnh cũng phải \"linh hoạt\" trong tổ chức thực hiện, trong trường hợp cá biệt cụ thể, điều chỉnh mục tiêu cho phù hợp. Sáng cùng ngày, Phó Thủ tướng Trần Lưu Quang đã dự Lễ khởi công dự án Nhà máy Pegavision Việt Nam tại khu công nghiệp Liên Hà Thái, huyện Thái Thụy, tỉnh Thái Bình
- Ngữ cảnh 2: Bình Định được định hướng là trung tâm khoa học, công nghệ đổi mới sáng tạo Tỉnh Bình Định được định hướng phát triển ngành công nghiệp phát triển theo hướng hiện đại, quy mô lớn, trở thành một trong những trung tâm công nghiệp chế biến chế tạo và công nghệ cao của vùng Bắc Trung Bộ và duyên hải Trung Bộ. Theo Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 vừa được Thủ tướng Chính phủ phê duyệt, tỉnh Bình Định được định hướng phát triển ngành công nghiệp phát triển theo hướng hiện đại, quy mô lớn, trở thành một trong những trung tâm công nghiệp chế biến chế tạo và công nghệ cao của vùng Bắc Trung Bộ và duyên hải Trung Bộ. Ngành công nghiệp tăng trưởng nhanh, bền vững, hướng tới tăng trưởng xanh, kinh tế tuần hoàn là trụ cột để phát triển và chuyển dịch cơ cấu kinh tế của tỉnh. Ngành chế biến, chế tạo công nghệ cao (dịch chuyển ngành công nghiệp chế biến, chế tạo sang lĩnh vực sản xuất có giá trị gia tăng cao như: chế biến sâu nông - thủy - hải sản, linh kiện điện tử, bán dẫn, dược phẩm), công nghệ thông tin, trí tuệ nhân tạo trở thành một trong những lĩnh vực đột phá, góp phần đưa tỉnh Bình Định trở thành một trung tâm khoa học, công nghệ đổi mới sáng tạo của vùng và cả nước. Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 đặt ra yêu cầu tỉnh này phải chú trọng thu hút đầu tư phát triển năng lượng tái tạo, năng lượng sạch như điện gió ven bờ, điện gió ngoài khơi, điện mặt trời, điện sinh khối và nguồn năng lượng mới (hydrogen/amoniac xanh…); các dự án sản xuất thép quy mô lớn, đóng tàu, sản xuất thiết bị phụ trợ điện gió có công nghệ tiên tiến để nâng cấp xây dựng hạ tầng kỹ thuật sản xuất, thúc đẩy chuyển dịch kinh tế. Quy hoạch tỉnh Bình Định thời kỳ 2021 - 2030, tầm nhìn đến năm 2050 cũng đặt ra mục tiêu đến năm 2030, Bình Định trở thành tỉnh phát triển thuộc nhóm dẫn đầu vùng Bắc Trung Bộ và duyên hải Trung Bộ, là trung tâm công nghiệp chế biến, chế tạo, dịch vụ, du lịch và văn hóa phía Nam của vùng; trung tâm lớn của cả nước về phát triển kinh tế biển; trọng điểm du lịch quốc gia và quốc tế với hệ thống kết cấu hạ tầng kinh tế đồng bộ, hiện đại; kinh tế của tỉnh phát triển nhanh, bền vững và xanh dựa trên các trụ cột tăng trưởng công nghiệp, dịch vụ du lịch, cảng biển - logistics; nông nghiệp ứng dụng công nghệ cao; đô thị hóa; thực hiện thành công các mục tiêu chuyển đổi số, đổi mới sáng tạo, cải thiện mạnh mẽ môi trường đầu tư kinh doanh, trở thành điểm đến đầu tư hấp dẫn của các doanh nghiệp lớn trong và ngoài nước; chỉ số năng lực cạnh tranh cấp tỉnh thuộc nhóm cao của cả nước; kết cấu hạ tầng kinh tế - xã hội đồng bộ, hiện đại, hệ thống đô thị phát triển theo hướng đô thị thông minh, kết nối thuận tiện với các trung tâm kinh tế của vùng, cả nước và quốc tế.
- Ngữ cảnh 3: . Chủ tịch UBND tỉnh Quảng Ninh cho biết, tỉnh đặt mục tiêu hướng đến năm 2030 trở thành một tỉnh tiêu biểu của cả nước về mọi mặt; tỉnh kiểu mẫu giàu đẹp, văn minh, hiện đại, nâng cao đời sống mọi mặt của nhân dân; cực tăng trưởng của khu vực phía Bắc, một trong những trung tâm phát triển năng động, toàn diện; trung tâm du lịch quốc tế, trung tâm kinh tế biển, cửa ngõ của Vùng kinh tế trọng điểm Bắc Bộ và cả nước. Để đạt được những mục tiêu trên, tỉnh Quảng Ninh xác định sự đóng góp, quan tâm của cộng đồng doanh nghiệp, nhất là các doanh nghiệp hàng đầu Việt Nam “các sếu đầu đàn” là một trong những yếu tố then chốt quyết định. Do vậy, tỉnh Quảng Ninh rất mong nhận được sự quan tâm, nghiên cứu đầu tư hợp tác của các Doanh nghiệp hàng đầu Việt Nam trong thời gian tới, nhất là trong việc đầu tư các dự án có hàm lượng công nghệ cao, công nghệ tiên tiến, quản trị hiện đại, giá trị gia tăng cao, có tác động lan tỏa. Tỉnh Quảng Ninh cam kết tạo điều kiện thuận lợi nhất cho doanh nghiệp phát triển hơn nữa khi đầu tư kinh doanh trên địa bàn tỉnh; cam kết đồng hành, lắng nghe tiếng nói của cộng đồng doanh nghiệp, các nhà đầu tư; cùng trăn trở, trách nhiệm, giải quyết thấu đáo, vào cuộc thực chất, hiệu quả đối với từng khó khăn, vướng mắc với mục tiêu tăng cường niềm tin và nâng cao sự hài lòng của cá nhân, tổ chức, doanh nghiệp là thước đo đánh giá chất lượng phục vụ, chất lượng điều hành của cơ quan hành chính nhà nước từ cấp tỉnh đến các cấp cơ sở. Tỉnh Quảng Ninh sẽ triển khai một cách hiệu quả nhất Quy hoạch tỉnh, quy hoạch các địa phương; chuẩn bị mặt bằng sạch; quỹ đất tái định cư; đảm bảo nguồn cung điện, nước và viễn thông theo yêu cầu; xây dựng các chính sách đột phá để thu hút nguồn nhân lực chất lượng cao thông qua việc cải thiện nhà ở và cơ sở văn hóa để đảm bảo đời sống vật chất và tinh thần cho người lao động. Đồng thời bảo vệ quyền và lợi ích chính đáng, hợp pháp của nhà đầu tư, doanh nghiệp vào tỉnh; bảo đảm hài hòa lợi ích giữa nhà nước, nhà đầu tư, người lao động và người dân.'''
question = '''Theo quy hoạch tỉnh Tây Ninh thời kỳ 2021 - 2030, tầm nhìn đến năm 2050, tỉnh sẽ tập trung phát triển công nghiệp và dịch vụ du lịch. Trong bối cảnh cạnh tranh với các tỉnh lân cận, yếu tố nào được coi là quan trọng nhất để Tây Ninh có thể thu hút đầu tư và phát triển nhanh, toàn diện và bền vững?'''
'''Trả lời:
Trong ngữ cảnh được cung cấp, không có thông tin về quy hoạch tỉnh Tây Ninh thời kỳ 2021 - 2030, tầm nhìn đến năm 2050. Do đó, không có câu trả lời chính xác dựa trên các thông tin đã cung cấp. Nếu bạn có thêm thông tin về quy hoạch tỉnh Tây Ninh, vui lòng cung cấp để tôi có thể trả lời câu hỏi một cách chính xác và đầy đủ.
'''
```
#### 4. Tích hợp thông tin
```python
context = '''- Ngữ cảnh 1: 43/2020/nđ-cp Điều 3. Chế độ, chính sách đối với người tham gia thi hành án tử hình\n1. Người tham gia Đội thi hành án tử hình được hưởng chế độ bồi dưỡng bằng ba lần mức lương cơ sở khi thi hành án tử hình đối với 01 người và nghỉ dưỡng 10 ngày theo quy định chung về chế độ nghỉ dưỡng đối với cán bộ, chiến sĩ Công an nhân dân, Quân đội nhân dân.\n2. Người tham gia Hội đồng thi hành án tử hình, cán bộ quản giáo, người ghi âm, ghi hình, chụp ảnh, phiên dịch, thực hiện lăn tay người bị thi hành án tử hình, khâm liệm, mai táng tử thi được hưởng chế độ bồi dưỡng bằng một lần mức lương cơ sở khi thi hành án tử hình đối với 01 người.\n3. Người tham gia bảo đảm an ninh, trật tự; đại diện Ủy ban nhân dân cấp xã; Điều tra viên được hưởng chế độ bồi dưỡng bằng một phần hai mức lương cơ sở khi thi hành án tử hình đối với 01 người.
- Ngữ cảnh 2: 53/2010/qh12 Điều 60. Giải quyết việc xin nhận tử thi, hài cốt của người bị thi hành án tử hình\n1. Việc giải quyết nhận tử thi được thực hiện như sau:\na) Trước khi thi hành án tử hình, thân nhân hoặc người đại diện hợp pháp của người chấp hành án được làm đơn có xác nhận của Ủy ban nhân dân cấp xã nơi cư trú gửi Chánh án Tòa án đã xét xử sơ thẩm đề nghị giải quyết cho nhận tử thi của người chấp hành án để an táng; trường hợp người chấp hành án là người nước ngoài thì đơn phải có xác nhận của cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người chấp hành án mang quốc tịch và phải được dịch ra tiếng Việt. Đơn phải ghi rõ họ tên, địa chỉ người nhận tử thi, quan hệ với người chấp hành án; cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí;\nb) Chánh án Tòa án đã xét xử sơ thẩm thông báo bằng văn bản cho người có đơn đề nghị về việc cho nhận tử thi hoặc không cho nhận tử thi khi có căn cứ cho rằng việc nhận tử thi ảnh hưởng đến an ninh, trật tự, vệ sinh môi trường. Trường hợp người chấp hành án là người nước ngoài, thì Chánh án Tòa án đã xét xử sơ thẩm có trách nhiệm thông báo bằng văn bản cho Bộ Ngoại giao Việt Nam để thông báo cho cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người đó mang quốc tịch;\nc) Cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm thông báo cho người có đơn đề nghị ngay sau khi thi hành án để đến nhận tử thi về an táng. Việc giao nhận tử thi phải được thực hiện trong thời hạn 24 giờ kể từ khi thông báo và phải lập biên bản, có chữ ký của các bên giao, nhận; hết thời hạn này mà người có đơn đề nghị không đến nhận tử thi thì cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm an táng.\n2. Trường hợp không được nhận tử thi hoặc thân nhân của người bị thi hành án không có đơn đề nghị được nhận tử thi về an táng thì cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu tổ chức việc an táng. Sau 03 năm kể từ ngày thi hành án, thân nhân hoặc đại diện hợp pháp của người đã bị thi hành án được làm đơn có xác nhận của Ủy ban nhân dân cấp xã nơi cư trú đề nghị Cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu nơi đã thi hành án cho nhận hài cốt. Đơn đề nghị phải ghi rõ họ tên, địa chỉ người nhận hài cốt, quan hệ với người bị thi hành án; cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí. Trong thời hạn 07 ngày, kể từ ngày nhận được đơn, cơ quan thi hành án hình sự Công an cấp tỉnh, cơ quan thi hành án hình sự cấp quân khu có trách nhiệm xem xét, giải quyết.\nTrường hợp người bị thi hành án là người nước ngoài thì đơn đề nghị phải có xác nhận của cơ quan có thẩm quyền hoặc cơ quan đại diện ngoại giao tại Việt Nam của nước mà người bị thi hành án mang quốc tịch và phải được dịch ra tiếng Việt. Việc giải quyết cho nhận hài cốt do cơ quan quản lý thi hành án hình sự xem xét, quyết định.
- Ngữ cảnh 3: 53/2010/qh12 Điều 57. Chế độ quản lý giam giữ, ăn, ở, mặc, sinh hoạt, gửi và nhận thư, nhận đồ vật, tiền mặt, gặp thân nhân, chăm sóc y tế\nChế độ quản lý giam giữ, ăn, ở, mặc, sinh hoạt, gửi và nhận thư, nhận đồ vật, tiền mặt, gặp thân nhân, chăm sóc y tế đối với người bị kết án tử hình trong thời gian chờ thi hành án thực hiện theo quy định của pháp luật về tạm giam.
- Ngữ cảnh 4: 82/2011/nđ-cp Điều 9. Chi phí mai táng\nChi phí mai táng người bị thi hành án tử hình bao gồm: 01 quan tài bằng gỗ thường, 01 bộ quần áo thường, 04 m vải liệm, hương, nến, rượu, cồn để làm vệ sinh khi liệm tử thi và các chi phí mai táng khác.\nTrường hợp thân nhân hoặc người đại diện hợp pháp của người bị kết án tử hình được nhận tử thi về mai táng thì phải tự chịu chi phí đưa di chuyển tử thi và phải cam kết chấp hành đúng quy định của pháp luật về bảo đảm an ninh, trật tự.
- Ngữ cảnh 5: 82/2011/nđ-cp Điều 5. Chế độ, chính sách đối với người tham gia thi hành án tử hình\n1. Người tham gia Đội thi hành án tử hình và bác sĩ xác định tĩnh mạch được hưởng chế độ bồi dưỡng bằng hai lần mức lương tối thiểu chung cho mỗi lần thi hành án và được nghỉ dưỡng 10 ngày theo quy định chung về chế độ nghỉ dưỡng đối với cán bộ, chiến sĩ Công an nhân dân, Quân đội nhân dân.\n2. Người tham gia Hội đồng thi hành án tử hình, bác sĩ pháp y, cán bộ chuyên môn, người thực hiện lăn tay người bị thi hành án tử hình, khâm liệm, an táng tử thi được hưởng chế độ bồi dưỡng bằng một lần mức lương tối thiểu chung quy định cho mỗi lần thi hành án.\n3. Những người tham gia khác được hưởng chế độ bồi dưỡng bằng một phần hai mức lương tối thiểu cho mỗi lần thi hành án.'''
question = '''Tử tù chết sau khi tiêm thuốc độc mà không có thân nhân nhận xác thì được hỗ trợ mai táng như thế nào?'''
'''Trả lời:
Dựa trên ngữ cảnh 2 (53/2010/qh12 Điều 60), nếu tử tù chết sau khi tiêm thuốc độc mà không có thân nhân nhận xác, thì cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu sẽ tổ chức việc an táng. Cụ thể:
1. **An táng**: Cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu sẽ chịu trách nhiệm an táng tử tù nếu không có thân nhân hoặc người đại diện hợp pháp đề nghị nhận tử thi.
2. **Hài cốt sau 3 năm**: Sau 3 năm kể từ ngày thi hành án, nếu thân nhân hoặc đại diện hợp pháp của người đã bị thi hành án vẫn chưa đề nghị nhận hài cốt, họ có thể làm đơn đề nghị Cơ quan thi hành án hình sự Công an cấp tỉnh hoặc cơ quan thi hành án hình sự cấp quân khu nơi đã thi hành án cho nhận hài cốt. Đơn đề nghị phải ghi rõ họ tên, địa chỉ người nhận hài cốt, quan hệ với người bị thi hành án, cam kết bảo đảm yêu cầu về an ninh, trật tự, vệ sinh môi trường và tự chịu chi phí.
3. **Chi phí mai táng**: Theo ngữ cảnh 4 (82/2011/nđ-cp Điều 9), chi phí mai táng bao gồm:
- 01 quan tài bằng gỗ thường.
- 01 bộ quần áo thường.
- 04 m vải liệm.
- Hương, nến, rượu, cồn để làm vệ sinh khi liệm tử thi.
- Các chi phí mai táng khác.
Nếu thân nhân hoặc người đại diện hợp pháp đề nghị nhận tử thi về mai táng, họ phải tự chịu chi phí đưa di chuyển tử thi và cam kết chấp hành đúng quy định của pháp luật về bảo đảm an ninh, trật tự.
'''
```
- Ngữ cảnh 2 và ngữ cảnh 4 có chứa phần thông tin cho việc trả lời câu hỏi.
### 5.Xác định positive/negative
```python
system_prompt = "Bạn là một trợ lí Tiếng Việt nhiệt tình và trung thực. Hãy luôn trả lời một cách hữu ích nhất có thể."
template = '''Hãy xác định xem ngữ cảnh có chứa đầy đủ thông tin để trả lời câu hỏi hay không.
Chỉ cần đưa ra 1 trong 2 đáp án trong phần trả lời là "Có" hoặc "Không".
### Ngữ cảnh :
{context}
### Câu hỏi :
{question}
### Trả lời :'''
context = '''Công dụng thuốc Xelocapec Capecitabine là một hoạt chất gây độc chọn lọc với tế bào ung thư. Hoạt chất này có trong thuốc Xelocapec. Vậy thuốc Xelocapec có tác dụng gì và cần lưu ý những vấn đề nào khi điều trị bằng sản phẩm này? 1. Xelocapec là thuốc gì? Xelocapec chứa hoạt chất Capecitabine hàm lượng 500mg. Thuốc Xelocapec bào chế dạng viên nén bao phim và đóng gói mỗi hộp 3 vỉ x 10 viên. Xelocapec chứa hoạt chất Capecitabine là một dẫn chất Fluoropyrimidine carbamate với tác dụng gây độc chọn lọc với các tế bào ung thư . Mặc dù trên in vitro Capecitabine không cho thấy tác dụng độc tế bào nhưng trên in vivo, Xelocapec biến đổi liên tiếp thành chất gây độc tế bào là 5-fluorouracil (5-FU). Sự hình thành 5-FU tại khối u thông qua xúc tác một cách tối ưu của yếu tố tạo mạch liên quan là Thymidine phosphorylase, qua đó hạn chế tối đa mức độ ảnh hưởng đến nhu mô lành của 5-FU. 2. Thuốc Xelocapec có tác dụng gì? Thuốc Xelocapec được chỉ định điều trị đơn lẻ hoặc kết hợp với các liệu pháp điều trị ung thư. Xelocapec làm chậm hoặc ngăn chặn sự tăng trưởng của tế bào ung thư, do đó giảm kích thước khối u trong những trường hợp sau: Ung thư vú : Xelocapec phối hợp với Docetaxel được chỉ định điều trị ung thư vú thể tiến triển tại chỗ hoặc di căn sau khi đã thất bại với liệu pháp hóa trị; Ung thư đại trực tràng : Xelocapec được chỉ định hỗ trợ điều trị ung thư đại tràng sau phẫu thuật hoặc ung thư đại trực tràng di căn; Ung thư dạ dày : Xelocapec phối hợp với hợp chất platin được chỉ định điều trị khởi đầu cho những bệnh nhân ung thư dạ dày. Chống chỉ định của thuốc Xelocapec : Bệnh nhân quá mẫn cảm với Capecitabine hay các thành phần khác có trong Xelocapec ; Người có tiền sử gặp các phản ứng không mong muốn nghiêm trọng khi điều trị với Fluoropyrimidine; Người đang mang thai hoặc cho con bú; Suy thận nặng (độ thanh thải Creatinin <30mL/phút); Bệnh nhân đang điều trị ung thư với Sorivudin hoặc chất tương tự về mặt hóa học như Brivudin; Bệnh nhân thiếu hụt Dihydropyrimidin dehydrogenase; Bệnh nhân giảm số lượng bạch cầu hoặc tiểu cầu nặng; Suy gan nặng. 3. Liều dùng của thuốc Xelocapec Liều dùng của Xelocapec khi điều trị đơn lẻ: Ung thư đại trực tràng, ung thư vú: 1250mg/m2, uống 2 lần mỗi ngày trong thời gian 14 ngày, tiếp sau đó là 7 ngày ngưng thuốc. Liều Xelocapec trong điều trị phối hợp: Ung thư vú: Liều khởi đầu là 1250mg/m2, 2 lần uống mỗi ngày trong 2 tuần dùng phối hợp với Docetaxel, tiếp sau đó lá 1 tuần ngưng thuốc; Ung thư dạ dày, đại trực tràng: Liều khuyến cáo là 800-1000mg/m2/lần x 2 lần/ngày trong thời gian 2 tuần, sau đó 7 ngày ngưng thuốc hoặc 625mg/m2/lần x 2 lần mỗi ngày khi điều trị liên tục. Thuốc Xelocapec nên uống cùng với thức ăn, do đó thời điểm tốt nhất là trong vòng 30 phút sau bữa ăn. 4. Tác dụng phụ của thuốc Xelocapec Các triệu chứng bất thường như buồn nôn, nôn ói, giảm cảm giác ngon miệng, táo bón, cơ thể mệt mỏi, yếu sức, đau đầu, chóng mặt, khó ngủ có thể xảy ra trong thời gian dùng Xelocapec . Trong đó, tình trạng buồn nôn và nôn ói có thể nghiêm trọng nên đôi khi cần được bác sĩ chỉ định thuốc kiểm soát phù hợp. Tiêu chảy là một tác dụng phụ phổ biến khác của thuốc Xelocapec . Bệnh nhân cần uống nhiều nước khi điều trị bằng Xelocapec trừ khi bác sĩ có hướng dẫn khác. Nôn ói hoặc tiêu chảy kéo dài do thuốc Xelocapec có thể dẫn đến mất nước nghiêm trọng, vì vậy người bệnh hãy liên hệ ngay với bác sĩ nếu có các triệu chứng mất nước như giảm đi tiểu, khô miệng, tăng cảm giác khát nước hoặc chóng mặt. Tình trạng rụng tóc tạm thời xảy ra trong thời gian dùng thuốc Xelocapec và có thể hồi phục sau khi điều trị đã kết thúc. Một số bệnh nhân ghi nhận hiện tượng thay đổi móng tay tạm thời. Đối với nam giới và phụ nữ trong độ tuổi sinh đẻ, thuốc Xelocapec có thể ảnh hưởng đến khả năng có con của bệnh nhân. Bệnh nhân hãy tham khảo ý kiến bác sĩ để biết thêm chi tiết. Thuốc Xelocapec có thể làm giảm khả năng miễn dịch của cơ thể với các tác nhân nhiễm trùng, dẫn đến tăng nguy cơ mắc các bệnh nhiễm trùng nghiêm trọng (nhưng hiếm khi gây tử vong) hoặc làm cho bệnh nhiễm trùng hiện mắc nghiêm trọng hơn. Phản ứng dị ứng rất nghiêm trọng với thuốc Xelocapec rất hiếm khi xảy ra. Tuy nhiên, bệnh nhân hãy liên hệ với bác sĩ ngay lập tức nếu xuất hiện các triệu chứng của phản ứng dị ứng nghiêm trọng như phát ban, sưng ngứa mặt/lưỡi/họng, chóng mặt nghiêm trọng hoặc khó thở. 5. Tương tác thuốc của Xelocapec Hoạt chất Capecitabine trong thuốc Xelocapec có thể xảy ra tương tác thuốc nghiêm trọng với một số thuốc sau: Thuốc chống đông máu Coumarin: Trong một nghiên cứu tương tác lâm sàng, sau khi dùng Warfarin liều đơn 20mg kết hợp với Capecitabine làm tăng AUC của S-warfarin khoảng 57% và giá trị INR tăng 91%.'''
question = '''Tại sao Capecitabine trong Xelocapec không gây độc tế bào trên in vitro nhưng lại biến đổi thành 5-fluorouracil (5-FU) gây độc tế bào trên in vivo, và cơ chế nào giúp hạn chế ảnh hưởng đến nhu mô lành của 5-FU?'''
'''Trả lời: Có'''
```
**Next release**
Mô hình phiên bản tiếp theo mạnh mẽ hơn, giải quyết các dữ liệu phức tạp hơn sẽ được public sớm nhất có thể.
**Developer**
Member: Nguyễn Nho Trung, Nguyễn Nhật Quang
## Contact
**Email**: nguyennhotrung3004@gmail.com
**LinkedIn**: [Trung Nguyen Nho](https://www.linkedin.com/in/trung-nguyen-nho-604288227/)
## Citation
```Plaintext
@misc{ViRAG-Gen,
title={ViRAG-Gen: Towards a specialized LLM for RAG task in Vietnamese language.}},
author={Nguyen Nho Trung, Nguyen Nhat Quang},
year={2024},
publisher={Huggingface},
}
```
| [
"CHIA"
] |
PocketDoc/Dans-PersonalityEngine-V1.1.0-12b | PocketDoc | text-generation | [
"transformers",
"safetensors",
"mistral",
"text-generation",
"general-purpose",
"roleplay",
"storywriting",
"chemistry",
"biology",
"code",
"climate",
"axolotl",
"text-generation-inference",
"finetune",
"conversational",
"en",
"dataset:PocketDoc/Dans-MemoryCore-CoreCurriculum-Small",
"dataset:AquaV/Energetic-Materials-Sharegpt",
"dataset:AquaV/Chemical-Biological-Safety-Applications-Sharegpt",
"dataset:AquaV/US-Army-Survival-Sharegpt",
"dataset:AquaV/Resistance-Sharegpt",
"dataset:AquaV/Interrogation-Sharegpt",
"dataset:AquaV/Multi-Environment-Operations-Sharegpt",
"dataset:PocketDoc/Dans-Mathmaxx",
"dataset:PocketDoc/Dans-Mathmaxx-Numina-CoT",
"dataset:PJMixers/Math-Multiturn-1K-ShareGPT",
"dataset:PocketDoc/Dans-Benchmaxx",
"dataset:PocketDoc/Dans-Benchmaxx-COT",
"dataset:PocketDoc/Dans-Codemaxx-LeetCode",
"dataset:PocketDoc/Dans-Codemaxx-CodeFeedback-Conversations",
"dataset:PocketDoc/Dans-Codemaxx-CodeFeedback-SingleTurn",
"dataset:PocketDoc/Dans-Codemaxx-Bigcode-SelfInstruct",
"dataset:PocketDoc/Dans-Taskmaxx",
"dataset:PocketDoc/Dans-Taskmaxx-DataPrepper",
"dataset:PocketDoc/Dans-Taskmaxx-ConcurrentQA-Reworked",
"dataset:PocketDoc/Dans-Taskmaxx-TableGPT",
"dataset:PocketDoc/Dans-Taskmaxx-SciRIFF",
"dataset:PocketDoc/Dans-Taskmaxx-Edit",
"dataset:PocketDoc/Dans-Systemmaxx",
"dataset:PocketDoc/Dans-Toolmaxx-Agent",
"dataset:PocketDoc/Dans-Toolmaxx-ShellCommands",
"dataset:PocketDoc/Dans-Toolmaxx-Functions-Toolbench",
"dataset:PocketDoc/Dans-Toolmaxx-Functions-ToolACE",
"dataset:PocketDoc/Dans-Toolmaxx-Functions-apigen",
"dataset:PocketDoc/Dans-ASCIIMaxx-Wordart",
"dataset:PocketDoc/Dans-Prosemaxx-Gutenberg",
"dataset:PocketDoc/Dans-Prosemaxx-Cowriter-M",
"dataset:PocketDoc/Dans-Prosemaxx-Adventure",
"dataset:PocketDoc/Dans-Prosemaxx-Gryphe-GPT4o-WritingPrompts",
"dataset:PocketDoc/Dans-Assistantmaxx-Sharegpt",
"dataset:PocketDoc/Dans-Assistantmaxx-OpenAssistant2",
"dataset:PocketDoc/Dans-Assistantmaxx-Opus-Merge",
"dataset:PocketDoc/Dans-Assistantmaxx-sonnetorca-subset",
"dataset:PocketDoc/Dans-Assistantmaxx-sonnetorca-subset-2",
"dataset:PocketDoc/Dans-Assistantmaxx-NoRobots",
"dataset:PocketDoc/Dans-Assistantmaxx-Synthia",
"dataset:PocketDoc/Dans-Assistantmaxx-ASL",
"dataset:PocketDoc/Dans-Assistantmaxx-PersonaLLM-Opus",
"dataset:PocketDoc/Dans-Assistantmaxx-UnnaturalInstructions-GPT4",
"dataset:PocketDoc/Dans-Assistantmaxx-LongAlign",
"dataset:PocketDoc/Dans-Assistantmaxx-EvolKit",
"dataset:PocketDoc/Dans-Assistantmaxx-Camel-GPT4",
"dataset:PocketDoc/Dans-Assistantmaxx-Tulu3-IF",
"dataset:PocketDoc/Dans-Logicmaxx-Skunkworks",
"dataset:PocketDoc/Dans-Logicmaxx-SAT-AP",
"dataset:PocketDoc/Dans-Logicmaxx-Magpie-Ultra",
"dataset:PJMixers/grimulkan_theory-of-mind-ShareGPT",
"dataset:PJMixers/grimulkan_physical-reasoning-ShareGPT",
"dataset:PocketDoc/Dans-Personamaxx",
"dataset:PocketDoc/Dans-Personamaxx-Rainy",
"dataset:PocketDoc/Dans-Personamaxx-Aesir",
"dataset:PocketDoc/Dans-Kinomaxx-VanillaBackrooms",
"base_model:mistralai/Mistral-Nemo-Base-2407",
"base_model:finetune:mistralai/Mistral-Nemo-Base-2407",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-12-19T04:38:41Z" | 2024-12-21T04:40:19+00:00 | 442 | 36 | ---
base_model:
- mistralai/Mistral-Nemo-Base-2407
datasets:
- PocketDoc/Dans-MemoryCore-CoreCurriculum-Small
- AquaV/Energetic-Materials-Sharegpt
- AquaV/Chemical-Biological-Safety-Applications-Sharegpt
- AquaV/US-Army-Survival-Sharegpt
- AquaV/Resistance-Sharegpt
- AquaV/Interrogation-Sharegpt
- AquaV/Multi-Environment-Operations-Sharegpt
- PocketDoc/Dans-Mathmaxx
- PocketDoc/Dans-Mathmaxx-Numina-CoT
- PJMixers/Math-Multiturn-1K-ShareGPT
- PocketDoc/Dans-Benchmaxx
- PocketDoc/Dans-Benchmaxx-COT
- PocketDoc/Dans-Codemaxx-LeetCode
- PocketDoc/Dans-Codemaxx-CodeFeedback-Conversations
- PocketDoc/Dans-Codemaxx-CodeFeedback-SingleTurn
- PocketDoc/Dans-Codemaxx-Bigcode-SelfInstruct
- PocketDoc/Dans-Taskmaxx
- PocketDoc/Dans-Taskmaxx-DataPrepper
- PocketDoc/Dans-Taskmaxx-ConcurrentQA-Reworked
- PocketDoc/Dans-Taskmaxx-TableGPT
- PocketDoc/Dans-Taskmaxx-SciRIFF
- PocketDoc/Dans-Taskmaxx-Edit
- PocketDoc/Dans-Systemmaxx
- PocketDoc/Dans-Toolmaxx-Agent
- PocketDoc/Dans-Toolmaxx-ShellCommands
- PocketDoc/Dans-Toolmaxx-Functions-Toolbench
- PocketDoc/Dans-Toolmaxx-Functions-ToolACE
- PocketDoc/Dans-Toolmaxx-Functions-apigen
- PocketDoc/Dans-ASCIIMaxx-Wordart
- PocketDoc/Dans-Prosemaxx-Gutenberg
- PocketDoc/Dans-Prosemaxx-Cowriter-M
- PocketDoc/Dans-Prosemaxx-Adventure
- PocketDoc/Dans-Prosemaxx-Gryphe-GPT4o-WritingPrompts
- PocketDoc/Dans-Assistantmaxx-Sharegpt
- PocketDoc/Dans-Assistantmaxx-OpenAssistant2
- PocketDoc/Dans-Assistantmaxx-Opus-Merge
- PocketDoc/Dans-Assistantmaxx-sonnetorca-subset
- PocketDoc/Dans-Assistantmaxx-sonnetorca-subset-2
- PocketDoc/Dans-Assistantmaxx-NoRobots
- PocketDoc/Dans-Assistantmaxx-Synthia
- PocketDoc/Dans-Assistantmaxx-ASL
- PocketDoc/Dans-Assistantmaxx-PersonaLLM-Opus
- PocketDoc/Dans-Assistantmaxx-UnnaturalInstructions-GPT4
- PocketDoc/Dans-Assistantmaxx-LongAlign
- PocketDoc/Dans-Assistantmaxx-EvolKit
- PocketDoc/Dans-Assistantmaxx-Camel-GPT4
- PocketDoc/Dans-Assistantmaxx-Tulu3-IF
- PocketDoc/Dans-Logicmaxx-Skunkworks
- PocketDoc/Dans-Logicmaxx-SAT-AP
- PocketDoc/Dans-Logicmaxx-Magpie-Ultra
- PJMixers/grimulkan_theory-of-mind-ShareGPT
- PJMixers/grimulkan_physical-reasoning-ShareGPT
- PocketDoc/Dans-Personamaxx
- PocketDoc/Dans-Personamaxx-Rainy
- PocketDoc/Dans-Personamaxx-Aesir
- PocketDoc/Dans-Kinomaxx-VanillaBackrooms
language:
- en
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
tags:
- general-purpose
- roleplay
- storywriting
- chemistry
- biology
- code
- climate
- axolotl
- text-generation-inference
- finetune
model-index:
- name: Dans-PersonalityEngine-V1.1.0-12b
results: []
---
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
</head>
<div class="crt-container">
<div class="crt-case">
<div class="crt-inner-case">
<div class="crt-bezel">
<div class="terminal-screen">
<h2>Dans-PersonalityEngine-V1.1.0-12b</h2>
<p>This model series is intended to be multifarious in its capabilities and should be quite capable at both co-writing and roleplay as well as find itself quite at home performing sentiment analysis or summarization as part of a pipeline. It has been trained on a wide array of one shot instructions, multi turn instructions, tool use, role playing scenarios, text adventure games, co-writing, and much more.</p>
<h3>Key Details</h3>
<pre class="code-block">
BASE MODEL: mistralai/Mistral-Nemo-Base-2407
LICENSE: apache-2.0
LANGUAGE: English
CONTEXT LENGTH: 32768 tokens</pre>
<h3>Recommended Settings</h3>
<pre class="code-block">
TEMPERATURE: 1.0
TOP_P: 0.95
MIN_P: 0.05</pre>
<h3>Prompting Format</h3>
<p>The model uses standard "ChatML" format:</p>
<pre class="code-block">
<|im_start|>system
system prompt<|im_end|>
<|im_start|>user
Hi there!<|im_end|>
<|im_start|>assistant
Nice to meet you!<|im_end|></pre>
<h3>SillyTavern Templates</h3>
<details>
<summary>Context Template</summary>
<pre class="code-block">
{
"story_string": "<|im_start|>system\n{{#if system}}{{system}}\n{{/if}}{{#if wiBefore}}{{wiBefore}}\n{{/if}}{{#if description}}{{description}}\n{{/if}}{{#if personality}}{{char}}'s personality: {{personality}}\n{{/if}}{{#if scenario}}Scenario: {{scenario}}\n{{/if}}{{#if wiAfter}}{{wiAfter}}\n{{/if}}{{#if persona}}{{persona}}\n{{/if}}{{trim}}<|im_end|>\n",
"example_separator": "",
"chat_start": "",
"use_stop_strings": false,
"allow_jailbreak": false,
"always_force_name2": false,
"trim_sentences": false,
"include_newline": false,
"single_line": false,
"name": "Dan-ChatML"
}</pre>
</details>
<details>
<summary>Instruct Template</summary>
<pre class="code-block">
{
"system_prompt": "Write {{char}}'s actions and dialogue, user will write {{user}}'s.",
"input_sequence": "<|im_start|>user\n",
"output_sequence": "<|im_start|>assistant\n",
"first_output_sequence": "",
"last_output_sequence": "",
"system_sequence_prefix": "",
"system_sequence_suffix": "",
"stop_sequence": "<|im_end|>",
"wrap": false,
"macro": true,
"names": false,
"names_force_groups": false,
"activation_regex": "",
"skip_examples": false,
"output_suffix": "<|im_end|>\n",
"input_suffix": "<|im_end|>\n",
"system_sequence": "<|im_start|>system\n",
"system_suffix": "<|im_end|>\n",
"user_alignment_message": "",
"last_system_sequence": "",
"system_same_as_user": false,
"first_input_sequence": "",
"last_input_sequence": "",
"name": "Dan-ChatML"
}</pre>
</details>
<h3>Example Outputs</h3>
<details>
<summary>General Assistance Example</summary>
<p><em>System: Act as a personable virtual assistant.</em></p>
<p><strong>User:</strong> I'm going to be in Chicago soon can you help me work up some restaurant ideas?</p>
<p><strong>Assistant:</strong> I'd love to! Here are a few suggestions for restaurants in Chicago:</p>
<ul>
<li>For deep dish pizza, I highly recommend Lou Malnati's Pizzeria or Giordano's. Both are iconic Chicago institutions and serve some of the best deep dish pizza in the city.</li>
<li>If you're looking for a classic steakhouse experience, Gibson's Bar & Steakhouse is a great choice. They offer high-quality steaks, seafood, and an extensive wine list in a sophisticated atmosphere.</li>
<li>For a more casual dining experience, check out Portillo's Hot Dogs. They're known for their Chicago-style hot dogs, Italian beef sandwiches, and other classic American fare.</li>
<li>For something a bit more unique, try Alinea. It's a high-end, avant-garde restaurant that offers a truly one-of-a-kind dining experience with innovative cuisine and cutting-edge presentation techniques.</li>
</ul>
<p>Let me know if you need any other recommendations or have specific dietary restrictions or preferences!</p>
</details>
<details>
<summary>Basic Programming Example</summary>
<p><em>System: Answer the question or perform the task.</em></p>
<p><strong>User:</strong> Explain the bug in the following code:</p>
<pre class="code-block">
from time import sleep
from multiprocessing.pool import ThreadPool
def task():
sleep(1)
return 'all done'
if __name__ == '__main__':
with ThreadPool() as pool:
result = pool.apply_async(task())</pre>
<p><strong>Assistant:</strong> The bug in this code is that the `task()` function is called immediately when the `apply_async()` method is invoked, not when the worker process is ready to execute it.</p>
<p>Specifically, the line `result = pool.apply_async(task())` calls `task()` right away and passes its return value to `apply_async()`. The worker process will execute whatever `task()` returns at that moment, which is `'all done'` since `task()` completes execution almost immediately.</p>
<p>To fix this bug, you need to pass the `task` function itself to `apply_async()`, rather than calling `task()` beforehand. Here's the corrected code:</p>
<pre class="code-block">
from time import sleep
from multiprocessing.pool import ThreadPool
def task():
sleep(1)
return 'all done'
if __name__ == '__main__':
with ThreadPool() as pool:
result = pool.apply_async(task) # Pass the function, not the result
value = result.get() # Wait for the result
print(value)</pre>
</details>
<details>
<summary>Writing Example</summary>
<p><em>System: Take the users messages and use them as a writing prompt. Write 6 sentences.</em></p>
<p><strong>User:</strong> Alpha Centauri has fallen, to the bears. This is a point of view tale about a soldier on the ground.</p>
<p><strong>Assistant:</strong> "Sir, they are gone sir." The young corporal was out of breath.</p>
<p>The soldier looked up. His grey eyes had been locked on his holofac display for hours now. He was in his late thirties and wore the rank of colonel. He was tall, with a well-muscled but slightly over weight body.</p>
<p><strong>User:</strong> "Who?"</p>
<p><strong>Assistant:</strong> He turned and looked at the corporal.</p>
<p>"The bears. They left. We checked the entire perimeter. No bears to be found sir."</p>
<p>"Did you check the air?"</p>
<p>"Not the air, sir. We didn't see any shuttles. But we have drones out now."</p>
<p><strong>User:</strong> "*Sigh*"</p>
<p><strong>Assistant:</strong> "Ok, get me the status report. I need to know how many dead we have."</p>
<p>"Yes sir." The corporal turned to leave.</p>
<p>"Corporal…"</p>
<p>"Sir?"</p>
<p>"Get me a fresh cup of coffee. Black. I need the caffeine."</p>
</details>
<details>
<summary>Text Editing Example</summary>
<p><em>System: You are a professional editor with decades of experience, help the user with any task they have for you.</em></p>
<p><strong>User:</strong> Can you rewrite this to flow better? "I knew I probably shouldnt have done that but ohhh well i guess i just going to live with it since whats done is done"</p>
<p><strong>Assistant:</strong> I knew I probably shouldn't have done that, but oh well. I guess I just going to live with it since what's done is done.</p>
</details>
<h3>Datasets</h3>
<details>
<summary>Breakdown</summary>
<ul>
<li>Memory & Core Curriculum
<ul>
<li>Dans-MemoryCore-CoreCurriculum-Small - Base knowledge</li>
</ul>
</li>
<li>Military & Survival Knowledge
<ul>
<li>Energetic-Materials - Understanding of explosives and related chemistry</li>
<li>Chemical-Biological-Safety-Applications - Safety protocols, handling procedures, etc.</li>
<li>US-Army-Survival - Survival techniques and field craft</li>
<li>Resistance - Resistance operations and tactics</li>
<li>Interrogation - Interview and interrogation techniques</li>
<li>Multi-Environment-Operations - Operations across different environments</li>
</ul>
</li>
<li>Mathematics & Problem Solving
<ul>
<li>Dans-Mathmaxx - Core mathematics capabilities</li>
<li>Dans-Mathmaxx-Numina-CoT - Chain of thought mathematical reasoning</li>
<li>Math-Multiturn-1K-ShareGPT - Multi-turn math problem solving</li>
</ul>
</li>
<li>Benchmarking & Testing
<ul>
<li>Dans-Benchmaxx - Prepares model for "answer only" style benchmarks. Helps prevent the model from talking too much when the situation calls for it.</li>
<li>Dans-Benchmaxx-COT - The same but for COT then answer based testing.</li>
</ul>
</li>
<li>Programming & Code
<ul>
<li>Dans-Codemaxx-LeetCode - Programmatically produced from rosettacode</li>
<li>Dans-Codemaxx-CodeFeedback - Dataset focused on correction after producing incorrect code.</li>
<li>Dans-Codemaxx-Bigcode-SelfInstruct - Subset from the Bigcode SelfInstruct dataset</li>
</ul>
</li>
<li>Task Specific Training
<ul>
<li>Dans-Taskmaxx - General task handling</li>
<li>Dans-Taskmaxx-DataPrepper - Data preparation and cleaning</li>
<li>Dans-Taskmaxx-ConcurrentQA - Multi hop retrieval based tasks</li>
<li>Dans-Taskmaxx-TableGPT - Table data processing</li>
<li>Dans-Taskmaxx-SciRIFF - Scientific paper analysis</li>
<li>Dans-Taskmaxx-Edit - Text editing and revision</li>
</ul>
</li>
<li>System & Tool Usage
<ul>
<li>Dans-Toolmaxx-Agent - Tool usage and agent behavior</li>
<li>Dans-Toolmaxx-ShellCommands - Command line operations</li>
<li>Dans-Toolmaxx-Functions - API and function calling</li>
</ul>
</li>
<li>Creative & Writing
<ul>
<li>Dans-ASCIIMaxx-Wordart - ASCII word art creation</li>
<li>Dans-Prosemaxx-Gutenberg - Summary style prompt writing instructions sourced from the Gutenberg project.</li>
<li>Dans-Prosemaxx-Cowriter - Back and forth co-writing dataset sourced from human written literature</li>
<li>Dans-Prosemaxx-Adventure - Interactive fiction throwbacks such as Zork, Anchorhead, and Hunting the Ripper</li>
<li>Dans-Prosemaxx-WritingPrompts - Prompt based writing instructions</li>
</ul>
</li>
<li>Assistant & Personality
<ul>
<li>Dans-Assistantmaxx series - Various assistant behaviors and capabilities</li>
<li>Dans-Personamaxx series - Personality and character development</li>
<li>Dans-Logicmaxx series - Logical reasoning and problem solving</li>
</ul>
</li>
<li>Instruction Following
<ul>
<li>Dans-Systemmaxx - System message training data optimized to help the model reject bad patterns</li>
</ul>
</li>
</ul>
</details>
<h3>Training</h3>
<p>Full finetuned for 2 epochs on 1x H200 SXM (88 hours of training)</p>
<p class="badge-container">
<a href="https://github.com/OpenAccess-AI-Collective/axolotl" target="_blank" rel="noopener noreferrer">
<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>
</a>
</p>
<h3>Support Development</h3>
<p>Development is limited by funding and resources. To help support:</p>
<p>- Contact on HF</p>
<p>- Email: visuallyadequate@gmail.com</p>
<p class="coffee-container">
<a href="https://www.buymeacoffee.com/visually" target="_blank" rel="noopener noreferrer">
<img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" height="45" width="162">
</a>
</p>
</div>
</div>
</div>
</div>
</div>
<style>
@import url('https://fonts.googleapis.com/css2?family=VT323&display=swap');
.crt-container {
padding: 10px;
max-width: 1000px;
margin: 0 auto;
width: 95%;
}
.crt-case {
background: #e8d7c3;
border-radius: 10px;
padding: 15px;
box-shadow: inset -2px -2px 5px rgba(0,0,0,0.3), 2px 2px 5px rgba(0,0,0,0.2);
}
.crt-inner-case {
background: #e8d7c3;
border-radius: 8px;
padding: 3px;
box-shadow: inset -1px -1px 4px rgba(0,0,0,0.3), 1px 1px 4px rgba(0,0,0,0.2);
}
.crt-bezel {
background: linear-gradient(145deg, #1a1a1a, #2a2a2a);
padding: 15px;
border-radius: 5px;
border: 3px solid #0a0a0a;
position: relative;
box-shadow:
inset 0 0 20px rgba(0,0,0,0.5),
inset 0 0 4px rgba(0,0,0,0.4),
inset 2px 2px 4px rgba(255,255,255,0.05),
inset -2px -2px 4px rgba(0,0,0,0.8),
0 0 2px rgba(0,0,0,0.6),
-1px -1px 4px rgba(255,255,255,0.1),
1px 1px 4px rgba(0,0,0,0.3);
}
.crt-bezel::before {
content: '';
position: absolute;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: linear-gradient(45deg,
rgba(255,255,255,0.03) 0%,
rgba(255,255,255,0) 40%,
rgba(0,0,0,0.1) 60%,
rgba(0,0,0,0.2) 100%);
border-radius: 3px;
pointer-events: none;
}
.terminal-screen {
background: #111112;
padding: 20px;
border-radius: 15px;
position: relative;
overflow: hidden;
font-family: 'VT323', monospace;
font-size: clamp(12px, 1.5vw, 16px);
color: #e49b3e;
line-height: 1.4;
text-shadow: 0 0 2px #e49b3e;
animation: flicker 0.15s infinite;
filter: brightness(1.1) contrast(1.1);
box-shadow:
inset 0 0 30px rgba(0,0,0,0.9),
inset 0 0 8px rgba(0,0,0,0.8),
0 0 5px rgba(0,0,0,0.6);
max-width: 80ch;
margin: 0 auto;
}
.terminal-screen h2, .terminal-screen h3 {
font-size: clamp(16px, 2vw, 20px);
margin-bottom: 1em;
color: #e49b3e;
}
.terminal-screen pre.code-block {
font-size: clamp(11px, 1.3vw, 14px);
white-space: pre-wrap;
margin: 1em 0;
background-color: #1a1a1a;
padding: 1em;
border-radius: 4px;
color: #e49b3e;
}
.terminal-screen::before {
content: "";
position: absolute;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: linear-gradient(rgba(18, 16, 16, 0) 50%, rgba(0, 0, 0, 0.25) 50%), url('data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAADIAAAAyBAMAAADsEZWCAAAAGFBMVEUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA4o8JoAAAAB3RSTlMAGwQIEQMYADcPzwAAACJJREFUKM9jYBgFo2AU0Beg+A8YMCLxGYZCbNQEo4BaAAD5TQiR5wU9vAAAAABJRU5ErkJggg==');
background-size: 100% 2.5px;
animation: scan 1s linear infinite;
pointer-events: none;
z-index: 2;
}
.terminal-screen::after {
content: "";
position: absolute;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: radial-gradient(circle at center,
rgba(17, 17, 18, 0) 0%,
rgba(17, 17, 18, 0.2) 50%,
rgba(17, 17, 18, 0.15) 100%
);
border-radius: 20px;
animation: vignette-pulse 3s infinite;
pointer-events: none;
z-index: 1;
}
.terminal-screen details {
margin: 1em 0;
padding: 0.5em;
border: 1px solid #e49b3e;
border-radius: 4px;
}
.terminal-screen summary {
cursor: pointer;
font-weight: bold;
margin: -0.5em;
padding: 0.5em;
border-bottom: 1px solid #e49b3e;
color: #e49b3e;
}
.terminal-screen details[open] summary {
margin-bottom: 0.5em;
}
.badge-container, .coffee-container {
text-align: center;
margin: 1em 0;
}
.badge-container img, .coffee-container img {
max-width: 100%;
height: auto;
}
.terminal-screen a {
color: #e49b3e;
text-decoration: underline;
transition: opacity 0.2s;
}
.terminal-screen a:hover {
opacity: 0.8;
}
.terminal-screen strong, .terminal-screen em {
color: #f0f0f0; /* off-white color for user/system messages */
}
.terminal-screen p {
color: #f0f0f0; /* off-white color for assistant responses */
}
.terminal-screen p, .terminal-screen li {
color: #e49b3e;
}
.terminal-screen code,
.terminal-screen kbd,
.terminal-screen samp {
color: #e49b3e;
font-family: 'VT323', monospace;
text-shadow: 0 0 2px #e49b3e;
background-color: #1a1a1a;
padding: 0.2em 0.4em;
border-radius: 4px;
}
.terminal-screen pre.code-block,
.terminal-screen pre {
font-size: clamp(11px, 1.3vw, 14px);
white-space: pre-wrap;
margin: 1em 0;
background-color: #1a1a1a;
padding: 1em;
border-radius: 4px;
color: #e49b3e;
}
@keyframes flicker {
0% { opacity: 0.98; }
50% { opacity: 1; }
100% { opacity: 0.99; }
}
@keyframes scan {
0% { transform: translateY(0); }
100% { transform: translateY(4px); }
}
@keyframes vignette-pulse {
0% { opacity: 0.8; }
50% { opacity: 1; }
100% { opacity: 0.8; }
}
</style> | [
"CRAFT"
] |
ricepaper/vi-gemma-2b-RAG | ricepaper | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gemma",
"text-generation",
"text-generation-inference",
"retrieval-augmented-generation",
"unsloth",
"trl",
"sft",
"conversational",
"en",
"vi",
"base_model:unsloth/gemma-1.1-2b-it-bnb-4bit",
"base_model:finetune:unsloth/gemma-1.1-2b-it-bnb-4bit",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-07-16T16:51:03Z" | 2024-08-05T17:32:25+00:00 | 440 | 13 | ---
base_model: unsloth/gemma-1.1-2b-it-bnb-4bit
language:
- en
- vi
license: apache-2.0
tags:
- text-generation-inference
- retrieval-augmented-generation
- transformers
- unsloth
- gemma
- trl
- sft
---
## Model Card: vi-gemma-2b-RAG
### (English below)
### Tiếng Việt (Vietnamese)
**Mô tả mô hình:**
vi-gemma-2b-RAG là một mô hình ngôn ngữ lớn được tinh chỉnh từ mô hình cơ sở [google/gemma-1.1-2b-it](https://huggingface.co/google/gemma-1.1-2b-it) sử dụng kỹ thuật LoRA. Mô hình được huấn luyện trên tập dữ liệu tiếng Việt với mục tiêu cải thiện khả năng xử lý ngôn ngữ tiếng Việt và nâng cao hiệu suất cho các tác vụ truy xuất thông tin mở (Retrieval Augmented Generation - RAG).
**Mục đích sử dụng:**
Mô hình vi-gemma-2b-RAG phù hợp cho các tác vụ sau:
* Trả lời câu hỏi dựa trên ngữ cảnh tiếng Việt.
* Tóm tắt văn bản tiếng Việt.
* Dịch máy tiếng Việt.
* Và các tác vụ tạo văn bản tiếng Việt khác.
**Giới hạn:**
Mặc dù đã được tinh chỉnh cho tiếng Việt, vi-gemma-2b-RAG vẫn có thể gặp phải một số hạn chế:
* Có thể tạo ra thông tin sai lệch hoặc không chính xác.
* Có thể thể hiện thành kiến hoặc quan điểm không phù hợp.
* Hiệu suất có thể bị ảnh hưởng bởi chất lượng của dữ liệu đầu vào.
**Cách sử dụng:**
Dưới đây chúng tôi chia sẻ một số đoạn mã về cách bắt đầu nhanh chóng để sử dụng mô hình. Trước tiên, hãy đảm bảo đã cài đặt `pip install -U transformers`, sau đó sao chép đoạn mã từ phần có liên quan đến usecase của bạn.
Chúng tôi khuyến nghị sử dụng `torch.bfloat16` làm mặc định.
```python
# pip install transformers torch accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
# Khởi tạo tokenizer và model từ checkpoint đã lưu
tokenizer = AutoTokenizer.from_pretrained("himmeow/vi-gemma-2b-RAG")
model = AutoModelForCausalLM.from_pretrained(
"himmeow/vi-gemma-2b-RAG",
device_map="auto",
torch_dtype=torch.bfloat16
)
# Sử dụng GPU nếu có
if torch.cuda.is_available():
model.to("cuda")
# Định dạng prompt cho model
prompt = """
### Instruction and Input:
Dựa vào ngữ cảnh/tài liệu sau:
{}
Hãy trả lời câu hỏi: {}
### Response:
{}
"""
# Chuẩn bị dữ liệu đầu vào
input_data = """
Short Tandem Repeats (STRs) là các trình tự DNA lặp lại ngắn (2- 6 nucleotides) xuất hiện phổ biến trong hệ gen của con người. Các trình tự này có tính đa hình rất cao trong tự nhiên, điều này khiến các STRs trở thành những markers di truyền rất quan trọng trong nghiên cứu bản đồ gen người và chuẩn đoán bệnh lý di truyền cũng như xác định danh tính trong lĩnh vực pháp y.
Các STRs trở nên phổ biến tại các phòng xét nghiệm pháp y bởi vì việc nhân bản và phân tích STRs chỉ cần lượng DNA rất thấp ngay cả khi ở dạng bị phân hủy việc đinh danh vẫn có thể được thực hiện thành công. Hơn nữa việc phát hiện và đánh giá sự nhiễm DNA mẫu trong các mẫu vật có thể được giải quyết nhanh với kết quả phân tích STRs. Ở Hoa Kỳ hiện nay, từ bộ 13 markers nay đã tăng lên 20 markers chính đang được sử dụng để tạo ra một cơ sở dữ liệu DNA trên toàn đất nước được gọi là The FBI Combined DNA Index System (Expaned CODIS).
CODIS và các cơ sử dữ liệu DNA tương tự đang được sử dụng thực sự thành công trong việc liên kết các hồ sơ DNA từ các tội phạm và các bằng chứng hiện trường vụ án. Kết quả định danh STRs cũng được sử dụng để hỗ trợ hàng trăm nghìn trường hợp xét nghiệm huyết thống cha con mỗi năm'
"""
query = "Hãy cho tôi biết một số tính chất của STRs được dùng để làm gì?"
# Định dạng input text
input_text = prompt.format(input_data, query," ")
# Mã hóa input text thành input ids
input_ids = tokenizer(input_text, return_tensors="pt")
# Sử dụng GPU cho input ids nếu có
if torch.cuda.is_available():
input_ids = input_ids.to("cuda")
# Tạo văn bản bằng model
outputs = model.generate(
**input_ids,
max_new_tokens=500,
no_repeat_ngram_size=5, # Ngăn chặn lặp lại các cụm từ 5 gram
# do_sample=True, # Kích hoạt chế độ tạo văn bản dựa trên lấy mẫu. Trong chế độ này, model sẽ chọn ngẫu nhiên token tiếp theo dựa trên xác suất được tính từ phân phối xác suất của các token.
# temperature=0.7, # Giảm temperature để kiểm soát tính ngẫu nhiên
# early_stopping=True, # Dừng tạo văn bản khi tìm thấy kết thúc phù hợp
)
# Giải mã và in kết quả
print(tokenizer.decode(outputs[0]))
'''
<bos>
### Instruction and Input:
Dựa vào ngữ cảnh/tài liệu sau:
Short Tandem Repeats (STRs) là các trình tự DNA lặp lại ngắn (2- 6 nucleotides) xuất hiện phổ biến trong hệ gen của con người. Các trình tự này có tính đa hình rất cao trong tự nhiên, điều này khiến các STRs trở thành những markers di truyền rất quan trọng trong nghiên cứu bản đồ gen người và chuẩn đoán bệnh lý di truyền cũng như xác định danh tính trong lĩnh vực pháp y.
Các STRs trở nên phổ biến tại các phòng xét nghiệm pháp y bởi vì việc nhân bản và phân tích STRs chỉ cần lượng DNA rất thấp ngay cả khi ở dạng bị phân hủy việc đinh danh vẫn có thể được thực hiện thành công. Hơn nữa việc phát hiện và đánh giá sự nhiễm DNA mẫu trong các mẫu vật có thể được giải quyết nhanh với kết quả phân tích STRs. Ở Hoa Kỳ hiện nay, từ bộ 13 markers nay đã tăng lên 20 markers chính đang được sử dụng để tạo ra một cơ sở dữ liệu DNA trên toàn đất nước được gọi là The FBI Combined DNA Index System (Expaned CODIS).
CODIS và các cơ sử dữ liệu DNA tương tự đang được sử dụng thực sự thành công trong việc liên kết các hồ sơ DNA từ các tội phạm và các bằng chứng hiện trường vụ án. Kết quả định danh STRs cũng được sử dụng để hỗ trợ hàng trăm nghìn trường hợp xét nghiệm huyết thống cha con mỗi năm'
Hãy trả lời câu hỏi: Hãy cho tôi biết một số tính chất của STRs được dùng để làm gì?
### Response:
STRs được sử dụng để xác định danh tính, chuẩn đoán bệnh lý và xác định bệnh lý di truyền.
<eos>
'''
```
**Huấn luyện:**
* **Mô hình cơ sở:** google/gemma-1.1-2b-it
* **Tập dữ liệu:** lamhieu/mabrycodes_dialogue_vi
* **Phương pháp tinh chỉnh:** LoRA, PEFT với Unsloth
## Model Card: vi-gemma-2b-RAG
### English
**Model Description:**
vi-gemma-2b-RAG is a large language model fine-tuned from the base model [google/gemma-1.1-2b-it](https://huggingface.co/google/gemma-1.1-2b-it) using LoRA. The model is trained on a Vietnamese dataset to improve its Vietnamese language processing capabilities and enhance its performance for Retrieval Augmented Generation (RAG) tasks.
**Intended Use:**
The vi-gemma-2b-RAG model is suitable for tasks such as:
* Vietnamese question answering.
* Vietnamese text summarization.
* Vietnamese machine translation.
* And other Vietnamese text generation tasks.
**Limitations:**
While fine-tuned for Vietnamese, vi-gemma-2b-RAG may still have some limitations:
* It may generate incorrect or misleading information.
* It may exhibit biases or inappropriate opinions.
* Its performance may be affected by the quality of the input data.
**How to Use:**
### Usage
Below we share some code snippets on how to get quickly started with running the model. First make sure to `pip install -U transformers`, then copy the snippet from the section that is relevant for your usecase.
We recommend `torch.bfloat16` as the default dtype.
```python
# pip install transformers torch accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
# Initialize the tokenizer and model from the saved checkpoint
tokenizer = AutoTokenizer.from_pretrained("himmeow/vi-gemma-2b-RAG")
model = AutoModelForCausalLM.from_pretrained(
"himmeow/vi-gemma-2b-RAG",
device_map="auto",
torch_dtype=torch.bfloat16
)
# Use GPU if available
if torch.cuda.is_available():
model.to("cuda")
# Define the prompt format for the model
prompt = """
### Instruction and Input:
Based on the following context/document:
{}
Please answer the question: {}
### Response:
{}
"""
# Prepare the input data
input_data = """
Short Tandem Repeats (STRs) are short (2-6 nucleotides) repeating DNA sequences that are widespread in the human genome. These sequences are highly polymorphic in nature, which makes STRs very important genetic markers in human gene mapping and diagnosis of hereditary diseases as well as identification in the field of forensics.
STRs have become popular in forensic laboratories because the replication and analysis of STRs requires very small amounts of DNA, even in decomposed form, identification can still be performed successfully. Furthermore, the detection and assessment of sample DNA contamination in specimens can be quickly resolved with STR analysis results. In the United States today, the set of 13 markers has now been increased to 20 main markers being used to create a nationwide DNA database called The FBI Combined DNA Index System (Expaned CODIS).
CODIS and similar DNA databases are being used very successfully in linking DNA records from criminals and crime scene evidence. STR identification results are also used to support hundreds of thousands of paternity test cases each year.'
"""
query = "Tell me what are some properties of STRs used for?"
# Format the input text
input_text = prompt.format(input_data, query," ")
# Encode the input text into input ids
input_ids = tokenizer(input_text, return_tensors="pt")
# Use GPU for input ids if available
if torch.cuda.is_available():
input_ids = input_ids.to("cuda")
# Generate text using the model
outputs = model.generate(
**input_ids,
max_new_tokens=500, # Limit the number of tokens generated
no_repeat_ngram_size=5, # Prevent repetition of 5-gram phrases
# do_sample=True,
# temperature=0.7, # Adjust the randomness of the generated text
# early_stopping=True, # Stop generating text when a suitable ending is found
)
# Decode and print the results
print(tokenizer.decode(outputs[0]))
```
**Training:**
* **Base Model:** google/gemma-1.1-2b-it
* **Dataset:** lamhieu/mabrycodes_dialogue_vi
* **Fine-tuning Method:** LoRA, PEFT and Unsloth
**Using example repository:** https://github.com/Martincrux/Vietnamese-RAG-system-building-with-vi-gemma-2b-RAG-and-halong_embedding
# Uploaded model
- **Developed by:** [hiieu](https://huggingface.co/hiieu), [himmeow the coder](https://huggingface.co/himmeow), [cuctrinh](https://www.linkedin.com/in/trinh-cuc-5722832b6)
- **License:** apache-2.0
- **Finetuned from model :** unsloth/gemma-1.1-2b-it-bnb-4bit
This gemma model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
| [
"CHIA"
] |
TheBloke/juanako-7B-v1-GGUF | TheBloke | null | [
"transformers",
"gguf",
"mistral",
"alignment-handbook",
"generated_from_trainer",
"dataset:HuggingFaceH4/ultrafeedback_binarized",
"base_model:fblgit/juanako-7b-v1",
"base_model:quantized:fblgit/juanako-7b-v1",
"license:artistic-2.0",
"region:us"
] | "2023-11-25T01:29:52Z" | 2023-11-25T01:33:55+00:00 | 439 | 1 | ---
base_model: fblgit/juanako-7b-v1
datasets:
- HuggingFaceH4/ultrafeedback_binarized
license: artistic-2.0
tags:
- alignment-handbook
- generated_from_trainer
inference: false
model_creator: FBL
model_type: mistral
prompt_template: '<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
'
quantized_by: TheBloke
model-index:
- name: juanako-7b-v1
results: []
---
<!-- markdownlint-disable MD041 -->
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# Juanako 7B V1 - GGUF
- Model creator: [FBL](https://huggingface.co/fblgit)
- Original model: [Juanako 7B V1](https://huggingface.co/fblgit/juanako-7b-v1)
<!-- description start -->
## Description
This repo contains GGUF format model files for [FBL's Juanako 7B V1](https://huggingface.co/fblgit/juanako-7b-v1).
These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
<!-- description end -->
<!-- README_GGUF.md-about-gguf start -->
### About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplete list of clients and libraries that are known to support GGUF:
* [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
* [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
* [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
* [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
* [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
* [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server.
* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
* [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
<!-- README_GGUF.md-about-gguf end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/juanako-7B-v1-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/juanako-7B-v1-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF)
* [FBL's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/fblgit/juanako-7b-v1)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: ChatML
```
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
<!-- prompt-template end -->
<!-- compatibility_gguf start -->
## Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
## Explanation of quantisation methods
<details>
<summary>Click to see details</summary>
The new methods available are:
* GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
* GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
* GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
* GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
</details>
<!-- compatibility_gguf end -->
<!-- README_GGUF.md-provided-files start -->
## Provided files
| Name | Quant method | Bits | Size | Max RAM required | Use case |
| ---- | ---- | ---- | ---- | ---- | ----- |
| [juanako-7b-v1.Q2_K.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q2_K.gguf) | Q2_K | 2 | 3.08 GB| 5.58 GB | smallest, significant quality loss - not recommended for most purposes |
| [juanako-7b-v1.Q3_K_S.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q3_K_S.gguf) | Q3_K_S | 3 | 3.16 GB| 5.66 GB | very small, high quality loss |
| [juanako-7b-v1.Q3_K_M.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q3_K_M.gguf) | Q3_K_M | 3 | 3.52 GB| 6.02 GB | very small, high quality loss |
| [juanako-7b-v1.Q3_K_L.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q3_K_L.gguf) | Q3_K_L | 3 | 3.82 GB| 6.32 GB | small, substantial quality loss |
| [juanako-7b-v1.Q4_0.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q4_0.gguf) | Q4_0 | 4 | 4.11 GB| 6.61 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| [juanako-7b-v1.Q4_K_S.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q4_K_S.gguf) | Q4_K_S | 4 | 4.14 GB| 6.64 GB | small, greater quality loss |
| [juanako-7b-v1.Q4_K_M.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q4_K_M.gguf) | Q4_K_M | 4 | 4.37 GB| 6.87 GB | medium, balanced quality - recommended |
| [juanako-7b-v1.Q5_0.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q5_0.gguf) | Q5_0 | 5 | 5.00 GB| 7.50 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| [juanako-7b-v1.Q5_K_S.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q5_K_S.gguf) | Q5_K_S | 5 | 5.00 GB| 7.50 GB | large, low quality loss - recommended |
| [juanako-7b-v1.Q5_K_M.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q5_K_M.gguf) | Q5_K_M | 5 | 5.13 GB| 7.63 GB | large, very low quality loss - recommended |
| [juanako-7b-v1.Q6_K.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q6_K.gguf) | Q6_K | 6 | 5.94 GB| 8.44 GB | very large, extremely low quality loss |
| [juanako-7b-v1.Q8_0.gguf](https://huggingface.co/TheBloke/juanako-7B-v1-GGUF/blob/main/juanako-7b-v1.Q8_0.gguf) | Q8_0 | 8 | 7.70 GB| 10.20 GB | very large, extremely low quality loss - not recommended |
**Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
<!-- README_GGUF.md-provided-files end -->
<!-- README_GGUF.md-how-to-download start -->
## How to download GGUF files
**Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
* LM Studio
* LoLLMS Web UI
* Faraday.dev
### In `text-generation-webui`
Under Download Model, you can enter the model repo: TheBloke/juanako-7B-v1-GGUF and below it, a specific filename to download, such as: juanako-7b-v1.Q4_K_M.gguf.
Then click Download.
### On the command line, including multiple files at once
I recommend using the `huggingface-hub` Python library:
```shell
pip3 install huggingface-hub
```
Then you can download any individual model file to the current directory, at high speed, with a command like this:
```shell
huggingface-cli download TheBloke/juanako-7B-v1-GGUF juanako-7b-v1.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
<details>
<summary>More advanced huggingface-cli download usage</summary>
You can also download multiple files at once with a pattern:
```shell
huggingface-cli download TheBloke/juanako-7B-v1-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
```
For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
```shell
pip3 install hf_transfer
```
And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
```shell
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/juanako-7B-v1-GGUF juanako-7b-v1.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
</details>
<!-- README_GGUF.md-how-to-download end -->
<!-- README_GGUF.md-how-to-run start -->
## Example `llama.cpp` command
Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
```shell
./main -ngl 32 -m juanako-7b-v1.Q4_K_M.gguf --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
```
Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change `-c 2048` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.
If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
## How to run in `text-generation-webui`
Further instructions can be found in the text-generation-webui documentation, here: [text-generation-webui/docs/04 ‐ Model Tab.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/04%20%E2%80%90%20Model%20Tab.md#llamacpp).
## How to run from Python code
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries.
### How to load this model in Python code, using ctransformers
#### First install the package
Run one of the following commands, according to your system:
```shell
# Base ctransformers with no GPU acceleration
pip install ctransformers
# Or with CUDA GPU acceleration
pip install ctransformers[cuda]
# Or with AMD ROCm GPU acceleration (Linux only)
CT_HIPBLAS=1 pip install ctransformers --no-binary ctransformers
# Or with Metal GPU acceleration for macOS systems only
CT_METAL=1 pip install ctransformers --no-binary ctransformers
```
#### Simple ctransformers example code
```python
from ctransformers import AutoModelForCausalLM
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = AutoModelForCausalLM.from_pretrained("TheBloke/juanako-7B-v1-GGUF", model_file="juanako-7b-v1.Q4_K_M.gguf", model_type="mistral", gpu_layers=50)
print(llm("AI is going to"))
```
## How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
* [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
* [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
<!-- README_GGUF.md-how-to-run end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
<!-- original-model-card start -->
# Original model card: FBL's Juanako 7B V1
# juanako-7b-v1
This model is a fine-tuned version of [fblgit/zephyr-lora-dpo-b1](https://huggingface.co/fblgit/zephyr-lora-dpo-b1) on the HuggingFaceH4/ultrafeedback_binarized dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4594
- Rewards/chosen: -1.1095
- Rewards/rejected: -2.3132
- Rewards/accuracies: 0.7964
- Rewards/margins: 1.2037
- Logps/rejected: -220.0052
- Logps/chosen: -217.5506
- Logits/rejected: -2.5535
- Logits/chosen: -2.7973
** Please feel free to run more tests and commit the results. Also if you are interested to participate in [UNA's paper research or GPU sponsorship](mailto:info@fblnet.net) **
## Model description
**It seems to outperforms the original Zephyr in most of the tasks.**
I trained Juanako with the same datasets and trainer from [alignment-handbook/zephyr-7b-sft-lora](https://huggingface.co/alignment-handbook/zephyr-7b-sft-lora)
* 1 epoch on DPO with transformers-UNA, the result is [fblgit/zephyr-lora-dpo-b1](https://huggingface.co/fblgit/zephyr-lora-dpo-b1) after merge using FastChat converter.
* finally 1 epoch on DPO with transformers-UNA to [fblgit/zephyr-lora-dpo-b1](https://huggingface.co/fblgit/zephyr-lora-dpo-b1).
Some other experiments were performed as well to test transformers-UNA capabilities on diverse scenarios and models.
**This is a complete version of the model, the result of converting LoRa's**
## Intended uses & limitations
Research purposes.
## Training and evaluation data
alignment-handbook DPO with UNA on top of the SFT lora.
### Evaluation lm-evaluation-harness
#### GSM8K
```
hf (pretrained=/root/juanako-7b-v1-beta,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 3, batch_size: 4
```
|Tasks|Version| Filter | Metric |Value | |Stderr|
|-----|-------|----------|-----------|-----:|---|-----:|
|gsm8k|Yaml |get-answer|exact_match|0.4556|± |0.0137|
#### 0-Shot
```
hf (pretrained=fblgit/juanako-7b-v1,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 0, batch_size: 8
```
| Tasks |Version|Filter| Metric | Value | |Stderr|
|-------------------|-------|------|-----------|------:|---|-----:|
|arc_challenge |Yaml |none |acc | 0.5691|± |0.0145|
| | |none |acc_norm | 0.6041|± |0.0143|
|arc_easy |Yaml |none |acc | 0.8363|± |0.0076|
| | |none |acc_norm | 0.8161|± |0.0079|
|hellaswag |Yaml |none |acc | 0.6554|± |0.0047|
| | |none |acc_norm | 0.8411|± |0.0036|
|boolq |Yaml |none |acc | 0.8355|± |0.0065|
|lambada |N/A |none |perplexity | 3.3607|± |0.1398|
| | |none |acc | 0.7309|± |0.0137|
|piqa |Yaml |none |acc | 0.8194|± |0.0090|
| | |none |acc_norm | 0.8335|± |0.0087|
|sciq |Yaml |none |acc | 0.9480|± |0.0070|
| | |none |acc_norm | 0.8960|± |0.0097|
|truthfulqa |N/A |none |bleu_max |26.0803|± |0.6528|
| - truthfulqa_mc1 |Yaml |none |acc | 0.4198|± |0.0173|
| - truthfulqa_mc2 |Yaml |none |acc | 0.5847|± |0.0153|
|winogrande |Yaml |none |acc | 0.7609|± |0.0120|
#### 1-Shot
```
hf (pretrained=fblgit/juanako-7b-v1,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 1, batch_size: 8
```
| Tasks |Version|Filter| Metric | Value | |Stderr|
|-------------------|-------|------|-----------|------:|---|-----:|
|arc_challenge |Yaml |none |acc | 0.6084|± |0.0143|
| | |none |acc_norm | 0.6357|± |0.0141|
|arc_easy |Yaml |none |acc | 0.8645|± |0.0070|
| | |none |acc_norm | 0.8645|± |0.0070|
|hellaswag |Yaml |none |acc | 0.6475|± |0.0048|
| | |none |acc_norm | 0.8372|± |0.0037|
|boolq |Yaml |none |acc | 0.8609|± |0.0061|
|lambada |N/A |none |perplexity | 3.5484|± |0.1034|
| | |none |acc | 0.7207|± |0.0107|
|piqa |Yaml |none |acc | 0.8259|± |0.0088|
| | |none |acc_norm | 0.8384|± |0.0086|
|sciq |Yaml |none |acc | 0.9730|± |0.0051|
| | |none |acc_norm | 0.9740|± |0.0050|
|truthfulqa |N/A |none |bleu_max |18.9814|± |0.4805|
| | |none |acc | 0.4856|± |0.0521|
| - truthfulqa_mc1 |Yaml |none |acc | 0.4333|± |0.0173|
| - truthfulqa_mc2 |Yaml |none |acc | 0.5903|± |0.0153|
|winogrande |Yaml |none |acc | 0.7609|± |0.0120|
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 12
- gradient_accumulation_steps: 16
- total_train_batch_size: 192
- total_eval_batch_size: 12
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.01
- num_epochs: 1
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.4966 | 0.15 | 50 | 0.4893 | -1.1759 | -2.2914 | 0.7485 | 1.1155 | -219.7872 | -218.2148 | -2.5450 | -2.7884 |
| 0.4522 | 0.31 | 100 | 0.4808 | -0.8099 | -1.8893 | 0.7784 | 1.0794 | -215.7659 | -214.5544 | -2.5644 | -2.8095 |
| 0.5048 | 0.46 | 150 | 0.4706 | -1.0526 | -2.1412 | 0.7725 | 1.0887 | -218.2852 | -216.9814 | -2.5638 | -2.8089 |
| 0.4853 | 0.62 | 200 | 0.4640 | -1.0787 | -2.2821 | 0.7725 | 1.2034 | -219.6941 | -217.2426 | -2.5460 | -2.7891 |
| 0.4639 | 0.77 | 250 | 0.4636 | -1.2348 | -2.4583 | 0.8084 | 1.2235 | -221.4559 | -218.8034 | -2.5533 | -2.7970 |
| 0.4634 | 0.93 | 300 | 0.4601 | -1.1370 | -2.3243 | 0.7964 | 1.1873 | -220.1163 | -217.8257 | -2.5540 | -2.7977 |
| - | 1.00 | 300 | 0.4594 | -1.1095 | -2.3132 | 0.7964 | 1.2037 | -220.0052 | -217.5506 | -2.5535 | -2.7973 |
### Framework versions
- Transformers 4.35.0-UNA
- Pytorch 2.1.0
- Datasets 2.14.6
- Tokenizers 0.14.1
## MMLU Results
#### 1-Shot
```
hf (pretrained=fblgit/juanako-7b-v1,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 1, batch_size: 1
```
| Tasks |Version|Filter|Metric|Value | |Stderr|
|---------------------------------------|-------|------|------|-----:|---|-----:|
|mmlu |N/A |none |acc |0.6085|± |0.1321|
| - humanities |N/A |none |acc |0.5405|± |0.1478|
| - formal_logic |Yaml |none |acc |0.4206|± |0.0442|
| - high_school_european_history |Yaml |none |acc |0.7576|± |0.0335|
| - high_school_us_history |Yaml |none |acc |0.8186|± |0.0270|
| - high_school_world_history |Yaml |none |acc |0.7890|± |0.0266|
| - international_law |Yaml |none |acc |0.7438|± |0.0398|
| - jurisprudence |Yaml |none |acc |0.8056|± |0.0383|
| - logical_fallacies |Yaml |none |acc |0.7791|± |0.0326|
| - moral_disputes |Yaml |none |acc |0.7023|± |0.0246|
| - moral_scenarios |Yaml |none |acc |0.2145|± |0.0137|
| - philosophy |Yaml |none |acc |0.7074|± |0.0258|
| - prehistory |Yaml |none |acc |0.7377|± |0.0245|
| - professional_law |Yaml |none |acc |0.4361|± |0.0127|
| - world_religions |Yaml |none |acc |0.8421|± |0.0280|
| - other |N/A |none |acc |0.6894|± |0.1091|
| - business_ethics |Yaml |none |acc |0.5600|± |0.0499|
| - clinical_knowledge |Yaml |none |acc |0.6981|± |0.0283|
| - college_medicine |Yaml |none |acc |0.6185|± |0.0370|
| - global_facts |Yaml |none |acc |0.3300|± |0.0473|
| - human_aging |Yaml |none |acc |0.6726|± |0.0315|
| - management |Yaml |none |acc |0.8058|± |0.0392|
| - marketing |Yaml |none |acc |0.8419|± |0.0239|
| - medical_genetics |Yaml |none |acc |0.7200|± |0.0451|
| - miscellaneous |Yaml |none |acc |0.8033|± |0.0142|
| - nutrition |Yaml |none |acc |0.7288|± |0.0255|
| - professional_accounting |Yaml |none |acc |0.4929|± |0.0298|
| - professional_medicine |Yaml |none |acc |0.6801|± |0.0283|
| - virology |Yaml |none |acc |0.5000|± |0.0389|
| - social_sciences |N/A |none |acc |0.7195|± |0.0676|
| - econometrics |Yaml |none |acc |0.5000|± |0.0470|
| - high_school_geography |Yaml |none |acc |0.7879|± |0.0291|
| - high_school_government_and_politics|Yaml |none |acc |0.8601|± |0.0250|
| - high_school_macroeconomics |Yaml |none |acc |0.6231|± |0.0246|
| - high_school_microeconomics |Yaml |none |acc |0.6471|± |0.0310|
| - high_school_psychology |Yaml |none |acc |0.8000|± |0.0171|
| - human_sexuality |Yaml |none |acc |0.7557|± |0.0377|
| - professional_psychology |Yaml |none |acc |0.6552|± |0.0192|
| - public_relations |Yaml |none |acc |0.6636|± |0.0453|
| - security_studies |Yaml |none |acc |0.7184|± |0.0288|
| - sociology |Yaml |none |acc |0.8358|± |0.0262|
| - us_foreign_policy |Yaml |none |acc |0.8500|± |0.0359|
| - stem |N/A |none |acc |0.5217|± |0.1149|
| - abstract_algebra |Yaml |none |acc |0.3000|± |0.0461|
| - anatomy |Yaml |none |acc |0.6222|± |0.0419|
| - astronomy |Yaml |none |acc |0.6711|± |0.0382|
| - college_biology |Yaml |none |acc |0.7361|± |0.0369|
| - college_chemistry |Yaml |none |acc |0.4400|± |0.0499|
| - college_computer_science |Yaml |none |acc |0.5000|± |0.0503|
| - college_mathematics |Yaml |none |acc |0.3100|± |0.0465|
| - college_physics |Yaml |none |acc |0.4902|± |0.0497|
| - computer_security |Yaml |none |acc |0.7100|± |0.0456|
| - conceptual_physics |Yaml |none |acc |0.5362|± |0.0326|
| - electrical_engineering |Yaml |none |acc |0.5862|± |0.0410|
| - elementary_mathematics |Yaml |none |acc |0.4365|± |0.0255|
| - high_school_biology |Yaml |none |acc |0.7129|± |0.0257|
| - high_school_chemistry |Yaml |none |acc |0.5074|± |0.0352|
| - high_school_computer_science |Yaml |none |acc |0.6500|± |0.0479|
| - high_school_mathematics |Yaml |none |acc |0.3259|± |0.0286|
| - high_school_physics |Yaml |none |acc |0.3709|± |0.0394|
| - high_school_statistics |Yaml |none |acc |0.5139|± |0.0341|
| - machine_learning |Yaml |none |acc |0.5089|± |0.0475|
| Groups |Version|Filter|Metric|Value | |Stderr|
|------------------|-------|------|------|-----:|---|-----:|
|mmlu |N/A |none |acc |0.6085|± |0.1321|
| - humanities |N/A |none |acc |0.5405|± |0.1478|
| - other |N/A |none |acc |0.6894|± |0.1091|
| - social_sciences|N/A |none |acc |0.7195|± |0.0676|
| - stem |N/A |none |acc |0.5217|± |0.1149|
<!-- original-model-card end -->
| [
"SCIQ"
] |
RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2404.14619",
"endpoints_compatible",
"region:us"
] | "2024-07-13T20:08:49Z" | 2024-07-13T22:26:15+00:00 | 438 | 1 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
OpenELM-1_1B-Instruct - GGUF
- Model creator: https://huggingface.co/apple/
- Original model: https://huggingface.co/apple/OpenELM-1_1B-Instruct/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [OpenELM-1_1B-Instruct.Q2_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q2_K.gguf) | Q2_K | 0.39GB |
| [OpenELM-1_1B-Instruct.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.IQ3_XS.gguf) | IQ3_XS | 0.44GB |
| [OpenELM-1_1B-Instruct.IQ3_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.IQ3_S.gguf) | IQ3_S | 0.46GB |
| [OpenELM-1_1B-Instruct.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q3_K_S.gguf) | Q3_K_S | 0.46GB |
| [OpenELM-1_1B-Instruct.IQ3_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.IQ3_M.gguf) | IQ3_M | 0.49GB |
| [OpenELM-1_1B-Instruct.Q3_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q3_K.gguf) | Q3_K | 0.52GB |
| [OpenELM-1_1B-Instruct.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q3_K_M.gguf) | Q3_K_M | 0.52GB |
| [OpenELM-1_1B-Instruct.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q3_K_L.gguf) | Q3_K_L | 0.56GB |
| [OpenELM-1_1B-Instruct.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.IQ4_XS.gguf) | IQ4_XS | 0.55GB |
| [OpenELM-1_1B-Instruct.Q4_0.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q4_0.gguf) | Q4_0 | 0.58GB |
| [OpenELM-1_1B-Instruct.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.IQ4_NL.gguf) | IQ4_NL | 0.58GB |
| [OpenELM-1_1B-Instruct.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q4_K_S.gguf) | Q4_K_S | 0.58GB |
| [OpenELM-1_1B-Instruct.Q4_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q4_K.gguf) | Q4_K | 0.63GB |
| [OpenELM-1_1B-Instruct.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q4_K_M.gguf) | Q4_K_M | 0.63GB |
| [OpenELM-1_1B-Instruct.Q4_1.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q4_1.gguf) | Q4_1 | 0.64GB |
| [OpenELM-1_1B-Instruct.Q5_0.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q5_0.gguf) | Q5_0 | 0.7GB |
| [OpenELM-1_1B-Instruct.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q5_K_S.gguf) | Q5_K_S | 0.7GB |
| [OpenELM-1_1B-Instruct.Q5_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q5_K.gguf) | Q5_K | 0.73GB |
| [OpenELM-1_1B-Instruct.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q5_K_M.gguf) | Q5_K_M | 0.73GB |
| [OpenELM-1_1B-Instruct.Q5_1.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q5_1.gguf) | Q5_1 | 0.76GB |
| [OpenELM-1_1B-Instruct.Q6_K.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q6_K.gguf) | Q6_K | 0.83GB |
| [OpenELM-1_1B-Instruct.Q8_0.gguf](https://huggingface.co/RichardErkhov/apple_-_OpenELM-1_1B-Instruct-gguf/blob/main/OpenELM-1_1B-Instruct.Q8_0.gguf) | Q8_0 | 1.07GB |
Original model description:
---
license: other
license_name: apple-sample-code-license
license_link: LICENSE
---
# OpenELM
*Sachin Mehta, Mohammad Hossein Sekhavat, Qingqing Cao, Maxwell Horton, Yanzi Jin, Chenfan Sun, Iman Mirzadeh, Mahyar Najibi, Dmitry Belenko, Peter Zatloukal, Mohammad Rastegari*
We introduce **OpenELM**, a family of **Open** **E**fficient **L**anguage **M**odels. OpenELM uses a layer-wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, leading to enhanced accuracy. We pretrained OpenELM models using the [CoreNet](https://github.com/apple/corenet) library. We release both pretrained and instruction tuned models with 270M, 450M, 1.1B and 3B parameters.
Our pre-training dataset contains RefinedWeb, deduplicated PILE, a subset of RedPajama, and a subset of Dolma v1.6, totaling approximately 1.8 trillion tokens. Please check license agreements and terms of these datasets before using them.
## Usage
We have provided an example function to generate output from OpenELM models loaded via [HuggingFace Hub](https://huggingface.co/docs/hub/) in `generate_openelm.py`.
You can try the model by running the following command:
```
python generate_openelm.py --model apple/OpenELM-1_1B-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2
```
Please refer to [this link](https://huggingface.co/docs/hub/security-tokens) to obtain your hugging face access token.
Additional arguments to the hugging face generate function can be passed via `generate_kwargs`. As an example, to speedup the inference, you can try [lookup token speculative generation](https://huggingface.co/docs/transformers/generation_strategies) by passing the `prompt_lookup_num_tokens` argument as follows:
```
python generate_openelm.py --model apple/OpenELM-1_1B-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2 prompt_lookup_num_tokens=10
```
Alternatively, try model-wise speculative generation with an [assistive model](https://huggingface.co/blog/assisted-generation) by passing a smaller model through the `assistant_model` argument, for example:
```
python generate_openelm.py --model apple/OpenELM-1_1B-Instruct --hf_access_token [HF_ACCESS_TOKEN] --prompt 'Once upon a time there was' --generate_kwargs repetition_penalty=1.2 --assistant_model [SMALLER_MODEL]
```
## Main Results
### Zero-Shot
| **Model Size** | **ARC-c** | **ARC-e** | **BoolQ** | **HellaSwag** | **PIQA** | **SciQ** | **WinoGrande** | **Average** |
|-----------------------------------------------------------------------------|-----------|-----------|-----------|---------------|-----------|-----------|----------------|-------------|
| [OpenELM-270M](https://huggingface.co/apple/OpenELM-270M) | 26.45 | 45.08 | **53.98** | 46.71 | 69.75 | **84.70** | **53.91** | 54.37 |
| [OpenELM-270M-Instruct](https://huggingface.co/apple/OpenELM-270M-Instruct) | **30.55** | **46.68** | 48.56 | **52.07** | **70.78** | 84.40 | 52.72 | **55.11** |
| [OpenELM-450M](https://huggingface.co/apple/OpenELM-450M) | 27.56 | 48.06 | 55.78 | 53.97 | 72.31 | 87.20 | 58.01 | 57.56 |
| [OpenELM-450M-Instruct](https://huggingface.co/apple/OpenELM-450M-Instruct) | **30.38** | **50.00** | **60.37** | **59.34** | **72.63** | **88.00** | **58.96** | **59.95** |
| [OpenELM-1_1B](https://huggingface.co/apple/OpenELM-1_1B) | 32.34 | **55.43** | 63.58 | 64.81 | **75.57** | **90.60** | 61.72 | 63.44 |
| [OpenELM-1_1B-Instruct](https://huggingface.co/apple/OpenELM-1_1B-Instruct) | **37.97** | 52.23 | **70.00** | **71.20** | 75.03 | 89.30 | **62.75** | **65.50** |
| [OpenELM-3B](https://huggingface.co/apple/OpenELM-3B) | 35.58 | 59.89 | 67.40 | 72.44 | 78.24 | **92.70** | 65.51 | 67.39 |
| [OpenELM-3B-Instruct](https://huggingface.co/apple/OpenELM-3B-Instruct) | **39.42** | **61.74** | **68.17** | **76.36** | **79.00** | 92.50 | **66.85** | **69.15** |
### LLM360
| **Model Size** | **ARC-c** | **HellaSwag** | **MMLU** | **TruthfulQA** | **WinoGrande** | **Average** |
|-----------------------------------------------------------------------------|-----------|---------------|-----------|----------------|----------------|-------------|
| [OpenELM-270M](https://huggingface.co/apple/OpenELM-270M) | 27.65 | 47.15 | 25.72 | **39.24** | **53.83** | 38.72 |
| [OpenELM-270M-Instruct](https://huggingface.co/apple/OpenELM-270M-Instruct) | **32.51** | **51.58** | **26.70** | 38.72 | 53.20 | **40.54** |
| [OpenELM-450M](https://huggingface.co/apple/OpenELM-450M) | 30.20 | 53.86 | **26.01** | 40.18 | 57.22 | 41.50 |
| [OpenELM-450M-Instruct](https://huggingface.co/apple/OpenELM-450M-Instruct) | **33.53** | **59.31** | 25.41 | **40.48** | **58.33** | **43.41** |
| [OpenELM-1_1B](https://huggingface.co/apple/OpenELM-1_1B) | 36.69 | 65.71 | **27.05** | 36.98 | 63.22 | 45.93 |
| [OpenELM-1_1B-Instruct](https://huggingface.co/apple/OpenELM-1_1B-Instruct) | **41.55** | **71.83** | 25.65 | **45.95** | **64.72** | **49.94** |
| [OpenELM-3B](https://huggingface.co/apple/OpenELM-3B) | 42.24 | 73.28 | **26.76** | 34.98 | 67.25 | 48.90 |
| [OpenELM-3B-Instruct](https://huggingface.co/apple/OpenELM-3B-Instruct) | **47.70** | **76.87** | 24.80 | **38.76** | **67.96** | **51.22** |
### OpenLLM Leaderboard
| **Model Size** | **ARC-c** | **CrowS-Pairs** | **HellaSwag** | **MMLU** | **PIQA** | **RACE** | **TruthfulQA** | **WinoGrande** | **Average** |
|-----------------------------------------------------------------------------|-----------|-----------------|---------------|-----------|-----------|-----------|----------------|----------------|-------------|
| [OpenELM-270M](https://huggingface.co/apple/OpenELM-270M) | 27.65 | **66.79** | 47.15 | 25.72 | 69.75 | 30.91 | **39.24** | **53.83** | 45.13 |
| [OpenELM-270M-Instruct](https://huggingface.co/apple/OpenELM-270M-Instruct) | **32.51** | 66.01 | **51.58** | **26.70** | **70.78** | 33.78 | 38.72 | 53.20 | **46.66** |
| [OpenELM-450M](https://huggingface.co/apple/OpenELM-450M) | 30.20 | **68.63** | 53.86 | **26.01** | 72.31 | 33.11 | 40.18 | 57.22 | 47.69 |
| [OpenELM-450M-Instruct](https://huggingface.co/apple/OpenELM-450M-Instruct) | **33.53** | 67.44 | **59.31** | 25.41 | **72.63** | **36.84** | **40.48** | **58.33** | **49.25** |
| [OpenELM-1_1B](https://huggingface.co/apple/OpenELM-1_1B) | 36.69 | **71.74** | 65.71 | **27.05** | **75.57** | 36.46 | 36.98 | 63.22 | 51.68 |
| [OpenELM-1_1B-Instruct](https://huggingface.co/apple/OpenELM-1_1B-Instruct) | **41.55** | 71.02 | **71.83** | 25.65 | 75.03 | **39.43** | **45.95** | **64.72** | **54.40** |
| [OpenELM-3B](https://huggingface.co/apple/OpenELM-3B) | 42.24 | **73.29** | 73.28 | **26.76** | 78.24 | **38.76** | 34.98 | 67.25 | 54.35 |
| [OpenELM-3B-Instruct](https://huggingface.co/apple/OpenELM-3B-Instruct) | **47.70** | 72.33 | **76.87** | 24.80 | **79.00** | 38.47 | **38.76** | **67.96** | **55.73** |
See the technical report for more results and comparison.
## Evaluation
### Setup
Install the following dependencies:
```bash
# install public lm-eval-harness
harness_repo="public-lm-eval-harness"
git clone https://github.com/EleutherAI/lm-evaluation-harness ${harness_repo}
cd ${harness_repo}
# use main branch on 03-15-2024, SHA is dc90fec
git checkout dc90fec
pip install -e .
cd ..
# 66d6242 is the main branch on 2024-04-01
pip install datasets@git+https://github.com/huggingface/datasets.git@66d6242
pip install tokenizers>=0.15.2 transformers>=4.38.2 sentencepiece>=0.2.0
```
### Evaluate OpenELM
```bash
# OpenELM-1_1B-Instruct
hf_model=apple/OpenELM-1_1B-Instruct
# this flag is needed because lm-eval-harness set add_bos_token to False by default, but OpenELM uses LLaMA tokenizer which requires add_bos_token to be True
tokenizer=meta-llama/Llama-2-7b-hf
add_bos_token=True
batch_size=1
mkdir lm_eval_output
shot=0
task=arc_challenge,arc_easy,boolq,hellaswag,piqa,race,winogrande,sciq,truthfulqa_mc2
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
shot=5
task=mmlu,winogrande
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
shot=25
task=arc_challenge,crows_pairs_english
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
shot=10
task=hellaswag
lm_eval --model hf \
--model_args pretrained=${hf_model},trust_remote_code=True,add_bos_token=${add_bos_token},tokenizer=${tokenizer} \
--tasks ${task} \
--device cuda:0 \
--num_fewshot ${shot} \
--output_path ./lm_eval_output/${hf_model//\//_}_${task//,/_}-${shot}shot \
--batch_size ${batch_size} 2>&1 | tee ./lm_eval_output/eval-${hf_model//\//_}_${task//,/_}-${shot}shot.log
```
## Bias, Risks, and Limitations
The release of OpenELM models aims to empower and enrich the open research community by providing access to state-of-the-art language models. Trained on publicly available datasets, these models are made available without any safety guarantees. Consequently, there exists the possibility of these models producing outputs that are inaccurate, harmful, biased, or objectionable in response to user prompts. Thus, it is imperative for users and developers to undertake thorough safety testing and implement appropriate filtering mechanisms tailored to their specific requirements.
## Citation
If you find our work useful, please cite:
```BibTex
@article{mehtaOpenELMEfficientLanguage2024,
title = {{OpenELM}: {An} {Efficient} {Language} {Model} {Family} with {Open} {Training} and {Inference} {Framework}},
shorttitle = {{OpenELM}},
url = {https://arxiv.org/abs/2404.14619v1},
language = {en},
urldate = {2024-04-24},
journal = {arXiv.org},
author = {Mehta, Sachin and Sekhavat, Mohammad Hossein and Cao, Qingqing and Horton, Maxwell and Jin, Yanzi and Sun, Chenfan and Mirzadeh, Iman and Najibi, Mahyar and Belenko, Dmitry and Zatloukal, Peter and Rastegari, Mohammad},
month = apr,
year = {2024},
}
@inproceedings{mehta2022cvnets,
author = {Mehta, Sachin and Abdolhosseini, Farzad and Rastegari, Mohammad},
title = {CVNets: High Performance Library for Computer Vision},
year = {2022},
booktitle = {Proceedings of the 30th ACM International Conference on Multimedia},
series = {MM '22}
}
```
| [
"SCIQ"
] |
QuantFactory/Meditron3-Gemma2-2B-GGUF | QuantFactory | text-generation | [
"gguf",
"medical",
"text-generation",
"en",
"base_model:google/gemma-2-2b-it",
"base_model:quantized:google/gemma-2-2b-it",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2025-01-26T11:16:29Z" | 2025-01-26T11:29:14+00:00 | 437 | 2 | ---
base_model:
- google/gemma-2-2b-it
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- medical
---
[](https://hf.co/QuantFactory)
# QuantFactory/Meditron3-Gemma2-2B-GGUF
This is quantized version of [OpenMeditron/Meditron3-Gemma2-2B](https://huggingface.co/OpenMeditron/Meditron3-Gemma2-2B) created using llama.cpp
# Original Model Card
### Model Card: Gemma2 Meditron-3[2B]
**Model Type:** Large Language Model (LLM)
**Specialization:** Medicine
**Focus:** General purpose including limited resource and humanitarian settings
**Description:**
Meditron is a suite of large language models specialized in clinical medicine. The models are co-designed with a diverse range of expert clinicians and humanitarian practitioners. Its training emphasizes equitable representation, contextual diversity, and actionable real-world evidence-based guidelines. We make a particular effort to represent limited-resource and humanitarian settings, neglected populations, and diseases. This release is trained on Gemma2[2B] base model and has the nomenclature Gemma2 Meditron-3[2B].
#### Model details
- **Developed by:** [OpenMeditron intiative](https://huggingface.co/OpenMeditron)
- **Model type:** Causal decoder-only transformer language model
- **Language(s):** English (mainly)
- **Finetuned from model:** [Gemma2-2B](https://huggingface.co/google/gemma-2-2b)
- **Input:** Text only
- **Output:** Text only
- **Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we enhance model's performance.
#### Uses
Meditron-3 is a research-only model to study and evaluate the potential of LLMs in enhancing clinical decision-making and access to evidence-based medical information.
#### Direct Use
Meditron-3 is a research-only model. It is not validated for medical use (see disclaimer below).
#### Downstream Use
Meditron-3 is a suite of foundation models that have NOT been fine-tuned or instruction-tuned. However, these models can be adapted to specific downstream tasks or applications using techniques such as Reinforcement Learning from Human Feedback (RLHF) or Direct Preference Optimization (DPO). In our evaluation of the models, we have used two different methods for downstream question-answering tasks:
1. In-context learning with k demonstrations added to the prompt.
2. Model fine-tuning for Q&A tasks using specific training datasets.
#### Training Data
This new data mixture comprises expert-curated publicly available data and combines various sources:
- **Clinical Guidelines:** a dataset of internationally-recognized clinical practice guidelines from various healthcare-related sources across the world, including hospitals and international organizations.
- **Peer-Reviewed Medical Publications:** full-text medical articles.
- **Synthetic Differential Diagnoses:** synthetic conversation like data for differential diagnosis.
- **Replay Data:** general domain pretraining data sampled from multiple state of the art pretraining and instruction tuning.
- **LLM-enhanced Medical MCQ:** medical multiple-choice questions enriched with LLMs.
Additional information about the datasets will be included in the Meditron-3 publication.
#### Evaluation
| Model Name | MedmcQA | MedQA | PubmedQA | Average |
|-----------------------------|---------|--------|----------|---------|
| google/gemma-2-2b | 40.31 | 34.80 | 74.20 | 49.77 |
| gemMeditron-2-2b-it | 42.51 | 38.81 | 75.40 | 52.24 |
| Difference (gemMeditron vs.)| 2.20 | 4.01 | 1.20 | 2.47 |
We evaluated Meditron on medical multiple-choice questions using [lm-harness](https://github.com/EleutherAI/lm-evaluation-harness) for reproducibility.
While MCQs are valuable for assessing exam-like performance, they fall short of capturing the model's real-world utility, especially in terms of contextual adaptation in under-represented settings. Medicine is not multiple choice and we need to go beyond accuracy to assess finer-grained issues like empathy, alignment to local guidelines, structure, completeness and safety. To address this, we have developed a platform to collect feedback directly from experts to continuously adapt to the changing contexts of clinical practice.
#### Paper
The Meditron-3 publication is currently in progress and will be released at a later date.
#### Legal Disclaimer
THIS SOFTWARE AND MODEL ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS, CONTRIBUTORS, OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
These models are a research tool intended for use in the field of computational linguistics and medicine. They are not intended to be used as diagnostic tools or for clinical decision-making without appropriate validation and regulatory approval. The content and data provided with the models do not replace the expertise of healthcare professionals. Healthcare professionals should use their professional judgment in evaluating the outputs of the Gemma models. Patients should not use the model outputs for self-diagnosis or treatment without consulting a qualified healthcare provider.
THE INFORMATION IS NOT INTENDED FOR CLINICAL DECISION-MAKING, IS NOT INTENDED TO BE USED IN THE DIAGNOSIS OR TREATMENT OF PATIENTS, AND MAY NOT BE USEFUL OR APPROPRIATE FOR ANY CLINICAL PURPOSE.
UNDER NO CIRCUMSTANCES CAN USERS USE THE NAME “YALE” OR "EPFL" OR “YALE UNIVERSITY,” OR ANY AFFILIATED INSTITUTION NOR ANY VARIATION OR ADAPTATION THEREOF, NOR ANY TRADEMARK, TRADENAME OR OTHER DESIGNATION OWNED BY YALE, NOR THE NAMES OF ANY OF ITS TRUSTEES, OFFICERS, FACULTY, STUDENTS, EMPLOYEES OR AGENTS, FOR ANY PURPOSE WITHOUT THE PRIOR WRITTEN CONSENT OF YALE IN EACH INSTANCE, SUCH CONSENT TO BE GRANTED OR WITHHELD BY YALE IN ITS SOLE DISCRETION.
| [
"MEDQA",
"PUBMEDQA"
] |
mradermacher/1.5-Pints-2K-v0.1-GGUF | mradermacher | null | [
"transformers",
"gguf",
"en",
"dataset:pints-ai/Expository-Prose-V1",
"dataset:HuggingFaceH4/ultrachat_200k",
"dataset:Open-Orca/SlimOrca-Dedup",
"dataset:meta-math/MetaMathQA",
"dataset:HuggingFaceH4/deita-10k-v0-sft",
"dataset:WizardLM/WizardLM_evol_instruct_V2_196k",
"dataset:togethercomputer/llama-instruct",
"dataset:LDJnr/Capybara",
"dataset:HuggingFaceH4/ultrafeedback_binarized",
"base_model:pints-ai/1.5-Pints-2K-v0.1",
"base_model:quantized:pints-ai/1.5-Pints-2K-v0.1",
"license:mit",
"endpoints_compatible",
"region:us",
"conversational"
] | "2025-02-24T16:02:16Z" | 2025-02-24T16:22:59+00:00 | 437 | 0 | ---
base_model: pints-ai/1.5-Pints-2K-v0.1
datasets:
- pints-ai/Expository-Prose-V1
- HuggingFaceH4/ultrachat_200k
- Open-Orca/SlimOrca-Dedup
- meta-math/MetaMathQA
- HuggingFaceH4/deita-10k-v0-sft
- WizardLM/WizardLM_evol_instruct_V2_196k
- togethercomputer/llama-instruct
- LDJnr/Capybara
- HuggingFaceH4/ultrafeedback_binarized
language:
- en
library_name: transformers
license: mit
extra_gated_fields:
Company: text
Country: country
I agree to use this model for in accordance to the afore-mentioned Terms of Use: checkbox
I want to use this model for:
options:
- Research
- Education
- label: Other
value: other
type: select
Specific date: date_picker
extra_gated_prompt: Though best efforts has been made to ensure, as much as possible,
that all texts in the training corpora are royalty free, this does not constitute
a legal guarantee that such is the case. **By using any of the models, corpora or
part thereof, the user agrees to bear full responsibility to do the necessary due
diligence to ensure that he / she is in compliance with their local copyright laws.
Additionally, the user agrees to bear any damages arising as a direct cause (or
otherwise) of using any artifacts released by the pints research team, as well as
full responsibility for the consequences of his / her usage (or implementation)
of any such released artifacts. The user also indemnifies Pints Research Team (and
any of its members or agents) of any damage, related or unrelated, to the release
or subsequent usage of any findings, artifacts or code by the team. For the avoidance
of doubt, any artifacts released by the Pints Research team are done so in accordance
with the 'fair use' clause of Copyright Law, in hopes that this will aid the research
community in bringing LLMs to the next frontier.
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/pints-ai/1.5-Pints-2K-v0.1
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q2_K.gguf) | Q2_K | 0.7 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q3_K_S.gguf) | Q3_K_S | 0.8 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q3_K_M.gguf) | Q3_K_M | 0.9 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q3_K_L.gguf) | Q3_K_L | 0.9 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.IQ4_XS.gguf) | IQ4_XS | 1.0 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q4_K_S.gguf) | Q4_K_S | 1.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q4_K_M.gguf) | Q4_K_M | 1.1 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q5_K_S.gguf) | Q5_K_S | 1.2 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q5_K_M.gguf) | Q5_K_M | 1.2 | |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q6_K.gguf) | Q6_K | 1.4 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.Q8_0.gguf) | Q8_0 | 1.8 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/1.5-Pints-2K-v0.1-GGUF/resolve/main/1.5-Pints-2K-v0.1.f16.gguf) | f16 | 3.2 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
| [
"BEAR"
] |
qinxianliu/FUE-v1 | qinxianliu | feature-extraction | [
"sentence-transformers",
"safetensors",
"mpnet",
"feature-extraction",
"sentence-similarity",
"mteb",
"en",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | "2024-09-09T06:03:02Z" | 2024-10-07T12:52:58+00:00 | 435 | 0 | ---
language: en
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- mteb
model-index:
- name: FUE-v1
results:
- task:
type: Retrieval
dataset:
name: MTEB ArguAna (default)
type: mteb/arguana
config: default
split: test
revision: c22ab2a51041ffd869aaddef7af8d8215647e41a
metrics:
- type: main_score
value: 28.538000000000004
- type: map_at_1
value: 12.518
- type: map_at_10
value: 22.123
- type: map_at_100
value: 23.69
- type: map_at_1000
value: 23.753
- type: map_at_20
value: 23.143
- type: map_at_3
value: 18.255
- type: map_at_5
value: 20.101
- type: mrr_at_1
value: 12.51778093883357
- type: mrr_at_10
value: 22.1181896181896
- type: mrr_at_100
value: 23.691403907137428
- type: mrr_at_1000
value: 23.754663976195605
- type: mrr_at_20
value: 23.14463626403498
- type: mrr_at_3
value: 18.25509720246565
- type: mrr_at_5
value: 20.090090090090076
- type: nauc_map_at_1000_diff1
value: 9.323271975614725
- type: nauc_map_at_1000_max
value: 14.254988429417367
- type: nauc_map_at_1000_std
value: 0.6719763425462996
- type: nauc_map_at_100_diff1
value: 9.282199840240164
- type: nauc_map_at_100_max
value: 14.292222368573587
- type: nauc_map_at_100_std
value: 0.7292665505578078
- type: nauc_map_at_10_diff1
value: 9.281903499487566
- type: nauc_map_at_10_max
value: 13.866426442021092
- type: nauc_map_at_10_std
value: 0.25246485902079857
- type: nauc_map_at_1_diff1
value: 17.7280206384371
- type: nauc_map_at_1_max
value: 10.815824715688484
- type: nauc_map_at_1_std
value: -2.743162847857448
- type: nauc_map_at_20_diff1
value: 9.102539119027215
- type: nauc_map_at_20_max
value: 14.315213373754535
- type: nauc_map_at_20_std
value: 0.4814890472114564
- type: nauc_map_at_3_diff1
value: 11.182980254921844
- type: nauc_map_at_3_max
value: 12.459436078396347
- type: nauc_map_at_3_std
value: -0.556503984217633
- type: nauc_map_at_5_diff1
value: 10.337883923056356
- type: nauc_map_at_5_max
value: 13.156434240892349
- type: nauc_map_at_5_std
value: -0.29354150586109307
- type: nauc_mrr_at_1000_diff1
value: 9.324556497025348
- type: nauc_mrr_at_1000_max
value: 14.253533211305847
- type: nauc_mrr_at_1000_std
value: 0.6697271579920419
- type: nauc_mrr_at_100_diff1
value: 9.283500049480148
- type: nauc_mrr_at_100_max
value: 14.29077459198726
- type: nauc_mrr_at_100_std
value: 0.7270254698558568
- type: nauc_mrr_at_10_diff1
value: 9.307586753777215
- type: nauc_mrr_at_10_max
value: 13.867744805840864
- type: nauc_mrr_at_10_std
value: 0.24609376657604679
- type: nauc_mrr_at_1_diff1
value: 17.7280206384371
- type: nauc_mrr_at_1_max
value: 10.815824715688484
- type: nauc_mrr_at_1_std
value: -2.743162847857448
- type: nauc_mrr_at_20_diff1
value: 9.104040005863022
- type: nauc_mrr_at_20_max
value: 14.313919541370158
- type: nauc_mrr_at_20_std
value: 0.47929028753819247
- type: nauc_mrr_at_3_diff1
value: 11.182980254921844
- type: nauc_mrr_at_3_max
value: 12.459436078396347
- type: nauc_mrr_at_3_std
value: -0.556503984217633
- type: nauc_mrr_at_5_diff1
value: 10.292026698958562
- type: nauc_mrr_at_5_max
value: 13.162311512830788
- type: nauc_mrr_at_5_std
value: -0.279085086218627
- type: nauc_ndcg_at_1000_diff1
value: 7.3163309764159825
- type: nauc_ndcg_at_1000_max
value: 16.010286453339404
- type: nauc_ndcg_at_1000_std
value: 2.607890495864114
- type: nauc_ndcg_at_100_diff1
value: 6.345607449579556
- type: nauc_ndcg_at_100_max
value: 17.107603089582387
- type: nauc_ndcg_at_100_std
value: 4.24098978361439
- type: nauc_ndcg_at_10_diff1
value: 5.8109195734245125
- type: nauc_ndcg_at_10_max
value: 15.749698291184078
- type: nauc_ndcg_at_10_std
value: 1.7791956012747472
- type: nauc_ndcg_at_1_diff1
value: 17.7280206384371
- type: nauc_ndcg_at_1_max
value: 10.815824715688484
- type: nauc_ndcg_at_1_std
value: -2.743162847857448
- type: nauc_ndcg_at_20_diff1
value: 4.8931445052931535
- type: nauc_ndcg_at_20_max
value: 17.242324916281724
- type: nauc_ndcg_at_20_std
value: 2.5398984271374716
- type: nauc_ndcg_at_3_diff1
value: 9.692595930124401
- type: nauc_ndcg_at_3_max
value: 13.040710081661585
- type: nauc_ndcg_at_3_std
value: 0.04190136287761992
- type: nauc_ndcg_at_5_diff1
value: 8.29716057792536
- type: nauc_ndcg_at_5_max
value: 14.202672828576501
- type: nauc_ndcg_at_5_std
value: 0.4915852638473377
- type: nauc_precision_at_1000_diff1
value: -4.340157000443621
- type: nauc_precision_at_1000_max
value: 47.2664467039377
- type: nauc_precision_at_1000_std
value: 55.01988662253597
- type: nauc_precision_at_100_diff1
value: -7.7805105013646445
- type: nauc_precision_at_100_max
value: 35.464559183683306
- type: nauc_precision_at_100_std
value: 32.179756475948615
- type: nauc_precision_at_10_diff1
value: -2.8054318748220983
- type: nauc_precision_at_10_max
value: 20.50642719670412
- type: nauc_precision_at_10_std
value: 5.658488748996167
- type: nauc_precision_at_1_diff1
value: 17.7280206384371
- type: nauc_precision_at_1_max
value: 10.815824715688484
- type: nauc_precision_at_1_std
value: -2.743162847857448
- type: nauc_precision_at_20_diff1
value: -8.125102884571286
- type: nauc_precision_at_20_max
value: 26.61981123742234
- type: nauc_precision_at_20_std
value: 9.015277052264246
- type: nauc_precision_at_3_diff1
value: 6.293777341889125
- type: nauc_precision_at_3_max
value: 14.423054474164651
- type: nauc_precision_at_3_std
value: 1.4341093423522946
- type: nauc_precision_at_5_diff1
value: 3.7181074720510505
- type: nauc_precision_at_5_max
value: 16.654168420354303
- type: nauc_precision_at_5_std
value: 2.2783035538057934
- type: nauc_recall_at_1000_diff1
value: -4.340157000443143
- type: nauc_recall_at_1000_max
value: 47.26644670393844
- type: nauc_recall_at_1000_std
value: 55.01988662253498
- type: nauc_recall_at_100_diff1
value: -7.780510501364643
- type: nauc_recall_at_100_max
value: 35.46455918368321
- type: nauc_recall_at_100_std
value: 32.179756475948565
- type: nauc_recall_at_10_diff1
value: -2.8054318748220988
- type: nauc_recall_at_10_max
value: 20.50642719670411
- type: nauc_recall_at_10_std
value: 5.658488748996162
- type: nauc_recall_at_1_diff1
value: 17.7280206384371
- type: nauc_recall_at_1_max
value: 10.815824715688484
- type: nauc_recall_at_1_std
value: -2.743162847857448
- type: nauc_recall_at_20_diff1
value: -8.125102884571232
- type: nauc_recall_at_20_max
value: 26.619811237422397
- type: nauc_recall_at_20_std
value: 9.015277052264283
- type: nauc_recall_at_3_diff1
value: 6.293777341889125
- type: nauc_recall_at_3_max
value: 14.423054474164635
- type: nauc_recall_at_3_std
value: 1.4341093423523195
- type: nauc_recall_at_5_diff1
value: 3.7181074720510505
- type: nauc_recall_at_5_max
value: 16.65416842035427
- type: nauc_recall_at_5_std
value: 2.278303553805766
- type: ndcg_at_1
value: 12.518
- type: ndcg_at_10
value: 28.538000000000004
- type: ndcg_at_100
value: 36.315
- type: ndcg_at_1000
value: 37.905
- type: ndcg_at_20
value: 32.235
- type: ndcg_at_3
value: 20.277
- type: ndcg_at_5
value: 23.625
- type: precision_at_1
value: 12.518
- type: precision_at_10
value: 4.957000000000001
- type: precision_at_100
value: 0.864
- type: precision_at_1000
value: 0.099
- type: precision_at_20
value: 3.2079999999999997
- type: precision_at_3
value: 8.725
- type: precision_at_5
value: 6.8709999999999996
- type: recall_at_1
value: 12.518
- type: recall_at_10
value: 49.573
- type: recall_at_100
value: 86.41499999999999
- type: recall_at_1000
value: 98.72
- type: recall_at_20
value: 64.154
- type: recall_at_3
value: 26.173999999999996
- type: recall_at_5
value: 34.353
- task:
type: Reranking
dataset:
name: MTEB AskUbuntuDupQuestions (default)
type: mteb/askubuntudupquestions-reranking
config: default
split: test
revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
metrics:
- type: main_score
value: 66.82497796556063
- type: map
value: 66.82497796556063
- type: mrr
value: 79.41322604757507
- type: nAUC_map_diff1
value: 12.416876133855089
- type: nAUC_map_max
value: 26.748567859708082
- type: nAUC_map_std
value: 17.369392917676496
- type: nAUC_mrr_diff1
value: 24.35046473918137
- type: nAUC_mrr_max
value: 41.748545887921786
- type: nAUC_mrr_std
value: 20.095859022985742
- task:
type: STS
dataset:
name: MTEB BIOSSES (default)
type: mteb/biosses-sts
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cosine_pearson
value: 77.26959202649614
- type: cosine_spearman
value: 72.60346154803956
- type: euclidean_pearson
value: 75.60706813872336
- type: euclidean_spearman
value: 72.60346154803956
- type: main_score
value: 72.60346154803956
- type: manhattan_pearson
value: 74.85441649457519
- type: manhattan_spearman
value: 71.82211844340206
- type: pearson
value: 77.26960125718689
- type: spearman
value: 72.60346154803956
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackAndroidRetrieval (default)
type: mteb/cqadupstack-android
config: default
split: test
revision: f46a197baaae43b4f621051089b82a364682dfeb
metrics:
- type: main_score
value: 52.916
- type: map_at_1
value: 32.649
- type: map_at_10
value: 45.772
- type: map_at_100
value: 47.457
- type: map_at_1000
value: 47.599999999999994
- type: map_at_20
value: 46.725
- type: map_at_3
value: 41.692
- type: map_at_5
value: 44.046
- type: mrr_at_1
value: 40.629470672389125
- type: mrr_at_10
value: 51.90226400526829
- type: mrr_at_100
value: 52.55196625361068
- type: mrr_at_1000
value: 52.59335586794691
- type: mrr_at_20
value: 52.2668351149164
- type: mrr_at_3
value: 49.40391034811636
- type: mrr_at_5
value: 51.09203624225081
- type: nauc_map_at_1000_diff1
value: 45.289329882047205
- type: nauc_map_at_1000_max
value: 39.28004476774031
- type: nauc_map_at_1000_std
value: -10.47186686937888
- type: nauc_map_at_100_diff1
value: 45.27568029871466
- type: nauc_map_at_100_max
value: 39.30223620117373
- type: nauc_map_at_100_std
value: -10.410329154110126
- type: nauc_map_at_10_diff1
value: 45.788018257879656
- type: nauc_map_at_10_max
value: 38.87347267479215
- type: nauc_map_at_10_std
value: -11.99683042659094
- type: nauc_map_at_1_diff1
value: 53.35615479144534
- type: nauc_map_at_1_max
value: 31.475294365337536
- type: nauc_map_at_1_std
value: -14.003054216071481
- type: nauc_map_at_20_diff1
value: 45.490662130375064
- type: nauc_map_at_20_max
value: 39.090398903912536
- type: nauc_map_at_20_std
value: -10.96175771068962
- type: nauc_map_at_3_diff1
value: 46.8301634760078
- type: nauc_map_at_3_max
value: 36.2671266498893
- type: nauc_map_at_3_std
value: -13.38271575807636
- type: nauc_map_at_5_diff1
value: 46.28657566818991
- type: nauc_map_at_5_max
value: 37.8207988324964
- type: nauc_map_at_5_std
value: -12.314684453880659
- type: nauc_mrr_at_1000_diff1
value: 43.14531221378786
- type: nauc_mrr_at_1000_max
value: 41.03230454152977
- type: nauc_mrr_at_1000_std
value: -7.879125647271273
- type: nauc_mrr_at_100_diff1
value: 43.12790875666598
- type: nauc_mrr_at_100_max
value: 41.03454971653736
- type: nauc_mrr_at_100_std
value: -7.858776508352652
- type: nauc_mrr_at_10_diff1
value: 42.87454261242089
- type: nauc_mrr_at_10_max
value: 41.108557872693055
- type: nauc_mrr_at_10_std
value: -8.099855590270796
- type: nauc_mrr_at_1_diff1
value: 49.61818728047409
- type: nauc_mrr_at_1_max
value: 38.804257214142154
- type: nauc_mrr_at_1_std
value: -10.72284382304455
- type: nauc_mrr_at_20_diff1
value: 43.0725399972107
- type: nauc_mrr_at_20_max
value: 41.08996625932272
- type: nauc_mrr_at_20_std
value: -7.8159114035841695
- type: nauc_mrr_at_3_diff1
value: 43.267966736078975
- type: nauc_mrr_at_3_max
value: 40.36006635996485
- type: nauc_mrr_at_3_std
value: -8.754877467052037
- type: nauc_mrr_at_5_diff1
value: 42.75118896375678
- type: nauc_mrr_at_5_max
value: 40.91174373590108
- type: nauc_mrr_at_5_std
value: -8.082572960635977
- type: nauc_ndcg_at_1000_diff1
value: 42.8206024836842
- type: nauc_ndcg_at_1000_max
value: 41.23107259743807
- type: nauc_ndcg_at_1000_std
value: -7.267656950359476
- type: nauc_ndcg_at_100_diff1
value: 42.28641440933444
- type: nauc_ndcg_at_100_max
value: 41.6734450569554
- type: nauc_ndcg_at_100_std
value: -6.224022095206258
- type: nauc_ndcg_at_10_diff1
value: 42.753045687362324
- type: nauc_ndcg_at_10_max
value: 41.47728394469051
- type: nauc_ndcg_at_10_std
value: -9.82176692905538
- type: nauc_ndcg_at_1_diff1
value: 49.61818728047409
- type: nauc_ndcg_at_1_max
value: 38.804257214142154
- type: nauc_ndcg_at_1_std
value: -10.72284382304455
- type: nauc_ndcg_at_20_diff1
value: 42.79059001163042
- type: nauc_ndcg_at_20_max
value: 41.45466723327685
- type: nauc_ndcg_at_20_std
value: -7.8811099324857095
- type: nauc_ndcg_at_3_diff1
value: 42.777535675427956
- type: nauc_ndcg_at_3_max
value: 39.005245346467646
- type: nauc_ndcg_at_3_std
value: -9.754879407450163
- type: nauc_ndcg_at_5_diff1
value: 42.66583257245142
- type: nauc_ndcg_at_5_max
value: 40.326265568150504
- type: nauc_ndcg_at_5_std
value: -9.230270533786904
- type: nauc_precision_at_1000_diff1
value: -16.70348014597805
- type: nauc_precision_at_1000_max
value: -8.386803902715592
- type: nauc_precision_at_1000_std
value: 0.13502316171005296
- type: nauc_precision_at_100_diff1
value: -15.998085065118264
- type: nauc_precision_at_100_max
value: 8.956110379153944
- type: nauc_precision_at_100_std
value: 16.51962286328538
- type: nauc_precision_at_10_diff1
value: -1.001533004655409
- type: nauc_precision_at_10_max
value: 29.82358647130004
- type: nauc_precision_at_10_std
value: 10.818324954671196
- type: nauc_precision_at_1_diff1
value: 49.61818728047409
- type: nauc_precision_at_1_max
value: 38.804257214142154
- type: nauc_precision_at_1_std
value: -10.72284382304455
- type: nauc_precision_at_20_diff1
value: -7.810474487909365
- type: nauc_precision_at_20_max
value: 21.77756355634436
- type: nauc_precision_at_20_std
value: 16.63426939452981
- type: nauc_precision_at_3_diff1
value: 18.512689708793893
- type: nauc_precision_at_3_max
value: 39.095246356190245
- type: nauc_precision_at_3_std
value: -1.0810033734203999
- type: nauc_precision_at_5_diff1
value: 8.138752687073158
- type: nauc_precision_at_5_max
value: 36.10702475124429
- type: nauc_precision_at_5_std
value: 7.333980251486291
- type: nauc_recall_at_1000_diff1
value: 24.267062537529018
- type: nauc_recall_at_1000_max
value: 54.99733856577079
- type: nauc_recall_at_1000_std
value: 35.868095521705776
- type: nauc_recall_at_100_diff1
value: 26.74575223566034
- type: nauc_recall_at_100_max
value: 47.652482792272785
- type: nauc_recall_at_100_std
value: 19.054850321156742
- type: nauc_recall_at_10_diff1
value: 34.19555084006223
- type: nauc_recall_at_10_max
value: 41.8550922310514
- type: nauc_recall_at_10_std
value: -8.506390007838977
- type: nauc_recall_at_1_diff1
value: 53.35615479144534
- type: nauc_recall_at_1_max
value: 31.475294365337536
- type: nauc_recall_at_1_std
value: -14.003054216071481
- type: nauc_recall_at_20_diff1
value: 33.77586137392995
- type: nauc_recall_at_20_max
value: 42.954168251101486
- type: nauc_recall_at_20_std
value: 0.3955721013883589
- type: nauc_recall_at_3_diff1
value: 38.445298323492345
- type: nauc_recall_at_3_max
value: 35.55313976386901
- type: nauc_recall_at_3_std
value: -11.509187665960084
- type: nauc_recall_at_5_diff1
value: 35.789287343837884
- type: nauc_recall_at_5_max
value: 38.63482405526856
- type: nauc_recall_at_5_std
value: -8.350167399589925
- type: ndcg_at_1
value: 40.629
- type: ndcg_at_10
value: 52.916
- type: ndcg_at_100
value: 58.07600000000001
- type: ndcg_at_1000
value: 59.73500000000001
- type: ndcg_at_20
value: 54.974000000000004
- type: ndcg_at_3
value: 47.547
- type: ndcg_at_5
value: 50.295
- type: precision_at_1
value: 40.629
- type: precision_at_10
value: 10.700999999999999
- type: precision_at_100
value: 1.6820000000000002
- type: precision_at_1000
value: 0.22300000000000003
- type: precision_at_20
value: 6.345000000000001
- type: precision_at_3
value: 23.796
- type: precision_at_5
value: 17.596999999999998
- type: recall_at_1
value: 32.649
- type: recall_at_10
value: 66.116
- type: recall_at_100
value: 87.51
- type: recall_at_1000
value: 97.829
- type: recall_at_20
value: 73.379
- type: recall_at_3
value: 50.613
- type: recall_at_5
value: 58.01
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackEnglishRetrieval (default)
type: mteb/cqadupstack-english
config: default
split: test
revision: ad9991cb51e31e31e430383c75ffb2885547b5f0
metrics:
- type: main_score
value: 43.505
- type: map_at_1
value: 27.195000000000004
- type: map_at_10
value: 37.580000000000005
- type: map_at_100
value: 39.01
- type: map_at_1000
value: 39.159
- type: map_at_20
value: 38.330999999999996
- type: map_at_3
value: 34.412
- type: map_at_5
value: 36.193
- type: mrr_at_1
value: 35.28662420382165
- type: mrr_at_10
value: 43.60006571630778
- type: mrr_at_100
value: 44.353842651523586
- type: mrr_at_1000
value: 44.4053387476111
- type: mrr_at_20
value: 44.04638739149498
- type: mrr_at_3
value: 41.18895966029725
- type: mrr_at_5
value: 42.61252653927815
- type: nauc_map_at_1000_diff1
value: 42.38618266865594
- type: nauc_map_at_1000_max
value: 41.491554181716175
- type: nauc_map_at_1000_std
value: -3.5386577633054737
- type: nauc_map_at_100_diff1
value: 42.40564292548819
- type: nauc_map_at_100_max
value: 41.45373569198577
- type: nauc_map_at_100_std
value: -3.6625477908993473
- type: nauc_map_at_10_diff1
value: 42.625665623454125
- type: nauc_map_at_10_max
value: 40.51903544452516
- type: nauc_map_at_10_std
value: -5.536441154309886
- type: nauc_map_at_1_diff1
value: 47.914129207001
- type: nauc_map_at_1_max
value: 36.47093132755044
- type: nauc_map_at_1_std
value: -9.621259227944329
- type: nauc_map_at_20_diff1
value: 42.57383348544318
- type: nauc_map_at_20_max
value: 41.02679079990154
- type: nauc_map_at_20_std
value: -4.490511334672925
- type: nauc_map_at_3_diff1
value: 43.59030079409757
- type: nauc_map_at_3_max
value: 39.93229570655855
- type: nauc_map_at_3_std
value: -7.175841169162778
- type: nauc_map_at_5_diff1
value: 42.98743128411056
- type: nauc_map_at_5_max
value: 40.150925311516275
- type: nauc_map_at_5_std
value: -6.424502709519393
- type: nauc_mrr_at_1000_diff1
value: 42.41841677865114
- type: nauc_mrr_at_1000_max
value: 42.247852660053745
- type: nauc_mrr_at_1000_std
value: -1.165540535556555
- type: nauc_mrr_at_100_diff1
value: 42.420657267186726
- type: nauc_mrr_at_100_max
value: 42.251825675553704
- type: nauc_mrr_at_100_std
value: -1.1636278747245774
- type: nauc_mrr_at_10_diff1
value: 42.3138037346923
- type: nauc_mrr_at_10_max
value: 42.10074065067146
- type: nauc_mrr_at_10_std
value: -1.6076100571015888
- type: nauc_mrr_at_1_diff1
value: 46.14573077561728
- type: nauc_mrr_at_1_max
value: 42.061927948085334
- type: nauc_mrr_at_1_std
value: -3.7673030766056828
- type: nauc_mrr_at_20_diff1
value: 42.40273873695689
- type: nauc_mrr_at_20_max
value: 42.171375510351766
- type: nauc_mrr_at_20_std
value: -1.3515543593263308
- type: nauc_mrr_at_3_diff1
value: 42.65327763586051
- type: nauc_mrr_at_3_max
value: 42.60487343560702
- type: nauc_mrr_at_3_std
value: -2.017909554093815
- type: nauc_mrr_at_5_diff1
value: 42.55441855170127
- type: nauc_mrr_at_5_max
value: 42.24074898539688
- type: nauc_mrr_at_5_std
value: -1.6335691035307471
- type: nauc_ndcg_at_1000_diff1
value: 40.99947003301228
- type: nauc_ndcg_at_1000_max
value: 42.59731002851968
- type: nauc_ndcg_at_1000_std
value: 1.5617506389566693
- type: nauc_ndcg_at_100_diff1
value: 41.05947202800858
- type: nauc_ndcg_at_100_max
value: 42.655256081496375
- type: nauc_ndcg_at_100_std
value: 1.1275622124800324
- type: nauc_ndcg_at_10_diff1
value: 40.71646296399764
- type: nauc_ndcg_at_10_max
value: 41.32474748899915
- type: nauc_ndcg_at_10_std
value: -2.838548394405895
- type: nauc_ndcg_at_1_diff1
value: 46.14573077561728
- type: nauc_ndcg_at_1_max
value: 42.061927948085334
- type: nauc_ndcg_at_1_std
value: -3.7673030766056828
- type: nauc_ndcg_at_20_diff1
value: 40.94701485601509
- type: nauc_ndcg_at_20_max
value: 41.89909312421838
- type: nauc_ndcg_at_20_std
value: -1.0729170787288922
- type: nauc_ndcg_at_3_diff1
value: 41.57176168658056
- type: nauc_ndcg_at_3_max
value: 42.089267442299075
- type: nauc_ndcg_at_3_std
value: -3.6656009457600476
- type: nauc_ndcg_at_5_diff1
value: 41.312525235264545
- type: nauc_ndcg_at_5_max
value: 41.40459679814617
- type: nauc_ndcg_at_5_std
value: -3.607343043079315
- type: nauc_precision_at_1000_diff1
value: -10.389355556009154
- type: nauc_precision_at_1000_max
value: 11.213997730937681
- type: nauc_precision_at_1000_std
value: 30.484993965189755
- type: nauc_precision_at_100_diff1
value: -4.589336722169161
- type: nauc_precision_at_100_max
value: 23.61692037737193
- type: nauc_precision_at_100_std
value: 34.58390587538388
- type: nauc_precision_at_10_diff1
value: 11.420232344757583
- type: nauc_precision_at_10_max
value: 33.61211581898657
- type: nauc_precision_at_10_std
value: 17.67212437703975
- type: nauc_precision_at_1_diff1
value: 46.14573077561728
- type: nauc_precision_at_1_max
value: 42.061927948085334
- type: nauc_precision_at_1_std
value: -3.7673030766056828
- type: nauc_precision_at_20_diff1
value: 5.338962369182836
- type: nauc_precision_at_20_max
value: 31.49712758851038
- type: nauc_precision_at_20_std
value: 26.273239812959265
- type: nauc_precision_at_3_diff1
value: 25.760340841656195
- type: nauc_precision_at_3_max
value: 40.78701062437991
- type: nauc_precision_at_3_std
value: 6.786760881569201
- type: nauc_precision_at_5_diff1
value: 20.210043555954318
- type: nauc_precision_at_5_max
value: 37.031291554404085
- type: nauc_precision_at_5_std
value: 10.611181228801739
- type: nauc_recall_at_1000_diff1
value: 33.476332225623814
- type: nauc_recall_at_1000_max
value: 47.867568065614016
- type: nauc_recall_at_1000_std
value: 43.50634640789991
- type: nauc_recall_at_100_diff1
value: 35.07854220105017
- type: nauc_recall_at_100_max
value: 42.9081089829942
- type: nauc_recall_at_100_std
value: 19.93173296454809
- type: nauc_recall_at_10_diff1
value: 35.186657922090845
- type: nauc_recall_at_10_max
value: 36.89789950808192
- type: nauc_recall_at_10_std
value: -3.0377254637259083
- type: nauc_recall_at_1_diff1
value: 47.914129207001
- type: nauc_recall_at_1_max
value: 36.47093132755044
- type: nauc_recall_at_1_std
value: -9.621259227944329
- type: nauc_recall_at_20_diff1
value: 35.33528482662295
- type: nauc_recall_at_20_max
value: 38.328398726744055
- type: nauc_recall_at_20_std
value: 3.9623726501092533
- type: nauc_recall_at_3_diff1
value: 39.2886333384052
- type: nauc_recall_at_3_max
value: 38.57303028073727
- type: nauc_recall_at_3_std
value: -6.903543957557018
- type: nauc_recall_at_5_diff1
value: 37.06028417057189
- type: nauc_recall_at_5_max
value: 36.99251102238125
- type: nauc_recall_at_5_std
value: -5.207245708092033
- type: ndcg_at_1
value: 35.287
- type: ndcg_at_10
value: 43.505
- type: ndcg_at_100
value: 48.502
- type: ndcg_at_1000
value: 50.782000000000004
- type: ndcg_at_20
value: 45.37
- type: ndcg_at_3
value: 39.074
- type: ndcg_at_5
value: 41.123
- type: precision_at_1
value: 35.287
- type: precision_at_10
value: 8.643
- type: precision_at_100
value: 1.4829999999999999
- type: precision_at_1000
value: 0.202
- type: precision_at_20
value: 5.188000000000001
- type: precision_at_3
value: 19.49
- type: precision_at_5
value: 13.975000000000001
- type: recall_at_1
value: 27.195000000000004
- type: recall_at_10
value: 54.089
- type: recall_at_100
value: 75.381
- type: recall_at_1000
value: 89.83
- type: recall_at_20
value: 60.99
- type: recall_at_3
value: 40.556
- type: recall_at_5
value: 46.573
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGamingRetrieval (default)
type: mteb/cqadupstack-gaming
config: default
split: test
revision: 4885aa143210c98657558c04aaf3dc47cfb54340
metrics:
- type: main_score
value: 52.82599999999999
- type: map_at_1
value: 35.96
- type: map_at_10
value: 47.109
- type: map_at_100
value: 48.227
- type: map_at_1000
value: 48.294
- type: map_at_20
value: 47.742000000000004
- type: map_at_3
value: 43.856
- type: map_at_5
value: 45.73
- type: mrr_at_1
value: 41.2539184952978
- type: mrr_at_10
value: 50.4839279494452
- type: mrr_at_100
value: 51.18997195908982
- type: mrr_at_1000
value: 51.22588707203708
- type: mrr_at_20
value: 50.87826541247888
- type: mrr_at_3
value: 47.732497387669824
- type: mrr_at_5
value: 49.31556948798333
- type: nauc_map_at_1000_diff1
value: 50.21912063624032
- type: nauc_map_at_1000_max
value: 41.05492601368038
- type: nauc_map_at_1000_std
value: -1.8335758997830354
- type: nauc_map_at_100_diff1
value: 50.191466738388776
- type: nauc_map_at_100_max
value: 41.04781068836281
- type: nauc_map_at_100_std
value: -1.8487668121623901
- type: nauc_map_at_10_diff1
value: 50.299475818245554
- type: nauc_map_at_10_max
value: 40.57210666375838
- type: nauc_map_at_10_std
value: -2.5349867924738354
- type: nauc_map_at_1_diff1
value: 54.309305748182524
- type: nauc_map_at_1_max
value: 34.78537970357836
- type: nauc_map_at_1_std
value: -4.367654821096338
- type: nauc_map_at_20_diff1
value: 50.17956579459495
- type: nauc_map_at_20_max
value: 40.845935693363586
- type: nauc_map_at_20_std
value: -2.189911133302338
- type: nauc_map_at_3_diff1
value: 50.863917087098066
- type: nauc_map_at_3_max
value: 39.469485934751866
- type: nauc_map_at_3_std
value: -4.334326050046052
- type: nauc_map_at_5_diff1
value: 50.53806135770955
- type: nauc_map_at_5_max
value: 39.751056790635424
- type: nauc_map_at_5_std
value: -3.486238828551465
- type: nauc_mrr_at_1000_diff1
value: 50.3689275270778
- type: nauc_mrr_at_1000_max
value: 42.72789427349376
- type: nauc_mrr_at_1000_std
value: -0.6875106248393903
- type: nauc_mrr_at_100_diff1
value: 50.35361102734404
- type: nauc_mrr_at_100_max
value: 42.743613265352224
- type: nauc_mrr_at_100_std
value: -0.6686536663032981
- type: nauc_mrr_at_10_diff1
value: 50.25968474187666
- type: nauc_mrr_at_10_max
value: 42.746468116591906
- type: nauc_mrr_at_10_std
value: -0.7353469482521312
- type: nauc_mrr_at_1_diff1
value: 54.681271473002916
- type: nauc_mrr_at_1_max
value: 40.441585770844284
- type: nauc_mrr_at_1_std
value: -2.839590354418767
- type: nauc_mrr_at_20_diff1
value: 50.30458869022417
- type: nauc_mrr_at_20_max
value: 42.69148052936814
- type: nauc_mrr_at_20_std
value: -0.7260575052437486
- type: nauc_mrr_at_3_diff1
value: 50.69068675878361
- type: nauc_mrr_at_3_max
value: 42.54723124779581
- type: nauc_mrr_at_3_std
value: -1.9548419929611167
- type: nauc_mrr_at_5_diff1
value: 50.37284804647469
- type: nauc_mrr_at_5_max
value: 42.48933463080673
- type: nauc_mrr_at_5_std
value: -1.1959350211193
- type: nauc_ndcg_at_1000_diff1
value: 49.13669459478487
- type: nauc_ndcg_at_1000_max
value: 43.09193372090789
- type: nauc_ndcg_at_1000_std
value: 1.0255400585004846
- type: nauc_ndcg_at_100_diff1
value: 48.46036764721693
- type: nauc_ndcg_at_100_max
value: 43.337874144896745
- type: nauc_ndcg_at_100_std
value: 1.4268868889619024
- type: nauc_ndcg_at_10_diff1
value: 48.5501585301524
- type: nauc_ndcg_at_10_max
value: 42.38370635551507
- type: nauc_ndcg_at_10_std
value: -0.126958393912763
- type: nauc_ndcg_at_1_diff1
value: 54.681271473002916
- type: nauc_ndcg_at_1_max
value: 40.441585770844284
- type: nauc_ndcg_at_1_std
value: -2.839590354418767
- type: nauc_ndcg_at_20_diff1
value: 48.37089809404846
- type: nauc_ndcg_at_20_max
value: 42.53664952827513
- type: nauc_ndcg_at_20_std
value: 0.051941093126791994
- type: nauc_ndcg_at_3_diff1
value: 49.486133964537785
- type: nauc_ndcg_at_3_max
value: 41.262493607776804
- type: nauc_ndcg_at_3_std
value: -2.7967155168398428
- type: nauc_ndcg_at_5_diff1
value: 48.96714073924463
- type: nauc_ndcg_at_5_max
value: 41.323528047385636
- type: nauc_ndcg_at_5_std
value: -1.5158330808056293
- type: nauc_precision_at_1000_diff1
value: -7.026558402946765
- type: nauc_precision_at_1000_max
value: 19.486730125805913
- type: nauc_precision_at_1000_std
value: 27.926974867437256
- type: nauc_precision_at_100_diff1
value: -2.2036370386128104
- type: nauc_precision_at_100_max
value: 28.340445317172758
- type: nauc_precision_at_100_std
value: 28.516212546705543
- type: nauc_precision_at_10_diff1
value: 17.901652875127454
- type: nauc_precision_at_10_max
value: 39.46667014199858
- type: nauc_precision_at_10_std
value: 14.874676790136363
- type: nauc_precision_at_1_diff1
value: 54.681271473002916
- type: nauc_precision_at_1_max
value: 40.441585770844284
- type: nauc_precision_at_1_std
value: -2.839590354418767
- type: nauc_precision_at_20_diff1
value: 9.992442152246879
- type: nauc_precision_at_20_max
value: 35.87159722623395
- type: nauc_precision_at_20_std
value: 19.593433922664403
- type: nauc_precision_at_3_diff1
value: 33.43340071813058
- type: nauc_precision_at_3_max
value: 42.813495259558984
- type: nauc_precision_at_3_std
value: 2.319939520883305
- type: nauc_precision_at_5_diff1
value: 26.73151055105659
- type: nauc_precision_at_5_max
value: 40.42707721448163
- type: nauc_precision_at_5_std
value: 7.084075741117675
- type: nauc_recall_at_1000_diff1
value: 40.976362341621
- type: nauc_recall_at_1000_max
value: 65.29728663348455
- type: nauc_recall_at_1000_std
value: 49.444127154114526
- type: nauc_recall_at_100_diff1
value: 33.852534329005536
- type: nauc_recall_at_100_max
value: 53.02330599056479
- type: nauc_recall_at_100_std
value: 23.794773788370286
- type: nauc_recall_at_10_diff1
value: 40.67129797892841
- type: nauc_recall_at_10_max
value: 42.66444189741527
- type: nauc_recall_at_10_std
value: 4.429365961370951
- type: nauc_recall_at_1_diff1
value: 54.309305748182524
- type: nauc_recall_at_1_max
value: 34.78537970357836
- type: nauc_recall_at_1_std
value: -4.367654821096338
- type: nauc_recall_at_20_diff1
value: 38.71080752436736
- type: nauc_recall_at_20_max
value: 43.5624279616423
- type: nauc_recall_at_20_std
value: 6.624168124956635
- type: nauc_recall_at_3_diff1
value: 45.11133844611515
- type: nauc_recall_at_3_max
value: 39.73140743866134
- type: nauc_recall_at_3_std
value: -4.333260589935666
- type: nauc_recall_at_5_diff1
value: 43.2408330778742
- type: nauc_recall_at_5_max
value: 39.765735398976986
- type: nauc_recall_at_5_std
value: -0.5671079053603477
- type: ndcg_at_1
value: 41.254000000000005
- type: ndcg_at_10
value: 52.82599999999999
- type: ndcg_at_100
value: 57.333
- type: ndcg_at_1000
value: 58.714
- type: ndcg_at_20
value: 54.559000000000005
- type: ndcg_at_3
value: 47.064
- type: ndcg_at_5
value: 49.91
- type: precision_at_1
value: 41.254000000000005
- type: precision_at_10
value: 8.577
- type: precision_at_100
value: 1.1900000000000002
- type: precision_at_1000
value: 0.135
- type: precision_at_20
value: 4.84
- type: precision_at_3
value: 20.899
- type: precision_at_5
value: 14.571000000000002
- type: recall_at_1
value: 35.96
- type: recall_at_10
value: 66.52799999999999
- type: recall_at_100
value: 86.284
- type: recall_at_1000
value: 96.279
- type: recall_at_20
value: 72.914
- type: recall_at_3
value: 51.03
- type: recall_at_5
value: 57.959
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackGisRetrieval (default)
type: mteb/cqadupstack-gis
config: default
split: test
revision: 5003b3064772da1887988e05400cf3806fe491f2
metrics:
- type: main_score
value: 41.058
- type: map_at_1
value: 26.669999999999998
- type: map_at_10
value: 35.673
- type: map_at_100
value: 36.869
- type: map_at_1000
value: 36.954
- type: map_at_20
value: 36.355
- type: map_at_3
value: 32.696999999999996
- type: map_at_5
value: 34.316
- type: mrr_at_1
value: 28.70056497175141
- type: mrr_at_10
value: 37.57080082503811
- type: mrr_at_100
value: 38.57318940120763
- type: mrr_at_1000
value: 38.628923670653904
- type: mrr_at_20
value: 38.142966912998894
- type: mrr_at_3
value: 34.7080979284369
- type: mrr_at_5
value: 36.27306967984933
- type: nauc_map_at_1000_diff1
value: 31.95025505717376
- type: nauc_map_at_1000_max
value: 27.445344687308342
- type: nauc_map_at_1000_std
value: -3.766659179631952
- type: nauc_map_at_100_diff1
value: 31.904113858753202
- type: nauc_map_at_100_max
value: 27.43329141824181
- type: nauc_map_at_100_std
value: -3.751434380621304
- type: nauc_map_at_10_diff1
value: 32.26387206275712
- type: nauc_map_at_10_max
value: 27.27250131155855
- type: nauc_map_at_10_std
value: -4.2570700844827805
- type: nauc_map_at_1_diff1
value: 37.676610770755644
- type: nauc_map_at_1_max
value: 24.743059176207872
- type: nauc_map_at_1_std
value: -7.899143095387766
- type: nauc_map_at_20_diff1
value: 31.870529576581475
- type: nauc_map_at_20_max
value: 27.305874069658703
- type: nauc_map_at_20_std
value: -3.879867258167858
- type: nauc_map_at_3_diff1
value: 33.69937975192626
- type: nauc_map_at_3_max
value: 27.2785915528158
- type: nauc_map_at_3_std
value: -4.550393882628957
- type: nauc_map_at_5_diff1
value: 32.50268016494544
- type: nauc_map_at_5_max
value: 26.628666672017225
- type: nauc_map_at_5_std
value: -4.888374245634325
- type: nauc_mrr_at_1000_diff1
value: 30.688875736244718
- type: nauc_mrr_at_1000_max
value: 28.36168526315933
- type: nauc_mrr_at_1000_std
value: -2.4134356134739363
- type: nauc_mrr_at_100_diff1
value: 30.643548147379505
- type: nauc_mrr_at_100_max
value: 28.360927430391857
- type: nauc_mrr_at_100_std
value: -2.388432251569784
- type: nauc_mrr_at_10_diff1
value: 30.838160632926026
- type: nauc_mrr_at_10_max
value: 28.274232987739524
- type: nauc_mrr_at_10_std
value: -2.6455491371420234
- type: nauc_mrr_at_1_diff1
value: 36.333747251739936
- type: nauc_mrr_at_1_max
value: 27.09211690724867
- type: nauc_mrr_at_1_std
value: -6.872327181645408
- type: nauc_mrr_at_20_diff1
value: 30.566159689071643
- type: nauc_mrr_at_20_max
value: 28.254100153054484
- type: nauc_mrr_at_20_std
value: -2.3863086501910877
- type: nauc_mrr_at_3_diff1
value: 31.995970169795008
- type: nauc_mrr_at_3_max
value: 28.672649281172863
- type: nauc_mrr_at_3_std
value: -3.0253479479372682
- type: nauc_mrr_at_5_diff1
value: 30.925479033010074
- type: nauc_mrr_at_5_max
value: 27.894579265110913
- type: nauc_mrr_at_5_std
value: -3.1633756284644305
- type: nauc_ndcg_at_1000_diff1
value: 29.480108448835164
- type: nauc_ndcg_at_1000_max
value: 28.694910139113766
- type: nauc_ndcg_at_1000_std
value: -0.9685609700216138
- type: nauc_ndcg_at_100_diff1
value: 28.20394217817361
- type: nauc_ndcg_at_100_max
value: 28.718549400317933
- type: nauc_ndcg_at_100_std
value: -0.2052052223285665
- type: nauc_ndcg_at_10_diff1
value: 29.33527460830841
- type: nauc_ndcg_at_10_max
value: 28.100629016562795
- type: nauc_ndcg_at_10_std
value: -1.9043904359384647
- type: nauc_ndcg_at_1_diff1
value: 36.333747251739936
- type: nauc_ndcg_at_1_max
value: 27.09211690724867
- type: nauc_ndcg_at_1_std
value: -6.872327181645408
- type: nauc_ndcg_at_20_diff1
value: 28.12694047381911
- type: nauc_ndcg_at_20_max
value: 28.07256049681584
- type: nauc_ndcg_at_20_std
value: -0.7546400633868358
- type: nauc_ndcg_at_3_diff1
value: 31.87777938588317
- type: nauc_ndcg_at_3_max
value: 28.084297522561176
- type: nauc_ndcg_at_3_std
value: -3.092215463329312
- type: nauc_ndcg_at_5_diff1
value: 29.881507389621103
- type: nauc_ndcg_at_5_max
value: 26.823659437194475
- type: nauc_ndcg_at_5_std
value: -3.351691772718416
- type: nauc_precision_at_1000_diff1
value: -6.384965239026326
- type: nauc_precision_at_1000_max
value: 10.400043080009187
- type: nauc_precision_at_1000_std
value: 13.493069987475284
- type: nauc_precision_at_100_diff1
value: -3.8311477783636785
- type: nauc_precision_at_100_max
value: 21.313719573692566
- type: nauc_precision_at_100_std
value: 15.340019805905872
- type: nauc_precision_at_10_diff1
value: 14.879866186868682
- type: nauc_precision_at_10_max
value: 29.443484927548557
- type: nauc_precision_at_10_std
value: 5.190205795872693
- type: nauc_precision_at_1_diff1
value: 36.333747251739936
- type: nauc_precision_at_1_max
value: 27.09211690724867
- type: nauc_precision_at_1_std
value: -6.872327181645408
- type: nauc_precision_at_20_diff1
value: 6.776608893898066
- type: nauc_precision_at_20_max
value: 25.915514134442724
- type: nauc_precision_at_20_std
value: 10.25138083695759
- type: nauc_precision_at_3_diff1
value: 24.58655147167322
- type: nauc_precision_at_3_max
value: 32.0175630253561
- type: nauc_precision_at_3_std
value: 1.4274592250651807
- type: nauc_precision_at_5_diff1
value: 18.590483368382866
- type: nauc_precision_at_5_max
value: 28.253561736970234
- type: nauc_precision_at_5_std
value: 0.829570400364922
- type: nauc_recall_at_1000_diff1
value: 10.810130884218827
- type: nauc_recall_at_1000_max
value: 47.13075325263327
- type: nauc_recall_at_1000_std
value: 33.15494499163207
- type: nauc_recall_at_100_diff1
value: 8.125366968609814
- type: nauc_recall_at_100_max
value: 33.38380343426024
- type: nauc_recall_at_100_std
value: 18.279628794274075
- type: nauc_recall_at_10_diff1
value: 20.649934311742626
- type: nauc_recall_at_10_max
value: 28.749953838066926
- type: nauc_recall_at_10_std
value: 4.047023543340581
- type: nauc_recall_at_1_diff1
value: 37.676610770755644
- type: nauc_recall_at_1_max
value: 24.743059176207872
- type: nauc_recall_at_1_std
value: -7.899143095387766
- type: nauc_recall_at_20_diff1
value: 14.783244276844618
- type: nauc_recall_at_20_max
value: 28.373124736783172
- type: nauc_recall_at_20_std
value: 9.128634320360753
- type: nauc_recall_at_3_diff1
value: 28.176705702681026
- type: nauc_recall_at_3_max
value: 28.30143052234742
- type: nauc_recall_at_3_std
value: -0.04083763472538744
- type: nauc_recall_at_5_diff1
value: 23.192105206068955
- type: nauc_recall_at_5_max
value: 25.258497503610215
- type: nauc_recall_at_5_std
value: -0.5987707205459003
- type: ndcg_at_1
value: 28.701
- type: ndcg_at_10
value: 41.058
- type: ndcg_at_100
value: 46.632
- type: ndcg_at_1000
value: 48.662
- type: ndcg_at_20
value: 43.363
- type: ndcg_at_3
value: 35.132999999999996
- type: ndcg_at_5
value: 37.881
- type: precision_at_1
value: 28.701
- type: precision_at_10
value: 6.3950000000000005
- type: precision_at_100
value: 0.9690000000000001
- type: precision_at_1000
value: 0.11900000000000001
- type: precision_at_20
value: 3.7510000000000003
- type: precision_at_3
value: 14.84
- type: precision_at_5
value: 10.508000000000001
- type: recall_at_1
value: 26.669999999999998
- type: recall_at_10
value: 55.92
- type: recall_at_100
value: 80.867
- type: recall_at_1000
value: 95.906
- type: recall_at_20
value: 64.586
- type: recall_at_3
value: 39.92
- type: recall_at_5
value: 46.396
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackMathematicaRetrieval (default)
type: mteb/cqadupstack-mathematica
config: default
split: test
revision: 90fceea13679c63fe563ded68f3b6f06e50061de
metrics:
- type: main_score
value: 31.507
- type: map_at_1
value: 16.398
- type: map_at_10
value: 25.485000000000003
- type: map_at_100
value: 26.967999999999996
- type: map_at_1000
value: 27.084999999999997
- type: map_at_20
value: 26.35
- type: map_at_3
value: 22.332
- type: map_at_5
value: 23.958
- type: mrr_at_1
value: 20.398009950248756
- type: mrr_at_10
value: 29.76614941167181
- type: mrr_at_100
value: 30.86747432087463
- type: mrr_at_1000
value: 30.919927410511455
- type: mrr_at_20
value: 30.443181439265278
- type: mrr_at_3
value: 26.637645107794373
- type: mrr_at_5
value: 28.31674958540631
- type: nauc_map_at_1000_diff1
value: 31.6249291191536
- type: nauc_map_at_1000_max
value: 29.55962360137207
- type: nauc_map_at_1000_std
value: 5.215881981383439
- type: nauc_map_at_100_diff1
value: 31.597179077805414
- type: nauc_map_at_100_max
value: 29.53626392781722
- type: nauc_map_at_100_std
value: 5.205272245991525
- type: nauc_map_at_10_diff1
value: 31.60302262185177
- type: nauc_map_at_10_max
value: 28.72958244724668
- type: nauc_map_at_10_std
value: 4.591243705917117
- type: nauc_map_at_1_diff1
value: 37.68937044932118
- type: nauc_map_at_1_max
value: 27.785294663519327
- type: nauc_map_at_1_std
value: 2.281125893959806
- type: nauc_map_at_20_diff1
value: 31.569405397103345
- type: nauc_map_at_20_max
value: 29.5374594662604
- type: nauc_map_at_20_std
value: 5.062837955779829
- type: nauc_map_at_3_diff1
value: 32.846629596852864
- type: nauc_map_at_3_max
value: 28.935149877956366
- type: nauc_map_at_3_std
value: 4.043651949362703
- type: nauc_map_at_5_diff1
value: 32.20424309358551
- type: nauc_map_at_5_max
value: 28.84626720504408
- type: nauc_map_at_5_std
value: 4.480982141190721
- type: nauc_mrr_at_1000_diff1
value: 29.91514359849449
- type: nauc_mrr_at_1000_max
value: 29.4880837184256
- type: nauc_mrr_at_1000_std
value: 6.90169972042484
- type: nauc_mrr_at_100_diff1
value: 29.90225503162752
- type: nauc_mrr_at_100_max
value: 29.468862402041644
- type: nauc_mrr_at_100_std
value: 6.891889857381346
- type: nauc_mrr_at_10_diff1
value: 29.935063368574966
- type: nauc_mrr_at_10_max
value: 29.144965203735662
- type: nauc_mrr_at_10_std
value: 6.69773269545311
- type: nauc_mrr_at_1_diff1
value: 34.64233696145788
- type: nauc_mrr_at_1_max
value: 29.76861665629048
- type: nauc_mrr_at_1_std
value: 6.050508042950772
- type: nauc_mrr_at_20_diff1
value: 29.869194769562036
- type: nauc_mrr_at_20_max
value: 29.52599000601207
- type: nauc_mrr_at_20_std
value: 6.833276125615728
- type: nauc_mrr_at_3_diff1
value: 30.73272542293855
- type: nauc_mrr_at_3_max
value: 29.646125101813958
- type: nauc_mrr_at_3_std
value: 6.596409692221532
- type: nauc_mrr_at_5_diff1
value: 30.22116393198506
- type: nauc_mrr_at_5_max
value: 29.3005889974291
- type: nauc_mrr_at_5_std
value: 6.761086751620502
- type: nauc_ndcg_at_1000_diff1
value: 29.53517185395757
- type: nauc_ndcg_at_1000_max
value: 30.156269324153744
- type: nauc_ndcg_at_1000_std
value: 7.756076732606718
- type: nauc_ndcg_at_100_diff1
value: 28.978423923406538
- type: nauc_ndcg_at_100_max
value: 29.7055702302561
- type: nauc_ndcg_at_100_std
value: 7.725429773260483
- type: nauc_ndcg_at_10_diff1
value: 28.95114011689634
- type: nauc_ndcg_at_10_max
value: 28.050859118724443
- type: nauc_ndcg_at_10_std
value: 5.564584606153562
- type: nauc_ndcg_at_1_diff1
value: 34.64233696145788
- type: nauc_ndcg_at_1_max
value: 29.76861665629048
- type: nauc_ndcg_at_1_std
value: 6.050508042950772
- type: nauc_ndcg_at_20_diff1
value: 28.792293298047504
- type: nauc_ndcg_at_20_max
value: 30.26649029003995
- type: nauc_ndcg_at_20_std
value: 6.692147001644501
- type: nauc_ndcg_at_3_diff1
value: 31.22691508724979
- type: nauc_ndcg_at_3_max
value: 29.41685209008128
- type: nauc_ndcg_at_3_std
value: 5.287699533527526
- type: nauc_ndcg_at_5_diff1
value: 30.126889228701526
- type: nauc_ndcg_at_5_max
value: 28.811536881225603
- type: nauc_ndcg_at_5_std
value: 5.567866298638633
- type: nauc_precision_at_1000_diff1
value: 1.0315570861883616
- type: nauc_precision_at_1000_max
value: 5.444321907094073
- type: nauc_precision_at_1000_std
value: 3.0310745219226525
- type: nauc_precision_at_100_diff1
value: 7.18318986657559
- type: nauc_precision_at_100_max
value: 17.459722160298842
- type: nauc_precision_at_100_std
value: 10.082153389290994
- type: nauc_precision_at_10_diff1
value: 17.142578413214434
- type: nauc_precision_at_10_max
value: 26.846895769037225
- type: nauc_precision_at_10_std
value: 8.568196201489595
- type: nauc_precision_at_1_diff1
value: 34.64233696145788
- type: nauc_precision_at_1_max
value: 29.76861665629048
- type: nauc_precision_at_1_std
value: 6.050508042950772
- type: nauc_precision_at_20_diff1
value: 13.674761586839344
- type: nauc_precision_at_20_max
value: 29.83399743832858
- type: nauc_precision_at_20_std
value: 11.562042971033899
- type: nauc_precision_at_3_diff1
value: 25.01590537073653
- type: nauc_precision_at_3_max
value: 30.253033767323938
- type: nauc_precision_at_3_std
value: 7.3087944205161515
- type: nauc_precision_at_5_diff1
value: 20.975487011820988
- type: nauc_precision_at_5_max
value: 29.173537748534212
- type: nauc_precision_at_5_std
value: 8.945752465905947
- type: nauc_recall_at_1000_diff1
value: 18.53296507398216
- type: nauc_recall_at_1000_max
value: 34.927013467781165
- type: nauc_recall_at_1000_std
value: 34.7934249117797
- type: nauc_recall_at_100_diff1
value: 18.01406190276854
- type: nauc_recall_at_100_max
value: 25.35501374220775
- type: nauc_recall_at_100_std
value: 18.19308339219603
- type: nauc_recall_at_10_diff1
value: 21.406147428374446
- type: nauc_recall_at_10_max
value: 22.774943611615736
- type: nauc_recall_at_10_std
value: 5.707280310402414
- type: nauc_recall_at_1_diff1
value: 37.68937044932118
- type: nauc_recall_at_1_max
value: 27.785294663519327
- type: nauc_recall_at_1_std
value: 2.281125893959806
- type: nauc_recall_at_20_diff1
value: 20.227099777176857
- type: nauc_recall_at_20_max
value: 29.344314385773657
- type: nauc_recall_at_20_std
value: 9.170201756879665
- type: nauc_recall_at_3_diff1
value: 26.785287089852357
- type: nauc_recall_at_3_max
value: 26.721156438701684
- type: nauc_recall_at_3_std
value: 5.14517396691279
- type: nauc_recall_at_5_diff1
value: 24.972251162551565
- type: nauc_recall_at_5_max
value: 25.44929193960884
- type: nauc_recall_at_5_std
value: 5.572691905709665
- type: ndcg_at_1
value: 20.398
- type: ndcg_at_10
value: 31.507
- type: ndcg_at_100
value: 38.116
- type: ndcg_at_1000
value: 40.564
- type: ndcg_at_20
value: 34.268
- type: ndcg_at_3
value: 25.358000000000004
- type: ndcg_at_5
value: 28.03
- type: precision_at_1
value: 20.398
- type: precision_at_10
value: 6.157
- type: precision_at_100
value: 1.0959999999999999
- type: precision_at_1000
value: 0.14400000000000002
- type: precision_at_20
value: 3.862
- type: precision_at_3
value: 12.272
- type: precision_at_5
value: 9.179
- type: recall_at_1
value: 16.398
- type: recall_at_10
value: 45.774
- type: recall_at_100
value: 74.099
- type: recall_at_1000
value: 90.979
- type: recall_at_20
value: 55.507
- type: recall_at_3
value: 29.176999999999996
- type: recall_at_5
value: 35.682
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackPhysicsRetrieval (default)
type: mteb/cqadupstack-physics
config: default
split: test
revision: 79531abbd1fb92d06c6d6315a0cbbbf5bb247ea4
metrics:
- type: main_score
value: 45.95
- type: map_at_1
value: 28.138999999999996
- type: map_at_10
value: 39.495000000000005
- type: map_at_100
value: 40.949000000000005
- type: map_at_1000
value: 41.07
- type: map_at_20
value: 40.300999999999995
- type: map_at_3
value: 35.853
- type: map_at_5
value: 38.004
- type: mrr_at_1
value: 34.93743984600577
- type: mrr_at_10
value: 44.76801564385776
- type: mrr_at_100
value: 45.65247517257724
- type: mrr_at_1000
value: 45.68956227855384
- type: mrr_at_20
value: 45.29344639385099
- type: mrr_at_3
value: 41.819056785370506
- type: mrr_at_5
value: 43.58036573628482
- type: nauc_map_at_1000_diff1
value: 47.277436726452734
- type: nauc_map_at_1000_max
value: 35.37747274882577
- type: nauc_map_at_1000_std
value: -1.7620121730183462
- type: nauc_map_at_100_diff1
value: 47.29530211449703
- type: nauc_map_at_100_max
value: 35.3267759107826
- type: nauc_map_at_100_std
value: -1.9003611271296315
- type: nauc_map_at_10_diff1
value: 47.66123826774245
- type: nauc_map_at_10_max
value: 34.898894823839974
- type: nauc_map_at_10_std
value: -2.7367802679721382
- type: nauc_map_at_1_diff1
value: 53.39584452417071
- type: nauc_map_at_1_max
value: 30.44376469140723
- type: nauc_map_at_1_std
value: -6.88828726087523
- type: nauc_map_at_20_diff1
value: 47.3836198057246
- type: nauc_map_at_20_max
value: 35.0413537966224
- type: nauc_map_at_20_std
value: -2.545827885505845
- type: nauc_map_at_3_diff1
value: 48.16522210457787
- type: nauc_map_at_3_max
value: 34.3965506492862
- type: nauc_map_at_3_std
value: -3.3433431726479834
- type: nauc_map_at_5_diff1
value: 47.98417848216568
- type: nauc_map_at_5_max
value: 34.73390747566828
- type: nauc_map_at_5_std
value: -2.8612602838895826
- type: nauc_mrr_at_1000_diff1
value: 45.71218991600523
- type: nauc_mrr_at_1000_max
value: 39.11582764653062
- type: nauc_mrr_at_1000_std
value: 2.2783759668804344
- type: nauc_mrr_at_100_diff1
value: 45.70154738130718
- type: nauc_mrr_at_100_max
value: 39.11273087180276
- type: nauc_mrr_at_100_std
value: 2.2710830483092987
- type: nauc_mrr_at_10_diff1
value: 45.642234982637824
- type: nauc_mrr_at_10_max
value: 38.92779723339438
- type: nauc_mrr_at_10_std
value: 1.9256549539298882
- type: nauc_mrr_at_1_diff1
value: 50.71909609236056
- type: nauc_mrr_at_1_max
value: 38.27951880430563
- type: nauc_mrr_at_1_std
value: 0.6510875710711332
- type: nauc_mrr_at_20_diff1
value: 45.695383873021726
- type: nauc_mrr_at_20_max
value: 39.022209591680394
- type: nauc_mrr_at_20_std
value: 2.0413367142919605
- type: nauc_mrr_at_3_diff1
value: 45.64927722234226
- type: nauc_mrr_at_3_max
value: 39.19282954961338
- type: nauc_mrr_at_3_std
value: 2.474812709395244
- type: nauc_mrr_at_5_diff1
value: 45.80017070276982
- type: nauc_mrr_at_5_max
value: 39.34045012221159
- type: nauc_mrr_at_5_std
value: 2.3303744020843107
- type: nauc_ndcg_at_1000_diff1
value: 45.1794919744782
- type: nauc_ndcg_at_1000_max
value: 37.73921904631251
- type: nauc_ndcg_at_1000_std
value: 2.26624679124494
- type: nauc_ndcg_at_100_diff1
value: 44.73702317994642
- type: nauc_ndcg_at_100_max
value: 37.0759462132415
- type: nauc_ndcg_at_100_std
value: 1.4931392395285414
- type: nauc_ndcg_at_10_diff1
value: 45.25647711557264
- type: nauc_ndcg_at_10_max
value: 35.70645701721464
- type: nauc_ndcg_at_10_std
value: -1.4004314196958476
- type: nauc_ndcg_at_1_diff1
value: 50.71909609236056
- type: nauc_ndcg_at_1_max
value: 38.27951880430563
- type: nauc_ndcg_at_1_std
value: 0.6510875710711332
- type: nauc_ndcg_at_20_diff1
value: 44.76174297209143
- type: nauc_ndcg_at_20_max
value: 35.822466503188686
- type: nauc_ndcg_at_20_std
value: -1.0518640293785047
- type: nauc_ndcg_at_3_diff1
value: 45.068368130065146
- type: nauc_ndcg_at_3_max
value: 37.02529090108255
- type: nauc_ndcg_at_3_std
value: 0.2157989475242898
- type: nauc_ndcg_at_5_diff1
value: 45.67384784064928
- type: nauc_ndcg_at_5_max
value: 36.47208549375304
- type: nauc_ndcg_at_5_std
value: -0.47643849090429163
- type: nauc_precision_at_1000_diff1
value: -17.125501936260747
- type: nauc_precision_at_1000_max
value: 9.558040560420771
- type: nauc_precision_at_1000_std
value: 31.800567902705996
- type: nauc_precision_at_100_diff1
value: -10.12964985687511
- type: nauc_precision_at_100_max
value: 17.81597956519217
- type: nauc_precision_at_100_std
value: 28.681055685422674
- type: nauc_precision_at_10_diff1
value: 11.455616137810187
- type: nauc_precision_at_10_max
value: 32.707133153254944
- type: nauc_precision_at_10_std
value: 15.18726232222438
- type: nauc_precision_at_1_diff1
value: 50.71909609236056
- type: nauc_precision_at_1_max
value: 38.27951880430563
- type: nauc_precision_at_1_std
value: 0.6510875710711332
- type: nauc_precision_at_20_diff1
value: 3.249020258643051
- type: nauc_precision_at_20_max
value: 27.111481841291123
- type: nauc_precision_at_20_std
value: 17.15563337285341
- type: nauc_precision_at_3_diff1
value: 25.76995146388162
- type: nauc_precision_at_3_max
value: 38.48807924293779
- type: nauc_precision_at_3_std
value: 10.764232529972658
- type: nauc_precision_at_5_diff1
value: 19.709759176886067
- type: nauc_precision_at_5_max
value: 36.27102876926324
- type: nauc_precision_at_5_std
value: 13.660107662673637
- type: nauc_recall_at_1000_diff1
value: 45.983533767225694
- type: nauc_recall_at_1000_max
value: 57.47414401478259
- type: nauc_recall_at_1000_std
value: 48.046202249413845
- type: nauc_recall_at_100_diff1
value: 32.07866475386506
- type: nauc_recall_at_100_max
value: 33.96383964758634
- type: nauc_recall_at_100_std
value: 12.252531516000694
- type: nauc_recall_at_10_diff1
value: 37.975164089633495
- type: nauc_recall_at_10_max
value: 30.871630327404432
- type: nauc_recall_at_10_std
value: -3.471373265508202
- type: nauc_recall_at_1_diff1
value: 53.39584452417071
- type: nauc_recall_at_1_max
value: 30.44376469140723
- type: nauc_recall_at_1_std
value: -6.88828726087523
- type: nauc_recall_at_20_diff1
value: 35.405722893633786
- type: nauc_recall_at_20_max
value: 30.02056108542106
- type: nauc_recall_at_20_std
value: -2.612688358596724
- type: nauc_recall_at_3_diff1
value: 41.68732922410159
- type: nauc_recall_at_3_max
value: 33.76501870587258
- type: nauc_recall_at_3_std
value: -2.1010829302018146
- type: nauc_recall_at_5_diff1
value: 41.076140933019545
- type: nauc_recall_at_5_max
value: 33.11420354771476
- type: nauc_recall_at_5_std
value: -2.37285059292278
- type: ndcg_at_1
value: 34.937000000000005
- type: ndcg_at_10
value: 45.95
- type: ndcg_at_100
value: 51.768
- type: ndcg_at_1000
value: 53.612
- type: ndcg_at_20
value: 48.309000000000005
- type: ndcg_at_3
value: 40.186
- type: ndcg_at_5
value: 43.111
- type: precision_at_1
value: 34.937000000000005
- type: precision_at_10
value: 8.73
- type: precision_at_100
value: 1.397
- type: precision_at_1000
value: 0.174
- type: precision_at_20
value: 5.135
- type: precision_at_3
value: 19.858999999999998
- type: precision_at_5
value: 14.456
- type: recall_at_1
value: 28.138999999999996
- type: recall_at_10
value: 59.646
- type: recall_at_100
value: 83.813
- type: recall_at_1000
value: 95.69800000000001
- type: recall_at_20
value: 68.09100000000001
- type: recall_at_3
value: 43.15
- type: recall_at_5
value: 50.876
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackProgrammersRetrieval (default)
type: mteb/cqadupstack-programmers
config: default
split: test
revision: 6184bc1440d2dbc7612be22b50686b8826d22b32
metrics:
- type: main_score
value: 38.568999999999996
- type: map_at_1
value: 23.288
- type: map_at_10
value: 32.771
- type: map_at_100
value: 34.146
- type: map_at_1000
value: 34.278
- type: map_at_20
value: 33.522999999999996
- type: map_at_3
value: 29.643000000000004
- type: map_at_5
value: 31.188
- type: mrr_at_1
value: 28.424657534246577
- type: mrr_at_10
value: 37.56790425454804
- type: mrr_at_100
value: 38.512708758710254
- type: mrr_at_1000
value: 38.58372536901956
- type: mrr_at_20
value: 38.13718150828231
- type: mrr_at_3
value: 34.874429223744286
- type: mrr_at_5
value: 36.34703196347027
- type: nauc_map_at_1000_diff1
value: 42.49932971514879
- type: nauc_map_at_1000_max
value: 34.44449252084838
- type: nauc_map_at_1000_std
value: 0.2820810427961318
- type: nauc_map_at_100_diff1
value: 42.48316842005717
- type: nauc_map_at_100_max
value: 34.439643249096655
- type: nauc_map_at_100_std
value: 0.2911248084637576
- type: nauc_map_at_10_diff1
value: 42.591951745001865
- type: nauc_map_at_10_max
value: 33.848598117618984
- type: nauc_map_at_10_std
value: -0.23217641524494498
- type: nauc_map_at_1_diff1
value: 48.9034806154442
- type: nauc_map_at_1_max
value: 29.896917123056483
- type: nauc_map_at_1_std
value: -2.439150046473878
- type: nauc_map_at_20_diff1
value: 42.52245588485265
- type: nauc_map_at_20_max
value: 34.33890957691394
- type: nauc_map_at_20_std
value: 0.04133780224148374
- type: nauc_map_at_3_diff1
value: 44.020947516815916
- type: nauc_map_at_3_max
value: 33.15603327298969
- type: nauc_map_at_3_std
value: -1.6868324998110078
- type: nauc_map_at_5_diff1
value: 42.82799596536379
- type: nauc_map_at_5_max
value: 33.5235389139028
- type: nauc_map_at_5_std
value: -1.233132343609442
- type: nauc_mrr_at_1000_diff1
value: 40.680718239040495
- type: nauc_mrr_at_1000_max
value: 35.142026511262294
- type: nauc_mrr_at_1000_std
value: 0.7212094420333764
- type: nauc_mrr_at_100_diff1
value: 40.674003566538424
- type: nauc_mrr_at_100_max
value: 35.14079055322123
- type: nauc_mrr_at_100_std
value: 0.73908883788415
- type: nauc_mrr_at_10_diff1
value: 40.65562886804231
- type: nauc_mrr_at_10_max
value: 34.98579748882993
- type: nauc_mrr_at_10_std
value: 0.4344145401378685
- type: nauc_mrr_at_1_diff1
value: 45.9859248383466
- type: nauc_mrr_at_1_max
value: 33.57626905034157
- type: nauc_mrr_at_1_std
value: -0.7348734835041326
- type: nauc_mrr_at_20_diff1
value: 40.6576258022829
- type: nauc_mrr_at_20_max
value: 35.154195627553406
- type: nauc_mrr_at_20_std
value: 0.634477977849603
- type: nauc_mrr_at_3_diff1
value: 41.488252532122885
- type: nauc_mrr_at_3_max
value: 35.1085446216396
- type: nauc_mrr_at_3_std
value: -0.26518839484649515
- type: nauc_mrr_at_5_diff1
value: 40.420541677182555
- type: nauc_mrr_at_5_max
value: 35.22172462701993
- type: nauc_mrr_at_5_std
value: -0.25848045409750375
- type: nauc_ndcg_at_1000_diff1
value: 40.546229948037514
- type: nauc_ndcg_at_1000_max
value: 35.55631228234972
- type: nauc_ndcg_at_1000_std
value: 2.9837182102880524
- type: nauc_ndcg_at_100_diff1
value: 40.09271767110789
- type: nauc_ndcg_at_100_max
value: 35.854478264686854
- type: nauc_ndcg_at_100_std
value: 3.792167239853245
- type: nauc_ndcg_at_10_diff1
value: 40.3083818339378
- type: nauc_ndcg_at_10_max
value: 34.51597389995391
- type: nauc_ndcg_at_10_std
value: 0.9186474992446686
- type: nauc_ndcg_at_1_diff1
value: 45.9859248383466
- type: nauc_ndcg_at_1_max
value: 33.57626905034157
- type: nauc_ndcg_at_1_std
value: -0.7348734835041326
- type: nauc_ndcg_at_20_diff1
value: 40.04167502138235
- type: nauc_ndcg_at_20_max
value: 35.66297308624731
- type: nauc_ndcg_at_20_std
value: 1.9289709947356306
- type: nauc_ndcg_at_3_diff1
value: 42.008639920375195
- type: nauc_ndcg_at_3_max
value: 35.013390036837244
- type: nauc_ndcg_at_3_std
value: -1.0593134654956005
- type: nauc_ndcg_at_5_diff1
value: 40.249687670087816
- type: nauc_ndcg_at_5_max
value: 34.6889269216984
- type: nauc_ndcg_at_5_std
value: -0.7707068993680779
- type: nauc_precision_at_1000_diff1
value: -8.682121420487263
- type: nauc_precision_at_1000_max
value: 0.7494169797223946
- type: nauc_precision_at_1000_std
value: 1.1392663482283298
- type: nauc_precision_at_100_diff1
value: 1.518759603729437
- type: nauc_precision_at_100_max
value: 16.851433601465114
- type: nauc_precision_at_100_std
value: 10.931374575070386
- type: nauc_precision_at_10_diff1
value: 16.794492920670503
- type: nauc_precision_at_10_max
value: 32.02893820418532
- type: nauc_precision_at_10_std
value: 6.61722505687105
- type: nauc_precision_at_1_diff1
value: 45.9859248383466
- type: nauc_precision_at_1_max
value: 33.57626905034157
- type: nauc_precision_at_1_std
value: -0.7348734835041326
- type: nauc_precision_at_20_diff1
value: 11.0013674864101
- type: nauc_precision_at_20_max
value: 30.81307549333853
- type: nauc_precision_at_20_std
value: 8.876146343359824
- type: nauc_precision_at_3_diff1
value: 30.358294150549263
- type: nauc_precision_at_3_max
value: 38.543301462870424
- type: nauc_precision_at_3_std
value: 2.867784756736603
- type: nauc_precision_at_5_diff1
value: 22.94487457136215
- type: nauc_precision_at_5_max
value: 37.54100803628419
- type: nauc_precision_at_5_std
value: 3.839791032775397
- type: nauc_recall_at_1000_diff1
value: 28.970034771420355
- type: nauc_recall_at_1000_max
value: 38.787768893627785
- type: nauc_recall_at_1000_std
value: 53.81040950703603
- type: nauc_recall_at_100_diff1
value: 29.583524837481896
- type: nauc_recall_at_100_max
value: 37.202619359626986
- type: nauc_recall_at_100_std
value: 24.086399035427473
- type: nauc_recall_at_10_diff1
value: 33.45660381923287
- type: nauc_recall_at_10_max
value: 31.879186393042907
- type: nauc_recall_at_10_std
value: 3.6546901922261297
- type: nauc_recall_at_1_diff1
value: 48.9034806154442
- type: nauc_recall_at_1_max
value: 29.896917123056483
- type: nauc_recall_at_1_std
value: -2.439150046473878
- type: nauc_recall_at_20_diff1
value: 31.821081500540053
- type: nauc_recall_at_20_max
value: 35.5790141630673
- type: nauc_recall_at_20_std
value: 7.549207597971351
- type: nauc_recall_at_3_diff1
value: 38.99447500514566
- type: nauc_recall_at_3_max
value: 33.57706513619638
- type: nauc_recall_at_3_std
value: -1.5778066212304678
- type: nauc_recall_at_5_diff1
value: 34.06626536124258
- type: nauc_recall_at_5_max
value: 32.71331577702591
- type: nauc_recall_at_5_std
value: -1.3814107710485413
- type: ndcg_at_1
value: 28.425
- type: ndcg_at_10
value: 38.568999999999996
- type: ndcg_at_100
value: 44.529
- type: ndcg_at_1000
value: 47.231
- type: ndcg_at_20
value: 40.877
- type: ndcg_at_3
value: 33.289
- type: ndcg_at_5
value: 35.480000000000004
- type: precision_at_1
value: 28.425
- type: precision_at_10
value: 7.317
- type: precision_at_100
value: 1.2109999999999999
- type: precision_at_1000
value: 0.163
- type: precision_at_20
value: 4.3549999999999995
- type: precision_at_3
value: 16.096
- type: precision_at_5
value: 11.507000000000001
- type: recall_at_1
value: 23.288
- type: recall_at_10
value: 51.010999999999996
- type: recall_at_100
value: 76.566
- type: recall_at_1000
value: 94.774
- type: recall_at_20
value: 59.24
- type: recall_at_3
value: 36.236000000000004
- type: recall_at_5
value: 42.243
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackRetrieval (default)
type: CQADupstackRetrieval_is_a_combined_dataset
config: default
split: test
revision: CQADupstackRetrieval_is_a_combined_dataset
metrics:
- type: main_score
value: 40.812583333333336
- type: ndcg_at_10
value: 40.812583333333336
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackStatsRetrieval (default)
type: mteb/cqadupstack-stats
config: default
split: test
revision: 65ac3a16b8e91f9cee4c9828cc7c335575432a2a
metrics:
- type: main_score
value: 34.053
- type: map_at_1
value: 21.135
- type: map_at_10
value: 29.336000000000002
- type: map_at_100
value: 30.429000000000002
- type: map_at_1000
value: 30.523
- type: map_at_20
value: 29.946
- type: map_at_3
value: 26.751
- type: map_at_5
value: 28.147
- type: mrr_at_1
value: 24.079754601226995
- type: mrr_at_10
value: 31.98570941669101
- type: mrr_at_100
value: 32.87704399381717
- type: mrr_at_1000
value: 32.950784120634474
- type: mrr_at_20
value: 32.48647576315244
- type: mrr_at_3
value: 29.933537832310854
- type: mrr_at_5
value: 30.99182004089981
- type: nauc_map_at_1000_diff1
value: 40.54921025385474
- type: nauc_map_at_1000_max
value: 34.77575879898821
- type: nauc_map_at_1000_std
value: 8.948775266395762
- type: nauc_map_at_100_diff1
value: 40.54349367224357
- type: nauc_map_at_100_max
value: 34.757684235242756
- type: nauc_map_at_100_std
value: 8.930005514047473
- type: nauc_map_at_10_diff1
value: 40.6730737054752
- type: nauc_map_at_10_max
value: 34.378525509672194
- type: nauc_map_at_10_std
value: 7.989204273097626
- type: nauc_map_at_1_diff1
value: 49.37407707269144
- type: nauc_map_at_1_max
value: 32.21678855112054
- type: nauc_map_at_1_std
value: 3.854422840353318
- type: nauc_map_at_20_diff1
value: 40.632949610967124
- type: nauc_map_at_20_max
value: 34.59900516061919
- type: nauc_map_at_20_std
value: 8.565328686423722
- type: nauc_map_at_3_diff1
value: 41.277384595957884
- type: nauc_map_at_3_max
value: 33.206185913755924
- type: nauc_map_at_3_std
value: 5.717825876602881
- type: nauc_map_at_5_diff1
value: 40.81938763414844
- type: nauc_map_at_5_max
value: 34.601640205395725
- type: nauc_map_at_5_std
value: 7.959501488950282
- type: nauc_mrr_at_1000_diff1
value: 41.643791133769284
- type: nauc_mrr_at_1000_max
value: 36.86860098522057
- type: nauc_mrr_at_1000_std
value: 11.16203371889211
- type: nauc_mrr_at_100_diff1
value: 41.638414475074086
- type: nauc_mrr_at_100_max
value: 36.8731893852322
- type: nauc_mrr_at_100_std
value: 11.160133493075861
- type: nauc_mrr_at_10_diff1
value: 41.75277747893673
- type: nauc_mrr_at_10_max
value: 36.66922318811368
- type: nauc_mrr_at_10_std
value: 10.524888054922569
- type: nauc_mrr_at_1_diff1
value: 49.68943847753842
- type: nauc_mrr_at_1_max
value: 35.88824714503076
- type: nauc_mrr_at_1_std
value: 7.9139229045391515
- type: nauc_mrr_at_20_diff1
value: 41.6522451889601
- type: nauc_mrr_at_20_max
value: 36.89632890368892
- type: nauc_mrr_at_20_std
value: 11.00866443920579
- type: nauc_mrr_at_3_diff1
value: 41.740543551542395
- type: nauc_mrr_at_3_max
value: 36.637108493657536
- type: nauc_mrr_at_3_std
value: 9.550432374486645
- type: nauc_mrr_at_5_diff1
value: 41.74370752738844
- type: nauc_mrr_at_5_max
value: 36.88456148864922
- type: nauc_mrr_at_5_std
value: 10.424655726913185
- type: nauc_ndcg_at_1000_diff1
value: 38.51408332292014
- type: nauc_ndcg_at_1000_max
value: 36.09571315118831
- type: nauc_ndcg_at_1000_std
value: 13.474870360532162
- type: nauc_ndcg_at_100_diff1
value: 38.019174212684376
- type: nauc_ndcg_at_100_max
value: 35.93119149466668
- type: nauc_ndcg_at_100_std
value: 13.2297549585252
- type: nauc_ndcg_at_10_diff1
value: 38.916423405975245
- type: nauc_ndcg_at_10_max
value: 35.08064464520372
- type: nauc_ndcg_at_10_std
value: 9.782593893240483
- type: nauc_ndcg_at_1_diff1
value: 49.68943847753842
- type: nauc_ndcg_at_1_max
value: 35.88824714503076
- type: nauc_ndcg_at_1_std
value: 7.9139229045391515
- type: nauc_ndcg_at_20_diff1
value: 38.63750293826562
- type: nauc_ndcg_at_20_max
value: 35.739247440468915
- type: nauc_ndcg_at_20_std
value: 11.539481993254595
- type: nauc_ndcg_at_3_diff1
value: 39.28042269065498
- type: nauc_ndcg_at_3_max
value: 34.63564871034042
- type: nauc_ndcg_at_3_std
value: 7.49191071861962
- type: nauc_ndcg_at_5_diff1
value: 39.003890765463304
- type: nauc_ndcg_at_5_max
value: 35.71265272584242
- type: nauc_ndcg_at_5_std
value: 9.927346415193906
- type: nauc_precision_at_1000_diff1
value: -4.072989249391745
- type: nauc_precision_at_1000_max
value: 19.44622146624579
- type: nauc_precision_at_1000_std
value: 27.981364695896975
- type: nauc_precision_at_100_diff1
value: 7.321844285435558
- type: nauc_precision_at_100_max
value: 30.52844606095081
- type: nauc_precision_at_100_std
value: 31.072929769770514
- type: nauc_precision_at_10_diff1
value: 21.853502178071917
- type: nauc_precision_at_10_max
value: 37.30474313788762
- type: nauc_precision_at_10_std
value: 21.687548959106135
- type: nauc_precision_at_1_diff1
value: 49.68943847753842
- type: nauc_precision_at_1_max
value: 35.88824714503076
- type: nauc_precision_at_1_std
value: 7.9139229045391515
- type: nauc_precision_at_20_diff1
value: 17.749818352515526
- type: nauc_precision_at_20_max
value: 35.765806180376416
- type: nauc_precision_at_20_std
value: 25.411075024216252
- type: nauc_precision_at_3_diff1
value: 29.632039358651944
- type: nauc_precision_at_3_max
value: 39.65790064788028
- type: nauc_precision_at_3_std
value: 16.540404435515757
- type: nauc_precision_at_5_diff1
value: 25.65567821866759
- type: nauc_precision_at_5_max
value: 41.39961363618346
- type: nauc_precision_at_5_std
value: 22.5672145347772
- type: nauc_recall_at_1000_diff1
value: 25.719490205486846
- type: nauc_recall_at_1000_max
value: 34.913128477025914
- type: nauc_recall_at_1000_std
value: 45.9261202538083
- type: nauc_recall_at_100_diff1
value: 25.23717510439502
- type: nauc_recall_at_100_max
value: 33.50919363278291
- type: nauc_recall_at_100_std
value: 26.975483944416958
- type: nauc_recall_at_10_diff1
value: 32.84494569562565
- type: nauc_recall_at_10_max
value: 32.92279016208031
- type: nauc_recall_at_10_std
value: 11.273741379749255
- type: nauc_recall_at_1_diff1
value: 49.37407707269144
- type: nauc_recall_at_1_max
value: 32.21678855112054
- type: nauc_recall_at_1_std
value: 3.854422840353318
- type: nauc_recall_at_20_diff1
value: 31.136285249615742
- type: nauc_recall_at_20_max
value: 34.60047592054581
- type: nauc_recall_at_20_std
value: 17.167916916392333
- type: nauc_recall_at_3_diff1
value: 34.20242999662983
- type: nauc_recall_at_3_max
value: 31.922366101615964
- type: nauc_recall_at_3_std
value: 5.5073356683470065
- type: nauc_recall_at_5_diff1
value: 33.737565539089545
- type: nauc_recall_at_5_max
value: 35.18290702820585
- type: nauc_recall_at_5_std
value: 11.516166260788946
- type: ndcg_at_1
value: 24.08
- type: ndcg_at_10
value: 34.053
- type: ndcg_at_100
value: 39.214
- type: ndcg_at_1000
value: 41.672
- type: ndcg_at_20
value: 35.998999999999995
- type: ndcg_at_3
value: 29.429
- type: ndcg_at_5
value: 31.480000000000004
- type: precision_at_1
value: 24.08
- type: precision_at_10
value: 5.752
- type: precision_at_100
value: 0.906
- type: precision_at_1000
value: 0.11900000000000001
- type: precision_at_20
value: 3.3970000000000002
- type: precision_at_3
value: 13.344000000000001
- type: precision_at_5
value: 9.417
- type: recall_at_1
value: 21.135
- type: recall_at_10
value: 45.82
- type: recall_at_100
value: 69.414
- type: recall_at_1000
value: 87.653
- type: recall_at_20
value: 53.044999999999995
- type: recall_at_3
value: 32.914
- type: recall_at_5
value: 37.967
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackTexRetrieval (default)
type: mteb/cqadupstack-tex
config: default
split: test
revision: 46989137a86843e03a6195de44b09deda022eec7
metrics:
- type: main_score
value: 31.079
- type: map_at_1
value: 17.498
- type: map_at_10
value: 25.761
- type: map_at_100
value: 27.117
- type: map_at_1000
value: 27.249000000000002
- type: map_at_20
value: 26.487
- type: map_at_3
value: 22.8
- type: map_at_5
value: 24.404
- type: mrr_at_1
value: 21.92016517549897
- type: mrr_at_10
value: 29.885472312347773
- type: mrr_at_100
value: 30.9160324896262
- type: mrr_at_1000
value: 30.991168841711332
- type: mrr_at_20
value: 30.452602988671234
- type: mrr_at_3
value: 27.208075246616247
- type: mrr_at_5
value: 28.703257627896395
- type: nauc_map_at_1000_diff1
value: 31.106783614540877
- type: nauc_map_at_1000_max
value: 28.6225885387827
- type: nauc_map_at_1000_std
value: 6.0841248710509515
- type: nauc_map_at_100_diff1
value: 31.08489579492722
- type: nauc_map_at_100_max
value: 28.557756475282236
- type: nauc_map_at_100_std
value: 6.052003313810615
- type: nauc_map_at_10_diff1
value: 31.48175165089029
- type: nauc_map_at_10_max
value: 28.385150030692763
- type: nauc_map_at_10_std
value: 5.495659063667348
- type: nauc_map_at_1_diff1
value: 38.27868635676493
- type: nauc_map_at_1_max
value: 26.68660889337588
- type: nauc_map_at_1_std
value: 3.160872467743023
- type: nauc_map_at_20_diff1
value: 31.207065661313653
- type: nauc_map_at_20_max
value: 28.409403987888872
- type: nauc_map_at_20_std
value: 5.663005924021793
- type: nauc_map_at_3_diff1
value: 33.081360769026695
- type: nauc_map_at_3_max
value: 28.052813330220555
- type: nauc_map_at_3_std
value: 4.494259362042225
- type: nauc_map_at_5_diff1
value: 31.969111899422543
- type: nauc_map_at_5_max
value: 27.92017739432563
- type: nauc_map_at_5_std
value: 4.811795252340103
- type: nauc_mrr_at_1000_diff1
value: 29.29292415286063
- type: nauc_mrr_at_1000_max
value: 29.213774281822502
- type: nauc_mrr_at_1000_std
value: 7.033800724186415
- type: nauc_mrr_at_100_diff1
value: 29.2626300487415
- type: nauc_mrr_at_100_max
value: 29.204727919066332
- type: nauc_mrr_at_100_std
value: 7.045304952598549
- type: nauc_mrr_at_10_diff1
value: 29.387229358683616
- type: nauc_mrr_at_10_max
value: 29.294544480801417
- type: nauc_mrr_at_10_std
value: 6.775611131886977
- type: nauc_mrr_at_1_diff1
value: 34.788621456797806
- type: nauc_mrr_at_1_max
value: 28.589390080521447
- type: nauc_mrr_at_1_std
value: 5.031155628869166
- type: nauc_mrr_at_20_diff1
value: 29.269757265619585
- type: nauc_mrr_at_20_max
value: 29.192868249024183
- type: nauc_mrr_at_20_std
value: 6.895053312400131
- type: nauc_mrr_at_3_diff1
value: 30.45886158188182
- type: nauc_mrr_at_3_max
value: 29.126637772912332
- type: nauc_mrr_at_3_std
value: 5.887845131702013
- type: nauc_mrr_at_5_diff1
value: 29.5830069273922
- type: nauc_mrr_at_5_max
value: 29.023599660723566
- type: nauc_mrr_at_5_std
value: 6.333484322706432
- type: nauc_ndcg_at_1000_diff1
value: 28.299997368370704
- type: nauc_ndcg_at_1000_max
value: 29.888737990959356
- type: nauc_ndcg_at_1000_std
value: 8.946268519586546
- type: nauc_ndcg_at_100_diff1
value: 27.626227219466298
- type: nauc_ndcg_at_100_max
value: 29.449363310563655
- type: nauc_ndcg_at_100_std
value: 9.152695620470102
- type: nauc_ndcg_at_10_diff1
value: 28.712684866650427
- type: nauc_ndcg_at_10_max
value: 29.193277467967206
- type: nauc_ndcg_at_10_std
value: 6.998413207889944
- type: nauc_ndcg_at_1_diff1
value: 34.788621456797806
- type: nauc_ndcg_at_1_max
value: 28.589390080521447
- type: nauc_ndcg_at_1_std
value: 5.031155628869166
- type: nauc_ndcg_at_20_diff1
value: 28.12146242755772
- type: nauc_ndcg_at_20_max
value: 28.9603042379236
- type: nauc_ndcg_at_20_std
value: 7.4783571354648055
- type: nauc_ndcg_at_3_diff1
value: 30.7052417569527
- type: nauc_ndcg_at_3_max
value: 29.20315655011043
- type: nauc_ndcg_at_3_std
value: 5.388849801543028
- type: nauc_ndcg_at_5_diff1
value: 29.36802229342131
- type: nauc_ndcg_at_5_max
value: 28.574869226548653
- type: nauc_ndcg_at_5_std
value: 5.792067593971809
- type: nauc_precision_at_1000_diff1
value: -1.0955902695363287
- type: nauc_precision_at_1000_max
value: 22.263203721640487
- type: nauc_precision_at_1000_std
value: 15.962211276640174
- type: nauc_precision_at_100_diff1
value: 1.697299406682818
- type: nauc_precision_at_100_max
value: 25.552155435237268
- type: nauc_precision_at_100_std
value: 19.630389535049883
- type: nauc_precision_at_10_diff1
value: 13.297756573713581
- type: nauc_precision_at_10_max
value: 30.477988337514095
- type: nauc_precision_at_10_std
value: 11.930425415137833
- type: nauc_precision_at_1_diff1
value: 34.788621456797806
- type: nauc_precision_at_1_max
value: 28.589390080521447
- type: nauc_precision_at_1_std
value: 5.031155628869166
- type: nauc_precision_at_20_diff1
value: 9.074294316929793
- type: nauc_precision_at_20_max
value: 28.035274038120022
- type: nauc_precision_at_20_std
value: 13.372503140399717
- type: nauc_precision_at_3_diff1
value: 22.46944870859152
- type: nauc_precision_at_3_max
value: 30.62699826426078
- type: nauc_precision_at_3_std
value: 7.446849435999603
- type: nauc_precision_at_5_diff1
value: 17.272667706952305
- type: nauc_precision_at_5_max
value: 29.24179840910647
- type: nauc_precision_at_5_std
value: 8.644639430581861
- type: nauc_recall_at_1000_diff1
value: 12.871611027617524
- type: nauc_recall_at_1000_max
value: 36.71435304856158
- type: nauc_recall_at_1000_std
value: 28.620443354999125
- type: nauc_recall_at_100_diff1
value: 14.961011668882907
- type: nauc_recall_at_100_max
value: 28.301006654430267
- type: nauc_recall_at_100_std
value: 19.460217217927635
- type: nauc_recall_at_10_diff1
value: 21.85511402624238
- type: nauc_recall_at_10_max
value: 27.432304380395227
- type: nauc_recall_at_10_std
value: 8.78952978670911
- type: nauc_recall_at_1_diff1
value: 38.27868635676493
- type: nauc_recall_at_1_max
value: 26.68660889337588
- type: nauc_recall_at_1_std
value: 3.160872467743023
- type: nauc_recall_at_20_diff1
value: 19.752305046339135
- type: nauc_recall_at_20_max
value: 26.236548609605155
- type: nauc_recall_at_20_std
value: 10.087194624804992
- type: nauc_recall_at_3_diff1
value: 27.751788017220413
- type: nauc_recall_at_3_max
value: 27.44474563712776
- type: nauc_recall_at_3_std
value: 5.050595287046713
- type: nauc_recall_at_5_diff1
value: 24.20044216232329
- type: nauc_recall_at_5_max
value: 26.30261864862886
- type: nauc_recall_at_5_std
value: 5.945475035324601
- type: ndcg_at_1
value: 21.92
- type: ndcg_at_10
value: 31.079
- type: ndcg_at_100
value: 37.254
- type: ndcg_at_1000
value: 40.075
- type: ndcg_at_20
value: 33.331
- type: ndcg_at_3
value: 25.889
- type: ndcg_at_5
value: 28.253
- type: precision_at_1
value: 21.92
- type: precision_at_10
value: 6.032
- type: precision_at_100
value: 1.088
- type: precision_at_1000
value: 0.152
- type: precision_at_20
value: 3.6839999999999997
- type: precision_at_3
value: 12.549
- type: precision_at_5
value: 9.36
- type: recall_at_1
value: 17.498
- type: recall_at_10
value: 42.824
- type: recall_at_100
value: 70.537
- type: recall_at_1000
value: 90.416
- type: recall_at_20
value: 51.13799999999999
- type: recall_at_3
value: 28.374
- type: recall_at_5
value: 34.452
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackUnixRetrieval (default)
type: mteb/cqadupstack-unix
config: default
split: test
revision: 6c6430d3a6d36f8d2a829195bc5dc94d7e063e53
metrics:
- type: main_score
value: 43.159
- type: map_at_1
value: 27.3
- type: map_at_10
value: 37.471
- type: map_at_100
value: 38.847
- type: map_at_1000
value: 38.95
- type: map_at_20
value: 38.238
- type: map_at_3
value: 34.343
- type: map_at_5
value: 35.731
- type: mrr_at_1
value: 32.46268656716418
- type: mrr_at_10
value: 41.2610681710495
- type: mrr_at_100
value: 42.31204325599387
- type: mrr_at_1000
value: 42.36679795654589
- type: mrr_at_20
value: 41.91619388699724
- type: mrr_at_3
value: 38.75932835820893
- type: mrr_at_5
value: 39.86473880597011
- type: nauc_map_at_1000_diff1
value: 38.41373782401318
- type: nauc_map_at_1000_max
value: 36.26799123342812
- type: nauc_map_at_1000_std
value: -1.6362179601270626
- type: nauc_map_at_100_diff1
value: 38.39090799924858
- type: nauc_map_at_100_max
value: 36.25156851372077
- type: nauc_map_at_100_std
value: -1.6616566136661386
- type: nauc_map_at_10_diff1
value: 38.696198873516394
- type: nauc_map_at_10_max
value: 36.16128806023898
- type: nauc_map_at_10_std
value: -2.130791362526739
- type: nauc_map_at_1_diff1
value: 43.37285405382472
- type: nauc_map_at_1_max
value: 31.642684614767276
- type: nauc_map_at_1_std
value: -5.540771932525815
- type: nauc_map_at_20_diff1
value: 38.4878820103522
- type: nauc_map_at_20_max
value: 36.2292631214085
- type: nauc_map_at_20_std
value: -1.843545810350474
- type: nauc_map_at_3_diff1
value: 39.95370528114123
- type: nauc_map_at_3_max
value: 34.23886393938281
- type: nauc_map_at_3_std
value: -4.809386044288499
- type: nauc_map_at_5_diff1
value: 39.16911049006275
- type: nauc_map_at_5_max
value: 35.2191378957676
- type: nauc_map_at_5_std
value: -3.5005953042339653
- type: nauc_mrr_at_1000_diff1
value: 36.18437691856584
- type: nauc_mrr_at_1000_max
value: 37.70260458140709
- type: nauc_mrr_at_1000_std
value: -0.11426638420398647
- type: nauc_mrr_at_100_diff1
value: 36.14459892119925
- type: nauc_mrr_at_100_max
value: 37.67599501718511
- type: nauc_mrr_at_100_std
value: -0.12595268985648042
- type: nauc_mrr_at_10_diff1
value: 36.27951441810734
- type: nauc_mrr_at_10_max
value: 37.8613553159427
- type: nauc_mrr_at_10_std
value: -0.20119672771815636
- type: nauc_mrr_at_1_diff1
value: 39.88957563396634
- type: nauc_mrr_at_1_max
value: 36.60257391624788
- type: nauc_mrr_at_1_std
value: -2.841791526837743
- type: nauc_mrr_at_20_diff1
value: 36.103682088537184
- type: nauc_mrr_at_20_max
value: 37.7488620725621
- type: nauc_mrr_at_20_std
value: -0.08430051028531159
- type: nauc_mrr_at_3_diff1
value: 36.8623555151354
- type: nauc_mrr_at_3_max
value: 37.36094868304334
- type: nauc_mrr_at_3_std
value: -1.8677123750289681
- type: nauc_mrr_at_5_diff1
value: 36.59448977125284
- type: nauc_mrr_at_5_max
value: 37.8339219805799
- type: nauc_mrr_at_5_std
value: -1.1457629757330636
- type: nauc_ndcg_at_1000_diff1
value: 36.0265401975872
- type: nauc_ndcg_at_1000_max
value: 37.45317419626094
- type: nauc_ndcg_at_1000_std
value: 1.6036524279201412
- type: nauc_ndcg_at_100_diff1
value: 35.29323774869888
- type: nauc_ndcg_at_100_max
value: 37.10845147549922
- type: nauc_ndcg_at_100_std
value: 1.5535678575966316
- type: nauc_ndcg_at_10_diff1
value: 36.180138557973876
- type: nauc_ndcg_at_10_max
value: 37.411761737356
- type: nauc_ndcg_at_10_std
value: 0.5165748912750906
- type: nauc_ndcg_at_1_diff1
value: 39.88957563396634
- type: nauc_ndcg_at_1_max
value: 36.60257391624788
- type: nauc_ndcg_at_1_std
value: -2.841791526837743
- type: nauc_ndcg_at_20_diff1
value: 35.610864685140825
- type: nauc_ndcg_at_20_max
value: 37.34524578592843
- type: nauc_ndcg_at_20_std
value: 1.2655274462928487
- type: nauc_ndcg_at_3_diff1
value: 37.581668636974165
- type: nauc_ndcg_at_3_max
value: 35.69832549832282
- type: nauc_ndcg_at_3_std
value: -3.6448795589159597
- type: nauc_ndcg_at_5_diff1
value: 36.991833727312326
- type: nauc_ndcg_at_5_max
value: 36.30298998287306
- type: nauc_ndcg_at_5_std
value: -2.2622164190279284
- type: nauc_precision_at_1000_diff1
value: -13.70160706618185
- type: nauc_precision_at_1000_max
value: 5.460656856541568
- type: nauc_precision_at_1000_std
value: 12.62097770232154
- type: nauc_precision_at_100_diff1
value: -6.499308352387093
- type: nauc_precision_at_100_max
value: 20.353724456457627
- type: nauc_precision_at_100_std
value: 17.95090127045124
- type: nauc_precision_at_10_diff1
value: 12.806170332318779
- type: nauc_precision_at_10_max
value: 39.23341562181076
- type: nauc_precision_at_10_std
value: 13.987639599584101
- type: nauc_precision_at_1_diff1
value: 39.88957563396634
- type: nauc_precision_at_1_max
value: 36.60257391624788
- type: nauc_precision_at_1_std
value: -2.841791526837743
- type: nauc_precision_at_20_diff1
value: 5.332995851304963
- type: nauc_precision_at_20_max
value: 34.53916605109217
- type: nauc_precision_at_20_std
value: 17.69712324408807
- type: nauc_precision_at_3_diff1
value: 25.4753762848704
- type: nauc_precision_at_3_max
value: 38.50846428335884
- type: nauc_precision_at_3_std
value: 1.9628066846211143
- type: nauc_precision_at_5_diff1
value: 20.301539878169468
- type: nauc_precision_at_5_max
value: 39.951735387397015
- type: nauc_precision_at_5_std
value: 6.793501039657923
- type: nauc_recall_at_1000_diff1
value: 22.203986100593294
- type: nauc_recall_at_1000_max
value: 48.075623185241035
- type: nauc_recall_at_1000_std
value: 50.75401433229061
- type: nauc_recall_at_100_diff1
value: 19.9395981394293
- type: nauc_recall_at_100_max
value: 32.7460953002592
- type: nauc_recall_at_100_std
value: 14.324454594663344
- type: nauc_recall_at_10_diff1
value: 29.346896169220532
- type: nauc_recall_at_10_max
value: 36.68032432093537
- type: nauc_recall_at_10_std
value: 6.713544668966169
- type: nauc_recall_at_1_diff1
value: 43.37285405382472
- type: nauc_recall_at_1_max
value: 31.642684614767276
- type: nauc_recall_at_1_std
value: -5.540771932525815
- type: nauc_recall_at_20_diff1
value: 26.448398177241923
- type: nauc_recall_at_20_max
value: 35.7341164237797
- type: nauc_recall_at_20_std
value: 9.476528957596907
- type: nauc_recall_at_3_diff1
value: 35.01162975513612
- type: nauc_recall_at_3_max
value: 32.6400708326739
- type: nauc_recall_at_3_std
value: -5.272713547738796
- type: nauc_recall_at_5_diff1
value: 32.78451551990977
- type: nauc_recall_at_5_max
value: 34.48194190611458
- type: nauc_recall_at_5_std
value: -1.532660412317024
- type: ndcg_at_1
value: 32.462999999999994
- type: ndcg_at_10
value: 43.159
- type: ndcg_at_100
value: 49.052
- type: ndcg_at_1000
value: 51.132
- type: ndcg_at_20
value: 45.57
- type: ndcg_at_3
value: 37.735
- type: ndcg_at_5
value: 39.556000000000004
- type: precision_at_1
value: 32.462999999999994
- type: precision_at_10
value: 7.593
- type: precision_at_100
value: 1.187
- type: precision_at_1000
value: 0.147
- type: precision_at_20
value: 4.515000000000001
- type: precision_at_3
value: 17.32
- type: precision_at_5
value: 11.996
- type: recall_at_1
value: 27.3
- type: recall_at_10
value: 56.642
- type: recall_at_100
value: 81.525
- type: recall_at_1000
value: 95.806
- type: recall_at_20
value: 65.11099999999999
- type: recall_at_3
value: 41.469
- type: recall_at_5
value: 46.163
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWebmastersRetrieval (default)
type: mteb/cqadupstack-webmasters
config: default
split: test
revision: 160c094312a0e1facb97e55eeddb698c0abe3571
metrics:
- type: main_score
value: 41.297
- type: map_at_1
value: 24.239
- type: map_at_10
value: 34.958
- type: map_at_100
value: 36.758
- type: map_at_1000
value: 36.989
- type: map_at_20
value: 35.9
- type: map_at_3
value: 31.673000000000002
- type: map_at_5
value: 33.675
- type: mrr_at_1
value: 28.458498023715418
- type: mrr_at_10
value: 39.1959972394755
- type: mrr_at_100
value: 40.18847772327865
- type: mrr_at_1000
value: 40.23773735721479
- type: mrr_at_20
value: 39.774954617486664
- type: mrr_at_3
value: 36.29776021080371
- type: mrr_at_5
value: 38.155467720685124
- type: nauc_map_at_1000_diff1
value: 36.874966531359455
- type: nauc_map_at_1000_max
value: 32.925100088208865
- type: nauc_map_at_1000_std
value: 2.9838945484541397
- type: nauc_map_at_100_diff1
value: 37.05086874383793
- type: nauc_map_at_100_max
value: 32.93389217056399
- type: nauc_map_at_100_std
value: 2.9144257345054467
- type: nauc_map_at_10_diff1
value: 37.37730997185654
- type: nauc_map_at_10_max
value: 32.06774672790129
- type: nauc_map_at_10_std
value: 1.8429020159075118
- type: nauc_map_at_1_diff1
value: 43.561442228636224
- type: nauc_map_at_1_max
value: 30.139564573234708
- type: nauc_map_at_1_std
value: -4.277872763081037
- type: nauc_map_at_20_diff1
value: 37.35659896740046
- type: nauc_map_at_20_max
value: 32.5517336397131
- type: nauc_map_at_20_std
value: 2.4435488993754753
- type: nauc_map_at_3_diff1
value: 37.473068855167796
- type: nauc_map_at_3_max
value: 31.59112945346696
- type: nauc_map_at_3_std
value: 1.6092320432651679
- type: nauc_map_at_5_diff1
value: 38.01435766254902
- type: nauc_map_at_5_max
value: 31.922726165398448
- type: nauc_map_at_5_std
value: 1.5628995834664126
- type: nauc_mrr_at_1000_diff1
value: 36.12153078238164
- type: nauc_mrr_at_1000_max
value: 32.22061637683918
- type: nauc_mrr_at_1000_std
value: 3.5366279283280204
- type: nauc_mrr_at_100_diff1
value: 36.1188638601524
- type: nauc_mrr_at_100_max
value: 32.20952783773952
- type: nauc_mrr_at_100_std
value: 3.53367306991397
- type: nauc_mrr_at_10_diff1
value: 36.24545839812258
- type: nauc_mrr_at_10_max
value: 32.02781480404713
- type: nauc_mrr_at_10_std
value: 3.2237214571781756
- type: nauc_mrr_at_1_diff1
value: 39.43286392756712
- type: nauc_mrr_at_1_max
value: 31.633687884951282
- type: nauc_mrr_at_1_std
value: -0.27764785706458617
- type: nauc_mrr_at_20_diff1
value: 36.13109306056499
- type: nauc_mrr_at_20_max
value: 32.24845518587824
- type: nauc_mrr_at_20_std
value: 3.5527414665451666
- type: nauc_mrr_at_3_diff1
value: 35.694561950510526
- type: nauc_mrr_at_3_max
value: 32.34372143829758
- type: nauc_mrr_at_3_std
value: 3.2831272959106585
- type: nauc_mrr_at_5_diff1
value: 36.47407512657775
- type: nauc_mrr_at_5_max
value: 32.245870671508776
- type: nauc_mrr_at_5_std
value: 3.22208428921342
- type: nauc_ndcg_at_1000_diff1
value: 34.94498046337473
- type: nauc_ndcg_at_1000_max
value: 34.00274024637903
- type: nauc_ndcg_at_1000_std
value: 6.037949553908587
- type: nauc_ndcg_at_100_diff1
value: 35.072050841400085
- type: nauc_ndcg_at_100_max
value: 33.77530570776558
- type: nauc_ndcg_at_100_std
value: 6.476581022520731
- type: nauc_ndcg_at_10_diff1
value: 35.17349390611081
- type: nauc_ndcg_at_10_max
value: 31.74636270166893
- type: nauc_ndcg_at_10_std
value: 3.825524111282119
- type: nauc_ndcg_at_1_diff1
value: 39.43286392756712
- type: nauc_ndcg_at_1_max
value: 31.633687884951282
- type: nauc_ndcg_at_1_std
value: -0.27764785706458617
- type: nauc_ndcg_at_20_diff1
value: 35.390740077541615
- type: nauc_ndcg_at_20_max
value: 33.026639113297236
- type: nauc_ndcg_at_20_std
value: 5.161352869053754
- type: nauc_ndcg_at_3_diff1
value: 34.784893735708245
- type: nauc_ndcg_at_3_max
value: 32.37270211170083
- type: nauc_ndcg_at_3_std
value: 4.002202121330911
- type: nauc_ndcg_at_5_diff1
value: 36.13692738773426
- type: nauc_ndcg_at_5_max
value: 32.39302727687912
- type: nauc_ndcg_at_5_std
value: 3.6394963350009557
- type: nauc_precision_at_1000_diff1
value: -22.965728978391358
- type: nauc_precision_at_1000_max
value: 0.991199838881063
- type: nauc_precision_at_1000_std
value: 9.683600354073281
- type: nauc_precision_at_100_diff1
value: -12.560615008152604
- type: nauc_precision_at_100_max
value: 14.334232028365351
- type: nauc_precision_at_100_std
value: 17.640635157521896
- type: nauc_precision_at_10_diff1
value: 11.206224974161422
- type: nauc_precision_at_10_max
value: 27.636112811297668
- type: nauc_precision_at_10_std
value: 13.844463698498116
- type: nauc_precision_at_1_diff1
value: 39.43286392756712
- type: nauc_precision_at_1_max
value: 31.633687884951282
- type: nauc_precision_at_1_std
value: -0.27764785706458617
- type: nauc_precision_at_20_diff1
value: 2.820754738992166
- type: nauc_precision_at_20_max
value: 24.548395859228734
- type: nauc_precision_at_20_std
value: 18.659599473117076
- type: nauc_precision_at_3_diff1
value: 23.700089513000997
- type: nauc_precision_at_3_max
value: 33.926097310453535
- type: nauc_precision_at_3_std
value: 11.528262810991514
- type: nauc_precision_at_5_diff1
value: 19.952717493700728
- type: nauc_precision_at_5_max
value: 32.73420866701224
- type: nauc_precision_at_5_std
value: 12.572768747428661
- type: nauc_recall_at_1000_diff1
value: 11.338759676414469
- type: nauc_recall_at_1000_max
value: 60.27964607676189
- type: nauc_recall_at_1000_std
value: 65.64600084082495
- type: nauc_recall_at_100_diff1
value: 25.61894545906908
- type: nauc_recall_at_100_max
value: 33.053374407768985
- type: nauc_recall_at_100_std
value: 25.841122943882226
- type: nauc_recall_at_10_diff1
value: 28.374429812654515
- type: nauc_recall_at_10_max
value: 25.752628109349747
- type: nauc_recall_at_10_std
value: 5.460179271646167
- type: nauc_recall_at_1_diff1
value: 43.561442228636224
- type: nauc_recall_at_1_max
value: 30.139564573234708
- type: nauc_recall_at_1_std
value: -4.277872763081037
- type: nauc_recall_at_20_diff1
value: 28.096290358405124
- type: nauc_recall_at_20_max
value: 29.47090259172014
- type: nauc_recall_at_20_std
value: 10.769843783609222
- type: nauc_recall_at_3_diff1
value: 31.442770007498943
- type: nauc_recall_at_3_max
value: 30.06591425778911
- type: nauc_recall_at_3_std
value: 5.640565369414359
- type: nauc_recall_at_5_diff1
value: 32.95442456012167
- type: nauc_recall_at_5_max
value: 28.473408007093287
- type: nauc_recall_at_5_std
value: 4.467307517857356
- type: ndcg_at_1
value: 28.458
- type: ndcg_at_10
value: 41.297
- type: ndcg_at_100
value: 47.797
- type: ndcg_at_1000
value: 50.059
- type: ndcg_at_20
value: 43.856
- type: ndcg_at_3
value: 35.656
- type: ndcg_at_5
value: 38.779
- type: precision_at_1
value: 28.458
- type: precision_at_10
value: 8.004
- type: precision_at_100
value: 1.68
- type: precision_at_1000
value: 0.254
- type: precision_at_20
value: 5.1979999999999995
- type: precision_at_3
value: 16.864
- type: precision_at_5
value: 12.727
- type: recall_at_1
value: 24.239
- type: recall_at_10
value: 54.26499999999999
- type: recall_at_100
value: 82.101
- type: recall_at_1000
value: 95.994
- type: recall_at_20
value: 63.649
- type: recall_at_3
value: 39.206
- type: recall_at_5
value: 46.604
- task:
type: Retrieval
dataset:
name: MTEB CQADupstackWordpressRetrieval (default)
type: mteb/cqadupstack-wordpress
config: default
split: test
revision: 4ffe81d471b1924886b33c7567bfb200e9eec5c4
metrics:
- type: main_score
value: 33.832
- type: map_at_1
value: 20.003
- type: map_at_10
value: 28.642
- type: map_at_100
value: 29.839
- type: map_at_1000
value: 29.958000000000002
- type: map_at_20
value: 29.349999999999998
- type: map_at_3
value: 25.535000000000004
- type: map_at_5
value: 27.400000000000002
- type: mrr_at_1
value: 21.811460258780038
- type: mrr_at_10
value: 30.64372267699438
- type: mrr_at_100
value: 31.73852756929626
- type: mrr_at_1000
value: 31.815679800702124
- type: mrr_at_20
value: 31.299814331394977
- type: mrr_at_3
value: 28.034504004929133
- type: mrr_at_5
value: 29.624152803450386
- type: nauc_map_at_1000_diff1
value: 27.06604986940547
- type: nauc_map_at_1000_max
value: 30.733957522966943
- type: nauc_map_at_1000_std
value: 5.707667228631227
- type: nauc_map_at_100_diff1
value: 27.069243825104223
- type: nauc_map_at_100_max
value: 30.697115186434498
- type: nauc_map_at_100_std
value: 5.694181242130099
- type: nauc_map_at_10_diff1
value: 27.287579843894623
- type: nauc_map_at_10_max
value: 30.416063160658435
- type: nauc_map_at_10_std
value: 5.001223879129747
- type: nauc_map_at_1_diff1
value: 34.766915543377216
- type: nauc_map_at_1_max
value: 31.68917356389495
- type: nauc_map_at_1_std
value: 4.63364234029443
- type: nauc_map_at_20_diff1
value: 26.988320257908132
- type: nauc_map_at_20_max
value: 30.59211599414314
- type: nauc_map_at_20_std
value: 5.5155380180944205
- type: nauc_map_at_3_diff1
value: 29.76134767596865
- type: nauc_map_at_3_max
value: 30.006661638483646
- type: nauc_map_at_3_std
value: 5.008441487743006
- type: nauc_map_at_5_diff1
value: 27.996100405808832
- type: nauc_map_at_5_max
value: 29.688313272752787
- type: nauc_map_at_5_std
value: 4.989016460968753
- type: nauc_mrr_at_1000_diff1
value: 27.026456330287647
- type: nauc_mrr_at_1000_max
value: 32.20560453205575
- type: nauc_mrr_at_1000_std
value: 5.546537805647763
- type: nauc_mrr_at_100_diff1
value: 27.02363656101913
- type: nauc_mrr_at_100_max
value: 32.196640350864875
- type: nauc_mrr_at_100_std
value: 5.518951009854468
- type: nauc_mrr_at_10_diff1
value: 27.00915074052057
- type: nauc_mrr_at_10_max
value: 32.001757674286736
- type: nauc_mrr_at_10_std
value: 5.072731140770552
- type: nauc_mrr_at_1_diff1
value: 34.845574431208064
- type: nauc_mrr_at_1_max
value: 34.49816690982053
- type: nauc_mrr_at_1_std
value: 5.41968335382083
- type: nauc_mrr_at_20_diff1
value: 26.84248184004788
- type: nauc_mrr_at_20_max
value: 32.07749657503134
- type: nauc_mrr_at_20_std
value: 5.441862772046592
- type: nauc_mrr_at_3_diff1
value: 29.505028408787286
- type: nauc_mrr_at_3_max
value: 32.86798316031229
- type: nauc_mrr_at_3_std
value: 4.8203112022331895
- type: nauc_mrr_at_5_diff1
value: 27.569200794858347
- type: nauc_mrr_at_5_max
value: 31.645968122456676
- type: nauc_mrr_at_5_std
value: 4.9761558440226334
- type: nauc_ndcg_at_1000_diff1
value: 23.88608662925394
- type: nauc_ndcg_at_1000_max
value: 31.479699528948128
- type: nauc_ndcg_at_1000_std
value: 7.525533932973386
- type: nauc_ndcg_at_100_diff1
value: 23.604927980525353
- type: nauc_ndcg_at_100_max
value: 31.17191900574252
- type: nauc_ndcg_at_100_std
value: 7.378799400265408
- type: nauc_ndcg_at_10_diff1
value: 23.62553795593263
- type: nauc_ndcg_at_10_max
value: 30.338342936369855
- type: nauc_ndcg_at_10_std
value: 4.953463227590225
- type: nauc_ndcg_at_1_diff1
value: 34.845574431208064
- type: nauc_ndcg_at_1_max
value: 34.49816690982053
- type: nauc_ndcg_at_1_std
value: 5.41968335382083
- type: nauc_ndcg_at_20_diff1
value: 22.70534696949328
- type: nauc_ndcg_at_20_max
value: 30.65105372647873
- type: nauc_ndcg_at_20_std
value: 6.53760429235338
- type: nauc_ndcg_at_3_diff1
value: 28.25657951417217
- type: nauc_ndcg_at_3_max
value: 30.680157197390457
- type: nauc_ndcg_at_3_std
value: 5.055222325853256
- type: nauc_ndcg_at_5_diff1
value: 25.051860390557394
- type: nauc_ndcg_at_5_max
value: 29.321858649131254
- type: nauc_ndcg_at_5_std
value: 5.084862957981524
- type: nauc_precision_at_1000_diff1
value: -13.576500693637458
- type: nauc_precision_at_1000_max
value: 14.89365822136165
- type: nauc_precision_at_1000_std
value: 2.561285129665411
- type: nauc_precision_at_100_diff1
value: -0.5356717684025231
- type: nauc_precision_at_100_max
value: 24.869247406121847
- type: nauc_precision_at_100_std
value: 11.862364821181174
- type: nauc_precision_at_10_diff1
value: 9.30582069803439
- type: nauc_precision_at_10_max
value: 30.923589035628858
- type: nauc_precision_at_10_std
value: 5.398201031301567
- type: nauc_precision_at_1_diff1
value: 34.845574431208064
- type: nauc_precision_at_1_max
value: 34.49816690982053
- type: nauc_precision_at_1_std
value: 5.41968335382083
- type: nauc_precision_at_20_diff1
value: 4.059202632041108
- type: nauc_precision_at_20_max
value: 30.894783817339388
- type: nauc_precision_at_20_std
value: 10.754015672232937
- type: nauc_precision_at_3_diff1
value: 22.522340581277586
- type: nauc_precision_at_3_max
value: 32.5100154377477
- type: nauc_precision_at_3_std
value: 6.522378511380731
- type: nauc_precision_at_5_diff1
value: 14.85331547455788
- type: nauc_precision_at_5_max
value: 30.095138012218207
- type: nauc_precision_at_5_std
value: 5.64736160156279
- type: nauc_recall_at_1000_diff1
value: -5.27332415745275
- type: nauc_recall_at_1000_max
value: 33.16050370515966
- type: nauc_recall_at_1000_std
value: 35.48080153587079
- type: nauc_recall_at_100_diff1
value: 9.401206314932313
- type: nauc_recall_at_100_max
value: 29.15466334221412
- type: nauc_recall_at_100_std
value: 15.925857450796613
- type: nauc_recall_at_10_diff1
value: 13.2795134118332
- type: nauc_recall_at_10_max
value: 27.4224535997482
- type: nauc_recall_at_10_std
value: 4.610601473431608
- type: nauc_recall_at_1_diff1
value: 34.766915543377216
- type: nauc_recall_at_1_max
value: 31.68917356389495
- type: nauc_recall_at_1_std
value: 4.63364234029443
- type: nauc_recall_at_20_diff1
value: 9.041251274906765
- type: nauc_recall_at_20_max
value: 27.833816433039633
- type: nauc_recall_at_20_std
value: 10.004419902367824
- type: nauc_recall_at_3_diff1
value: 24.312698907863194
- type: nauc_recall_at_3_max
value: 28.445506445251507
- type: nauc_recall_at_3_std
value: 4.954049242639007
- type: nauc_recall_at_5_diff1
value: 17.32213286984511
- type: nauc_recall_at_5_max
value: 25.38081324317843
- type: nauc_recall_at_5_std
value: 5.5041998917365875
- type: ndcg_at_1
value: 21.811
- type: ndcg_at_10
value: 33.832
- type: ndcg_at_100
value: 39.675
- type: ndcg_at_1000
value: 42.22
- type: ndcg_at_20
value: 36.246
- type: ndcg_at_3
value: 28.09
- type: ndcg_at_5
value: 31.173000000000002
- type: precision_at_1
value: 21.811
- type: precision_at_10
value: 5.601
- type: precision_at_100
value: 0.9259999999999999
- type: precision_at_1000
value: 0.126
- type: precision_at_20
value: 3.383
- type: precision_at_3
value: 12.261
- type: precision_at_5
value: 9.168
- type: recall_at_1
value: 20.003
- type: recall_at_10
value: 47.754000000000005
- type: recall_at_100
value: 74.575
- type: recall_at_1000
value: 93.118
- type: recall_at_20
value: 56.887
- type: recall_at_3
value: 32.584999999999994
- type: recall_at_5
value: 39.987
- task:
type: Retrieval
dataset:
name: MTEB ClimateFEVER (default)
type: mteb/climate-fever
config: default
split: test
revision: 47f2ac6acb640fc46020b02a5b59fdda04d39380
metrics:
- type: main_score
value: 17.039
- type: map_at_1
value: 5.803
- type: map_at_10
value: 11.043
- type: map_at_100
value: 12.575
- type: map_at_1000
value: 12.777
- type: map_at_20
value: 11.811
- type: map_at_3
value: 8.886
- type: map_at_5
value: 9.918000000000001
- type: mrr_at_1
value: 13.745928338762214
- type: mrr_at_10
value: 23.035856470709884
- type: mrr_at_100
value: 24.31853659598728
- type: mrr_at_1000
value: 24.381741087426335
- type: mrr_at_20
value: 23.826099691586087
- type: mrr_at_3
value: 19.641693811074926
- type: mrr_at_5
value: 21.462540716612363
- type: nauc_map_at_1000_diff1
value: 14.075861001605652
- type: nauc_map_at_1000_max
value: 32.50540299753915
- type: nauc_map_at_1000_std
value: 16.63440932146544
- type: nauc_map_at_100_diff1
value: 14.123217880033861
- type: nauc_map_at_100_max
value: 32.41146114250095
- type: nauc_map_at_100_std
value: 16.381092437163552
- type: nauc_map_at_10_diff1
value: 14.311158236352378
- type: nauc_map_at_10_max
value: 31.663465768164695
- type: nauc_map_at_10_std
value: 14.79004698936343
- type: nauc_map_at_1_diff1
value: 19.79138559884689
- type: nauc_map_at_1_max
value: 28.547089859427256
- type: nauc_map_at_1_std
value: 8.354323364048897
- type: nauc_map_at_20_diff1
value: 14.23157436194119
- type: nauc_map_at_20_max
value: 31.829323482590056
- type: nauc_map_at_20_std
value: 15.728487414739051
- type: nauc_map_at_3_diff1
value: 16.977223379413818
- type: nauc_map_at_3_max
value: 30.91365779223932
- type: nauc_map_at_3_std
value: 11.004013141499629
- type: nauc_map_at_5_diff1
value: 15.287594960641249
- type: nauc_map_at_5_max
value: 31.4855930880175
- type: nauc_map_at_5_std
value: 13.035283381488696
- type: nauc_mrr_at_1000_diff1
value: 15.19673911397727
- type: nauc_mrr_at_1000_max
value: 32.27281360470439
- type: nauc_mrr_at_1000_std
value: 17.982389020371247
- type: nauc_mrr_at_100_diff1
value: 15.20130608016213
- type: nauc_mrr_at_100_max
value: 32.26452236374923
- type: nauc_mrr_at_100_std
value: 18.002716153000627
- type: nauc_mrr_at_10_diff1
value: 15.1395436352143
- type: nauc_mrr_at_10_max
value: 32.2082338876027
- type: nauc_mrr_at_10_std
value: 17.838618343534538
- type: nauc_mrr_at_1_diff1
value: 18.75801367430074
- type: nauc_mrr_at_1_max
value: 29.522146316819242
- type: nauc_mrr_at_1_std
value: 12.322203243653137
- type: nauc_mrr_at_20_diff1
value: 15.15341535844078
- type: nauc_mrr_at_20_max
value: 32.149592234561844
- type: nauc_mrr_at_20_std
value: 18.05739998762671
- type: nauc_mrr_at_3_diff1
value: 16.713713609700807
- type: nauc_mrr_at_3_max
value: 31.073485728698007
- type: nauc_mrr_at_3_std
value: 14.685962809871809
- type: nauc_mrr_at_5_diff1
value: 15.20715918092642
- type: nauc_mrr_at_5_max
value: 31.91496536532501
- type: nauc_mrr_at_5_std
value: 16.32734830238726
- type: nauc_ndcg_at_1000_diff1
value: 11.479850187724962
- type: nauc_ndcg_at_1000_max
value: 36.373357396933336
- type: nauc_ndcg_at_1000_std
value: 25.683568919701514
- type: nauc_ndcg_at_100_diff1
value: 11.559236662789004
- type: nauc_ndcg_at_100_max
value: 34.90591391147814
- type: nauc_ndcg_at_100_std
value: 23.057965705153865
- type: nauc_ndcg_at_10_diff1
value: 12.544873936559739
- type: nauc_ndcg_at_10_max
value: 32.83663980495747
- type: nauc_ndcg_at_10_std
value: 19.238337421544006
- type: nauc_ndcg_at_1_diff1
value: 18.75801367430074
- type: nauc_ndcg_at_1_max
value: 29.522146316819242
- type: nauc_ndcg_at_1_std
value: 12.322203243653137
- type: nauc_ndcg_at_20_diff1
value: 12.361691440408999
- type: nauc_ndcg_at_20_max
value: 32.76528016217301
- type: nauc_ndcg_at_20_std
value: 21.05644041370163
- type: nauc_ndcg_at_3_diff1
value: 15.96832626064785
- type: nauc_ndcg_at_3_max
value: 31.44063980122053
- type: nauc_ndcg_at_3_std
value: 12.936413672431879
- type: nauc_ndcg_at_5_diff1
value: 13.713513523433363
- type: nauc_ndcg_at_5_max
value: 32.36446493796571
- type: nauc_ndcg_at_5_std
value: 15.76174671030862
- type: nauc_precision_at_1000_diff1
value: 1.34184180265124
- type: nauc_precision_at_1000_max
value: 23.682869852931777
- type: nauc_precision_at_1000_std
value: 30.11200034478147
- type: nauc_precision_at_100_diff1
value: 2.7429761602552545
- type: nauc_precision_at_100_max
value: 27.799306591674704
- type: nauc_precision_at_100_std
value: 28.63347129016136
- type: nauc_precision_at_10_diff1
value: 6.809778511951629
- type: nauc_precision_at_10_max
value: 31.549319057696106
- type: nauc_precision_at_10_std
value: 27.797058851811858
- type: nauc_precision_at_1_diff1
value: 18.75801367430074
- type: nauc_precision_at_1_max
value: 29.522146316819242
- type: nauc_precision_at_1_std
value: 12.322203243653137
- type: nauc_precision_at_20_diff1
value: 6.652188021122349
- type: nauc_precision_at_20_max
value: 28.77994242287882
- type: nauc_precision_at_20_std
value: 29.36148470486359
- type: nauc_precision_at_3_diff1
value: 14.712258200325262
- type: nauc_precision_at_3_max
value: 32.0031289884313
- type: nauc_precision_at_3_std
value: 16.234492809427824
- type: nauc_precision_at_5_diff1
value: 9.304224063707855
- type: nauc_precision_at_5_max
value: 33.271254238986785
- type: nauc_precision_at_5_std
value: 21.406431386266583
- type: nauc_recall_at_1000_diff1
value: 0.25017014524490533
- type: nauc_recall_at_1000_max
value: 38.04838759534934
- type: nauc_recall_at_1000_std
value: 40.585116054127084
- type: nauc_recall_at_100_diff1
value: 3.0042658139447593
- type: nauc_recall_at_100_max
value: 30.765498421078746
- type: nauc_recall_at_100_std
value: 26.344845670411445
- type: nauc_recall_at_10_diff1
value: 7.066884715859835
- type: nauc_recall_at_10_max
value: 30.432215207137187
- type: nauc_recall_at_10_std
value: 21.88096446786714
- type: nauc_recall_at_1_diff1
value: 19.79138559884689
- type: nauc_recall_at_1_max
value: 28.547089859427256
- type: nauc_recall_at_1_std
value: 8.354323364048897
- type: nauc_recall_at_20_diff1
value: 6.260902435695847
- type: nauc_recall_at_20_max
value: 28.085979713597105
- type: nauc_recall_at_20_std
value: 24.042661661305686
- type: nauc_recall_at_3_diff1
value: 13.85539993127579
- type: nauc_recall_at_3_max
value: 30.66048968710386
- type: nauc_recall_at_3_std
value: 12.49456367200617
- type: nauc_recall_at_5_diff1
value: 9.732734054264856
- type: nauc_recall_at_5_max
value: 31.558450901307765
- type: nauc_recall_at_5_std
value: 16.85189341873968
- type: ndcg_at_1
value: 13.746
- type: ndcg_at_10
value: 17.039
- type: ndcg_at_100
value: 24.251
- type: ndcg_at_1000
value: 28.407
- type: ndcg_at_20
value: 19.596
- type: ndcg_at_3
value: 12.825000000000001
- type: ndcg_at_5
value: 14.246
- type: precision_at_1
value: 13.746
- type: precision_at_10
value: 5.772
- type: precision_at_100
value: 1.345
- type: precision_at_1000
value: 0.211
- type: precision_at_20
value: 3.945
- type: precision_at_3
value: 9.815
- type: precision_at_5
value: 7.9479999999999995
- type: recall_at_1
value: 5.803
- type: recall_at_10
value: 22.320999999999998
- type: recall_at_100
value: 48.083
- type: recall_at_1000
value: 72.02199999999999
- type: recall_at_20
value: 29.693
- type: recall_at_3
value: 12.249
- type: recall_at_5
value: 15.925
- task:
type: Retrieval
dataset:
name: MTEB DBPedia (default)
type: mteb/dbpedia
config: default
split: dev
revision: c0f706b76e590d620bd6618b3ca8efdd34e2d659
metrics:
- type: main_score
value: 32.056000000000004
- type: map_at_1
value: 6.159
- type: map_at_10
value: 14.749
- type: map_at_100
value: 19.639
- type: map_at_1000
value: 20.902
- type: map_at_20
value: 16.320999999999998
- type: map_at_3
value: 10.598
- type: map_at_5
value: 12.232
- type: mrr_at_1
value: 55.223880597014926
- type: mrr_at_10
value: 62.61194029850745
- type: mrr_at_100
value: 63.161403712876506
- type: mrr_at_1000
value: 63.19045792845882
- type: mrr_at_20
value: 62.883310719131615
- type: mrr_at_3
value: 60.69651741293532
- type: mrr_at_5
value: 62.11442786069652
- type: nauc_map_at_1000_diff1
value: 14.66774420808803
- type: nauc_map_at_1000_max
value: 15.83673037882429
- type: nauc_map_at_1000_std
value: 5.780706425119277
- type: nauc_map_at_100_diff1
value: 15.155295918719005
- type: nauc_map_at_100_max
value: 14.072328538517597
- type: nauc_map_at_100_std
value: 1.770766267850918
- type: nauc_map_at_10_diff1
value: 14.98606047237713
- type: nauc_map_at_10_max
value: 6.187697786036102
- type: nauc_map_at_10_std
value: -6.360564380948436
- type: nauc_map_at_1_diff1
value: 28.930347534146826
- type: nauc_map_at_1_max
value: -22.946473489514478
- type: nauc_map_at_1_std
value: -5.478885357489002
- type: nauc_map_at_20_diff1
value: 13.225096470295508
- type: nauc_map_at_20_max
value: 8.134020886577796
- type: nauc_map_at_20_std
value: -3.688087435394214
- type: nauc_map_at_3_diff1
value: 23.967211717665325
- type: nauc_map_at_3_max
value: -8.54757081540758
- type: nauc_map_at_3_std
value: -11.854904684666472
- type: nauc_map_at_5_diff1
value: 18.69169934915187
- type: nauc_map_at_5_max
value: -2.681982078444763
- type: nauc_map_at_5_std
value: -9.677458332091078
- type: nauc_mrr_at_1000_diff1
value: 29.1326144948668
- type: nauc_mrr_at_1000_max
value: 30.32888792364254
- type: nauc_mrr_at_1000_std
value: 0.10599054230879215
- type: nauc_mrr_at_100_diff1
value: 29.170245504786546
- type: nauc_mrr_at_100_max
value: 30.334485258327394
- type: nauc_mrr_at_100_std
value: 0.073017997841517
- type: nauc_mrr_at_10_diff1
value: 28.61680486143009
- type: nauc_mrr_at_10_max
value: 29.886786779812883
- type: nauc_mrr_at_10_std
value: -0.7804123870159443
- type: nauc_mrr_at_1_diff1
value: 28.445954358025144
- type: nauc_mrr_at_1_max
value: 29.757901190323032
- type: nauc_mrr_at_1_std
value: -8.160393754836772
- type: nauc_mrr_at_20_diff1
value: 28.970583921994
- type: nauc_mrr_at_20_max
value: 30.297978189519036
- type: nauc_mrr_at_20_std
value: 0.025099380605112777
- type: nauc_mrr_at_3_diff1
value: 26.922893811846027
- type: nauc_mrr_at_3_max
value: 28.454141443011615
- type: nauc_mrr_at_3_std
value: 0.5404188302638873
- type: nauc_mrr_at_5_diff1
value: 28.260050983821404
- type: nauc_mrr_at_5_max
value: 30.662788723961064
- type: nauc_mrr_at_5_std
value: 0.25695689627892465
- type: nauc_ndcg_at_1000_diff1
value: 17.550822161705184
- type: nauc_ndcg_at_1000_max
value: 23.47112652887515
- type: nauc_ndcg_at_1000_std
value: 18.921600629635876
- type: nauc_ndcg_at_100_diff1
value: 19.75836373321895
- type: nauc_ndcg_at_100_max
value: 17.218863090775898
- type: nauc_ndcg_at_100_std
value: 6.188418551354186
- type: nauc_ndcg_at_10_diff1
value: 17.018486303230613
- type: nauc_ndcg_at_10_max
value: 13.172847759812633
- type: nauc_ndcg_at_10_std
value: -2.557915936016456
- type: nauc_ndcg_at_1_diff1
value: 28.30450276152453
- type: nauc_ndcg_at_1_max
value: 10.705906316378517
- type: nauc_ndcg_at_1_std
value: -11.557649340196585
- type: nauc_ndcg_at_20_diff1
value: 17.248069262429212
- type: nauc_ndcg_at_20_max
value: 9.849529486503714
- type: nauc_ndcg_at_20_std
value: -0.311186473192
- type: nauc_ndcg_at_3_diff1
value: 24.91723384151359
- type: nauc_ndcg_at_3_max
value: 12.746501741669613
- type: nauc_ndcg_at_3_std
value: -7.478244755823778
- type: nauc_ndcg_at_5_diff1
value: 20.109109265849096
- type: nauc_ndcg_at_5_max
value: 15.035625561879778
- type: nauc_ndcg_at_5_std
value: -4.249792039203992
- type: nauc_precision_at_1000_diff1
value: -1.6904232746143093
- type: nauc_precision_at_1000_max
value: 28.193351306594632
- type: nauc_precision_at_1000_std
value: 49.466555012944426
- type: nauc_precision_at_100_diff1
value: 5.522098718336858
- type: nauc_precision_at_100_max
value: 32.28843828110046
- type: nauc_precision_at_100_std
value: 38.855283968708925
- type: nauc_precision_at_10_diff1
value: 0.8754832613076982
- type: nauc_precision_at_10_max
value: 33.667753747656214
- type: nauc_precision_at_10_std
value: 11.811992543860297
- type: nauc_precision_at_1_diff1
value: 28.445954358025144
- type: nauc_precision_at_1_max
value: 29.757901190323032
- type: nauc_precision_at_1_std
value: -8.160393754836772
- type: nauc_precision_at_20_diff1
value: 3.06814758548795
- type: nauc_precision_at_20_max
value: 34.56674755156743
- type: nauc_precision_at_20_std
value: 20.6096728873244
- type: nauc_precision_at_3_diff1
value: 14.003111174047799
- type: nauc_precision_at_3_max
value: 26.627875058653572
- type: nauc_precision_at_3_std
value: -1.576904767201546
- type: nauc_precision_at_5_diff1
value: 4.2565090179589
- type: nauc_precision_at_5_max
value: 30.09262969886471
- type: nauc_precision_at_5_std
value: 3.37613085747931
- type: nauc_recall_at_1000_diff1
value: 1.318125788778631
- type: nauc_recall_at_1000_max
value: 23.347027051136532
- type: nauc_recall_at_1000_std
value: 38.6734430950855
- type: nauc_recall_at_100_diff1
value: 16.260906677583385
- type: nauc_recall_at_100_max
value: 13.269724794813417
- type: nauc_recall_at_100_std
value: 12.528814958715534
- type: nauc_recall_at_10_diff1
value: 12.703144510602954
- type: nauc_recall_at_10_max
value: 11.643752381855938
- type: nauc_recall_at_10_std
value: -4.93246361539532
- type: nauc_recall_at_1_diff1
value: 28.930347534146826
- type: nauc_recall_at_1_max
value: -22.946473489514478
- type: nauc_recall_at_1_std
value: -5.478885357489002
- type: nauc_recall_at_20_diff1
value: 8.197159305821131
- type: nauc_recall_at_20_max
value: 7.992590649067159
- type: nauc_recall_at_20_std
value: 2.9499995677138213
- type: nauc_recall_at_3_diff1
value: 24.578598477208853
- type: nauc_recall_at_3_max
value: -5.236092678096414
- type: nauc_recall_at_3_std
value: -11.675071458967711
- type: nauc_recall_at_5_diff1
value: 16.448098064378637
- type: nauc_recall_at_5_max
value: 0.5330990419735525
- type: nauc_recall_at_5_std
value: -8.139497416808533
- type: ndcg_at_1
value: 44.03
- type: ndcg_at_10
value: 32.056000000000004
- type: ndcg_at_100
value: 36.813
- type: ndcg_at_1000
value: 44.603
- type: ndcg_at_20
value: 30.522
- type: ndcg_at_3
value: 37.112
- type: ndcg_at_5
value: 34.353
- type: precision_at_1
value: 55.224
- type: precision_at_10
value: 24.776
- type: precision_at_100
value: 7.5520000000000005
- type: precision_at_1000
value: 1.361
- type: precision_at_20
value: 16.418
- type: precision_at_3
value: 39.303
- type: precision_at_5
value: 32.537
- type: recall_at_1
value: 6.159
- type: recall_at_10
value: 18.767
- type: recall_at_100
value: 42.281
- type: recall_at_1000
value: 69.05799999999999
- type: recall_at_20
value: 24.176000000000002
- type: recall_at_3
value: 11.350999999999999
- type: recall_at_5
value: 13.858
- task:
type: Retrieval
dataset:
name: MTEB DBPedia (default)
type: mteb/dbpedia
config: default
split: test
revision: c0f706b76e590d620bd6618b3ca8efdd34e2d659
metrics:
- type: main_score
value: 26.912000000000003
- type: map_at_1
value: 5.587000000000001
- type: map_at_10
value: 11.895
- type: map_at_100
value: 16.598
- type: map_at_1000
value: 17.812
- type: map_at_20
value: 13.679
- type: map_at_3
value: 8.475000000000001
- type: map_at_5
value: 9.864
- type: mrr_at_1
value: 43.5
- type: mrr_at_10
value: 55.057341269841274
- type: mrr_at_100
value: 55.7116763249525
- type: mrr_at_1000
value: 55.73114654313282
- type: mrr_at_20
value: 55.48448584654082
- type: mrr_at_3
value: 52.625000000000014
- type: mrr_at_5
value: 54.25
- type: nauc_map_at_1000_diff1
value: 9.222585526729162
- type: nauc_map_at_1000_max
value: 6.474554295414542
- type: nauc_map_at_1000_std
value: 19.530457057036873
- type: nauc_map_at_100_diff1
value: 9.11589434330461
- type: nauc_map_at_100_max
value: 3.0466163667963415
- type: nauc_map_at_100_std
value: 15.897958511594085
- type: nauc_map_at_10_diff1
value: 14.033788134812697
- type: nauc_map_at_10_max
value: -4.180301152674868
- type: nauc_map_at_10_std
value: 0.9565463929832962
- type: nauc_map_at_1_diff1
value: 24.88372785558863
- type: nauc_map_at_1_max
value: -7.903205339084808
- type: nauc_map_at_1_std
value: -9.025106405036246
- type: nauc_map_at_20_diff1
value: 10.982775028585708
- type: nauc_map_at_20_max
value: -1.9814898263360956
- type: nauc_map_at_20_std
value: 5.022744754836501
- type: nauc_map_at_3_diff1
value: 18.920355286487474
- type: nauc_map_at_3_max
value: -7.008953507731825
- type: nauc_map_at_3_std
value: -7.4302507983255035
- type: nauc_map_at_5_diff1
value: 15.649741124593813
- type: nauc_map_at_5_max
value: -6.470264312561451
- type: nauc_map_at_5_std
value: -3.629324410314669
- type: nauc_mrr_at_1000_diff1
value: 27.445673721571445
- type: nauc_mrr_at_1000_max
value: 30.848234089089377
- type: nauc_mrr_at_1000_std
value: 25.584068294106398
- type: nauc_mrr_at_100_diff1
value: 27.450436881700647
- type: nauc_mrr_at_100_max
value: 30.86805059900003
- type: nauc_mrr_at_100_std
value: 25.602642209980893
- type: nauc_mrr_at_10_diff1
value: 27.03642213567309
- type: nauc_mrr_at_10_max
value: 30.564766924576997
- type: nauc_mrr_at_10_std
value: 25.498766017362584
- type: nauc_mrr_at_1_diff1
value: 29.75429751366281
- type: nauc_mrr_at_1_max
value: 27.603423680456483
- type: nauc_mrr_at_1_std
value: 21.42210410437796
- type: nauc_mrr_at_20_diff1
value: 27.316158731916072
- type: nauc_mrr_at_20_max
value: 30.788911987885836
- type: nauc_mrr_at_20_std
value: 25.511317895342973
- type: nauc_mrr_at_3_diff1
value: 27.40278716897145
- type: nauc_mrr_at_3_max
value: 30.17795110537535
- type: nauc_mrr_at_3_std
value: 25.58949724207859
- type: nauc_mrr_at_5_diff1
value: 27.02036667526836
- type: nauc_mrr_at_5_max
value: 30.56172819314359
- type: nauc_mrr_at_5_std
value: 25.68106181481602
- type: nauc_ndcg_at_1000_diff1
value: 10.185469519827409
- type: nauc_ndcg_at_1000_max
value: 19.97222766004627
- type: nauc_ndcg_at_1000_std
value: 33.627761835495576
- type: nauc_ndcg_at_100_diff1
value: 10.34566025576966
- type: nauc_ndcg_at_100_max
value: 10.904699675777103
- type: nauc_ndcg_at_100_std
value: 25.902939752688408
- type: nauc_ndcg_at_10_diff1
value: 9.763184508901789
- type: nauc_ndcg_at_10_max
value: 15.84272775012148
- type: nauc_ndcg_at_10_std
value: 23.3644718110982
- type: nauc_ndcg_at_1_diff1
value: 27.53213753471065
- type: nauc_ndcg_at_1_max
value: 25.47433464492357
- type: nauc_ndcg_at_1_std
value: 22.927455121813495
- type: nauc_ndcg_at_20_diff1
value: 9.21982112040321
- type: nauc_ndcg_at_20_max
value: 10.855015325947646
- type: nauc_ndcg_at_20_std
value: 20.177684268239858
- type: nauc_ndcg_at_3_diff1
value: 16.480487570544028
- type: nauc_ndcg_at_3_max
value: 23.08069150528697
- type: nauc_ndcg_at_3_std
value: 24.654582040091068
- type: nauc_ndcg_at_5_diff1
value: 11.091173429430855
- type: nauc_ndcg_at_5_max
value: 18.943964454340282
- type: nauc_ndcg_at_5_std
value: 25.417120083736062
- type: nauc_precision_at_1000_diff1
value: 6.551654038336142
- type: nauc_precision_at_1000_max
value: 49.44788048053744
- type: nauc_precision_at_1000_std
value: 37.149790102492084
- type: nauc_precision_at_100_diff1
value: -3.0445732987998726
- type: nauc_precision_at_100_max
value: 35.41813437751633
- type: nauc_precision_at_100_std
value: 49.84817605241621
- type: nauc_precision_at_10_diff1
value: -1.9547816065045103
- type: nauc_precision_at_10_max
value: 28.351986746022433
- type: nauc_precision_at_10_std
value: 39.87641283361654
- type: nauc_precision_at_1_diff1
value: 29.75429751366281
- type: nauc_precision_at_1_max
value: 27.603423680456483
- type: nauc_precision_at_1_std
value: 21.42210410437796
- type: nauc_precision_at_20_diff1
value: -5.163004702977454
- type: nauc_precision_at_20_max
value: 29.15244120857218
- type: nauc_precision_at_20_std
value: 42.54647731285851
- type: nauc_precision_at_3_diff1
value: 8.549385971521142
- type: nauc_precision_at_3_max
value: 28.050026369321174
- type: nauc_precision_at_3_std
value: 31.253522187900256
- type: nauc_precision_at_5_diff1
value: -0.5560619777844419
- type: nauc_precision_at_5_max
value: 25.400998256732244
- type: nauc_precision_at_5_std
value: 36.987247493209956
- type: nauc_recall_at_1000_diff1
value: 6.3187672735979135
- type: nauc_recall_at_1000_max
value: 11.452062079279093
- type: nauc_recall_at_1000_std
value: 34.01011092060735
- type: nauc_recall_at_100_diff1
value: 10.256765153257081
- type: nauc_recall_at_100_max
value: 1.3948375294116233
- type: nauc_recall_at_100_std
value: 23.12466567545472
- type: nauc_recall_at_10_diff1
value: 9.429415669158093
- type: nauc_recall_at_10_max
value: -7.19845910730243
- type: nauc_recall_at_10_std
value: -0.7518308015760871
- type: nauc_recall_at_1_diff1
value: 24.88372785558863
- type: nauc_recall_at_1_max
value: -7.903205339084808
- type: nauc_recall_at_1_std
value: -9.025106405036246
- type: nauc_recall_at_20_diff1
value: 7.221922012299395
- type: nauc_recall_at_20_max
value: -5.373839522647143
- type: nauc_recall_at_20_std
value: 3.361400375981407
- type: nauc_recall_at_3_diff1
value: 15.217973070806593
- type: nauc_recall_at_3_max
value: -8.255546494136766
- type: nauc_recall_at_3_std
value: -8.0508215378933
- type: nauc_recall_at_5_diff1
value: 8.750499393201636
- type: nauc_recall_at_5_max
value: -9.456361840780644
- type: nauc_recall_at_5_std
value: -3.255892902113352
- type: ndcg_at_1
value: 32.25
- type: ndcg_at_10
value: 26.912000000000003
- type: ndcg_at_100
value: 31.477
- type: ndcg_at_1000
value: 38.78
- type: ndcg_at_20
value: 26.851999999999997
- type: ndcg_at_3
value: 29.343000000000004
- type: ndcg_at_5
value: 27.916
- type: precision_at_1
value: 43.5
- type: precision_at_10
value: 22.25
- type: precision_at_100
value: 7.3580000000000005
- type: precision_at_1000
value: 1.6549999999999998
- type: precision_at_20
value: 16.712
- type: precision_at_3
value: 33.667
- type: precision_at_5
value: 28.65
- type: recall_at_1
value: 5.587000000000001
- type: recall_at_10
value: 17.302999999999997
- type: recall_at_100
value: 39.731
- type: recall_at_1000
value: 63.751999999999995
- type: recall_at_20
value: 22.978
- type: recall_at_3
value: 9.884
- type: recall_at_5
value: 12.717999999999998
- task:
type: Retrieval
dataset:
name: MTEB FEVER (default)
type: mteb/fever
config: default
split: dev
revision: bea83ef9e8fb933d90a2f1d5515737465d613e12
metrics:
- type: main_score
value: 42.311
- type: map_at_1
value: 26.230999999999998
- type: map_at_10
value: 36.488
- type: map_at_100
value: 37.189
- type: map_at_1000
value: 37.238
- type: map_at_20
value: 36.908
- type: map_at_3
value: 33.711999999999996
- type: map_at_5
value: 35.363
- type: mrr_at_1
value: 28.5028502850285
- type: mrr_at_10
value: 39.190341653212904
- type: mrr_at_100
value: 39.873337606890374
- type: mrr_at_1000
value: 39.915366095710894
- type: mrr_at_20
value: 39.61445401700697
- type: mrr_at_3
value: 36.438643864386435
- type: mrr_at_5
value: 38.06705670567057
- type: nauc_map_at_1000_diff1
value: 38.544927343322755
- type: nauc_map_at_1000_max
value: 31.238008214807472
- type: nauc_map_at_1000_std
value: 17.340710731800087
- type: nauc_map_at_100_diff1
value: 38.53271731505664
- type: nauc_map_at_100_max
value: 31.24444002509582
- type: nauc_map_at_100_std
value: 17.336497143108794
- type: nauc_map_at_10_diff1
value: 38.47149376137689
- type: nauc_map_at_10_max
value: 31.057943306774234
- type: nauc_map_at_10_std
value: 16.88320995134788
- type: nauc_map_at_1_diff1
value: 42.64117606936473
- type: nauc_map_at_1_max
value: 25.235300368623836
- type: nauc_map_at_1_std
value: 11.099092252838478
- type: nauc_map_at_20_diff1
value: 38.53909982615329
- type: nauc_map_at_20_max
value: 31.25656920271077
- type: nauc_map_at_20_std
value: 17.27576321281346
- type: nauc_map_at_3_diff1
value: 38.70482758910707
- type: nauc_map_at_3_max
value: 29.343329165290143
- type: nauc_map_at_3_std
value: 14.573356497682216
- type: nauc_map_at_5_diff1
value: 38.45317210617443
- type: nauc_map_at_5_max
value: 30.410656735518977
- type: nauc_map_at_5_std
value: 16.04759185540847
- type: nauc_mrr_at_1000_diff1
value: 39.53064995242943
- type: nauc_mrr_at_1000_max
value: 32.077343336360904
- type: nauc_mrr_at_1000_std
value: 17.039796809778124
- type: nauc_mrr_at_100_diff1
value: 39.51767279358483
- type: nauc_mrr_at_100_max
value: 32.09214820002331
- type: nauc_mrr_at_100_std
value: 17.04820908126493
- type: nauc_mrr_at_10_diff1
value: 39.473737530652535
- type: nauc_mrr_at_10_max
value: 31.95485973713632
- type: nauc_mrr_at_10_std
value: 16.682344196814157
- type: nauc_mrr_at_1_diff1
value: 43.54209193151506
- type: nauc_mrr_at_1_max
value: 26.225221030688388
- type: nauc_mrr_at_1_std
value: 11.045465765362609
- type: nauc_mrr_at_20_diff1
value: 39.51856222703832
- type: nauc_mrr_at_20_max
value: 32.15641740823006
- type: nauc_mrr_at_20_std
value: 17.052323554211206
- type: nauc_mrr_at_3_diff1
value: 39.59110286581465
- type: nauc_mrr_at_3_max
value: 30.27336556208048
- type: nauc_mrr_at_3_std
value: 14.400464395275131
- type: nauc_mrr_at_5_diff1
value: 39.37475030822441
- type: nauc_mrr_at_5_max
value: 31.280543943287686
- type: nauc_mrr_at_5_std
value: 15.893234066653813
- type: nauc_ndcg_at_1000_diff1
value: 37.92867727075456
- type: nauc_ndcg_at_1000_max
value: 34.646736289585874
- type: nauc_ndcg_at_1000_std
value: 22.109283863938227
- type: nauc_ndcg_at_100_diff1
value: 37.598490502476295
- type: nauc_ndcg_at_100_max
value: 34.99139335810959
- type: nauc_ndcg_at_100_std
value: 22.209322757781297
- type: nauc_ndcg_at_10_diff1
value: 37.46143110942927
- type: nauc_ndcg_at_10_max
value: 34.384238800709674
- type: nauc_ndcg_at_10_std
value: 20.383762100255804
- type: nauc_ndcg_at_1_diff1
value: 43.54209193151506
- type: nauc_ndcg_at_1_max
value: 26.225221030688388
- type: nauc_ndcg_at_1_std
value: 11.045465765362609
- type: nauc_ndcg_at_20_diff1
value: 37.629804811157065
- type: nauc_ndcg_at_20_max
value: 35.13056902686399
- type: nauc_ndcg_at_20_std
value: 21.8348192764161
- type: nauc_ndcg_at_3_diff1
value: 37.80901242867197
- type: nauc_ndcg_at_3_max
value: 30.93143006645779
- type: nauc_ndcg_at_3_std
value: 15.69317657679832
- type: nauc_ndcg_at_5_diff1
value: 37.36938303664242
- type: nauc_ndcg_at_5_max
value: 32.78308787409404
- type: nauc_ndcg_at_5_std
value: 18.37346652477358
- type: nauc_precision_at_1000_diff1
value: 16.527943940474014
- type: nauc_precision_at_1000_max
value: 36.661745199874645
- type: nauc_precision_at_1000_std
value: 40.759235242152194
- type: nauc_precision_at_100_diff1
value: 24.54407163693533
- type: nauc_precision_at_100_max
value: 45.60883532038694
- type: nauc_precision_at_100_std
value: 42.91332524207595
- type: nauc_precision_at_10_diff1
value: 31.72535525745312
- type: nauc_precision_at_10_max
value: 45.349381448046586
- type: nauc_precision_at_10_std
value: 32.974173826381055
- type: nauc_precision_at_1_diff1
value: 43.54209193151506
- type: nauc_precision_at_1_max
value: 26.225221030688388
- type: nauc_precision_at_1_std
value: 11.045465765362609
- type: nauc_precision_at_20_diff1
value: 30.374414778965203
- type: nauc_precision_at_20_max
value: 48.38235848820685
- type: nauc_precision_at_20_std
value: 39.73784986913937
- type: nauc_precision_at_3_diff1
value: 34.95600573048938
- type: nauc_precision_at_3_max
value: 35.80713704633644
- type: nauc_precision_at_3_std
value: 19.412990960503265
- type: nauc_precision_at_5_diff1
value: 33.12845327196843
- type: nauc_precision_at_5_max
value: 40.27118733247272
- type: nauc_precision_at_5_std
value: 25.943558127871956
- type: nauc_recall_at_1000_diff1
value: 29.768973516222186
- type: nauc_recall_at_1000_max
value: 45.76807061498038
- type: nauc_recall_at_1000_std
value: 51.113086504789116
- type: nauc_recall_at_100_diff1
value: 29.22558023832464
- type: nauc_recall_at_100_max
value: 46.15003117187778
- type: nauc_recall_at_100_std
value: 42.80642812167553
- type: nauc_recall_at_10_diff1
value: 31.629369166217437
- type: nauc_recall_at_10_max
value: 42.076418890799836
- type: nauc_recall_at_10_std
value: 30.59336185155107
- type: nauc_recall_at_1_diff1
value: 42.64117606936473
- type: nauc_recall_at_1_max
value: 25.235300368623836
- type: nauc_recall_at_1_std
value: 11.099092252838478
- type: nauc_recall_at_20_diff1
value: 31.335469263874455
- type: nauc_recall_at_20_max
value: 45.181873226895455
- type: nauc_recall_at_20_std
value: 37.03396315996297
- type: nauc_recall_at_3_diff1
value: 33.30861190274494
- type: nauc_recall_at_3_max
value: 33.17199752470918
- type: nauc_recall_at_3_std
value: 18.300476180248737
- type: nauc_recall_at_5_diff1
value: 31.99541739955995
- type: nauc_recall_at_5_max
value: 37.327841772897685
- type: nauc_recall_at_5_std
value: 24.427296500445365
- type: ndcg_at_1
value: 28.503
- type: ndcg_at_10
value: 42.311
- type: ndcg_at_100
value: 45.739999999999995
- type: ndcg_at_1000
value: 47.116
- type: ndcg_at_20
value: 43.814
- type: ndcg_at_3
value: 36.844
- type: ndcg_at_5
value: 39.696999999999996
- type: precision_at_1
value: 28.503
- type: precision_at_10
value: 6.367000000000001
- type: precision_at_100
value: 0.822
- type: precision_at_1000
value: 0.096
- type: precision_at_20
value: 3.5159999999999996
- type: precision_at_3
value: 15.697
- type: precision_at_5
value: 10.975
- type: recall_at_1
value: 26.230999999999998
- type: recall_at_10
value: 57.830000000000005
- type: recall_at_100
value: 73.529
- type: recall_at_1000
value: 84.21
- type: recall_at_20
value: 63.580000000000005
- type: recall_at_3
value: 43.068
- type: recall_at_5
value: 49.99
- task:
type: Retrieval
dataset:
name: MTEB FEVER (default)
type: mteb/fever
config: default
split: test
revision: bea83ef9e8fb933d90a2f1d5515737465d613e12
metrics:
- type: main_score
value: 42.563
- type: map_at_1
value: 25.789
- type: map_at_10
value: 36.523
- type: map_at_100
value: 37.347
- type: map_at_1000
value: 37.393
- type: map_at_20
value: 37.027
- type: map_at_3
value: 33.483000000000004
- type: map_at_5
value: 35.412
- type: mrr_at_1
value: 27.752775277527753
- type: mrr_at_10
value: 38.92034441539394
- type: mrr_at_100
value: 39.701308768272966
- type: mrr_at_1000
value: 39.74002237607331
- type: mrr_at_20
value: 39.40306190517976
- type: mrr_at_3
value: 35.848584858485836
- type: mrr_at_5
value: 37.803280328032784
- type: nauc_map_at_1000_diff1
value: 34.17597190862309
- type: nauc_map_at_1000_max
value: 26.9900559130689
- type: nauc_map_at_1000_std
value: 12.695854413524419
- type: nauc_map_at_100_diff1
value: 34.159140742195376
- type: nauc_map_at_100_max
value: 26.991827854535916
- type: nauc_map_at_100_std
value: 12.700812019288938
- type: nauc_map_at_10_diff1
value: 34.08880322400212
- type: nauc_map_at_10_max
value: 26.686385238773536
- type: nauc_map_at_10_std
value: 12.106693100422692
- type: nauc_map_at_1_diff1
value: 39.75327570681995
- type: nauc_map_at_1_max
value: 22.516865174086742
- type: nauc_map_at_1_std
value: 7.352429770604897
- type: nauc_map_at_20_diff1
value: 34.09956129529087
- type: nauc_map_at_20_max
value: 26.88383408046557
- type: nauc_map_at_20_std
value: 12.522543335179536
- type: nauc_map_at_3_diff1
value: 34.4408475586939
- type: nauc_map_at_3_max
value: 25.32901294015423
- type: nauc_map_at_3_std
value: 10.261012213574276
- type: nauc_map_at_5_diff1
value: 33.839452205385015
- type: nauc_map_at_5_max
value: 26.162139149396758
- type: nauc_map_at_5_std
value: 11.319738955299071
- type: nauc_mrr_at_1000_diff1
value: 35.35911412311634
- type: nauc_mrr_at_1000_max
value: 27.663425924474506
- type: nauc_mrr_at_1000_std
value: 12.242914517959687
- type: nauc_mrr_at_100_diff1
value: 35.34838926395416
- type: nauc_mrr_at_100_max
value: 27.67860094508867
- type: nauc_mrr_at_100_std
value: 12.261677171728074
- type: nauc_mrr_at_10_diff1
value: 35.27102485638653
- type: nauc_mrr_at_10_max
value: 27.477689977850154
- type: nauc_mrr_at_10_std
value: 11.80499800891837
- type: nauc_mrr_at_1_diff1
value: 40.97638577245565
- type: nauc_mrr_at_1_max
value: 23.585524680635643
- type: nauc_mrr_at_1_std
value: 7.242935375380964
- type: nauc_mrr_at_20_diff1
value: 35.29506971612031
- type: nauc_mrr_at_20_max
value: 27.627615743060264
- type: nauc_mrr_at_20_std
value: 12.159260194079204
- type: nauc_mrr_at_3_diff1
value: 35.51192948129344
- type: nauc_mrr_at_3_max
value: 26.1168315974585
- type: nauc_mrr_at_3_std
value: 9.973253779504573
- type: nauc_mrr_at_5_diff1
value: 35.00372193705226
- type: nauc_mrr_at_5_max
value: 26.927656142079055
- type: nauc_mrr_at_5_std
value: 11.02304870858424
- type: nauc_ndcg_at_1000_diff1
value: 33.49991969615717
- type: nauc_ndcg_at_1000_max
value: 30.209587350888466
- type: nauc_ndcg_at_1000_std
value: 17.3604013033284
- type: nauc_ndcg_at_100_diff1
value: 33.09545326289088
- type: nauc_ndcg_at_100_max
value: 30.41263075949266
- type: nauc_ndcg_at_100_std
value: 17.74146555544644
- type: nauc_ndcg_at_10_diff1
value: 32.68974571152804
- type: nauc_ndcg_at_10_max
value: 29.01307655615002
- type: nauc_ndcg_at_10_std
value: 14.992165745891109
- type: nauc_ndcg_at_1_diff1
value: 40.97638577245565
- type: nauc_ndcg_at_1_max
value: 23.585524680635643
- type: nauc_ndcg_at_1_std
value: 7.242935375380964
- type: nauc_ndcg_at_20_diff1
value: 32.70131432794314
- type: nauc_ndcg_at_20_max
value: 29.726619193431663
- type: nauc_ndcg_at_20_std
value: 16.531701001153703
- type: nauc_ndcg_at_3_diff1
value: 33.2569305109641
- type: nauc_ndcg_at_3_max
value: 26.29321481604789
- type: nauc_ndcg_at_3_std
value: 11.081977028754668
- type: nauc_ndcg_at_5_diff1
value: 32.15552189165288
- type: nauc_ndcg_at_5_max
value: 27.759460485679003
- type: nauc_ndcg_at_5_std
value: 13.015694866625852
- type: nauc_precision_at_1000_diff1
value: 16.812073179287424
- type: nauc_precision_at_1000_max
value: 37.540985211601225
- type: nauc_precision_at_1000_std
value: 36.35910579672834
- type: nauc_precision_at_100_diff1
value: 21.3347768230779
- type: nauc_precision_at_100_max
value: 44.04414745027497
- type: nauc_precision_at_100_std
value: 41.39009594338739
- type: nauc_precision_at_10_diff1
value: 26.47938109887204
- type: nauc_precision_at_10_max
value: 37.19520608029416
- type: nauc_precision_at_10_std
value: 24.75031842784357
- type: nauc_precision_at_1_diff1
value: 40.97638577245565
- type: nauc_precision_at_1_max
value: 23.585524680635643
- type: nauc_precision_at_1_std
value: 7.242935375380964
- type: nauc_precision_at_20_diff1
value: 24.4933069920691
- type: nauc_precision_at_20_max
value: 40.181333968312195
- type: nauc_precision_at_20_std
value: 31.639505760980857
- type: nauc_precision_at_3_diff1
value: 29.55648660225873
- type: nauc_precision_at_3_max
value: 29.42746113719643
- type: nauc_precision_at_3_std
value: 13.134772598063801
- type: nauc_precision_at_5_diff1
value: 26.323829221173785
- type: nauc_precision_at_5_max
value: 32.98308960955934
- type: nauc_precision_at_5_std
value: 17.999660418650226
- type: nauc_recall_at_1000_diff1
value: 27.09758394920151
- type: nauc_recall_at_1000_max
value: 49.014623321157934
- type: nauc_recall_at_1000_std
value: 54.70002479259918
- type: nauc_recall_at_100_diff1
value: 24.828096194405543
- type: nauc_recall_at_100_max
value: 44.57384134801808
- type: nauc_recall_at_100_std
value: 45.47648726921974
- type: nauc_recall_at_10_diff1
value: 25.91218654921606
- type: nauc_recall_at_10_max
value: 34.13630206088126
- type: nauc_recall_at_10_std
value: 24.013987861347292
- type: nauc_recall_at_1_diff1
value: 39.75327570681995
- type: nauc_recall_at_1_max
value: 22.516865174086742
- type: nauc_recall_at_1_std
value: 7.352429770604897
- type: nauc_recall_at_20_diff1
value: 24.96068254117673
- type: nauc_recall_at_20_max
value: 37.43760229546025
- type: nauc_recall_at_20_std
value: 31.477306209504473
- type: nauc_recall_at_3_diff1
value: 28.060178345037656
- type: nauc_recall_at_3_max
value: 27.364528772057188
- type: nauc_recall_at_3_std
value: 13.396166074057525
- type: nauc_recall_at_5_diff1
value: 25.015826317297275
- type: nauc_recall_at_5_max
value: 30.444068792645858
- type: nauc_recall_at_5_std
value: 17.667570824756893
- type: ndcg_at_1
value: 27.753
- type: ndcg_at_10
value: 42.563
- type: ndcg_at_100
value: 46.572
- type: ndcg_at_1000
value: 47.825
- type: ndcg_at_20
value: 44.335
- type: ndcg_at_3
value: 36.54
- type: ndcg_at_5
value: 39.945
- type: precision_at_1
value: 27.753
- type: precision_at_10
value: 6.460000000000001
- type: precision_at_100
value: 0.859
- type: precision_at_1000
value: 0.098
- type: precision_at_20
value: 3.615
- type: precision_at_3
value: 15.542
- type: precision_at_5
value: 11.134
- type: recall_at_1
value: 25.789
- type: recall_at_10
value: 59.022
- type: recall_at_100
value: 77.49000000000001
- type: recall_at_1000
value: 87.075
- type: recall_at_20
value: 65.813
- type: recall_at_3
value: 42.931000000000004
- type: recall_at_5
value: 51.148
- task:
type: Retrieval
dataset:
name: MTEB FEVER (default)
type: mteb/fever
config: default
split: train
revision: bea83ef9e8fb933d90a2f1d5515737465d613e12
metrics:
- type: main_score
value: 39.544000000000004
- type: map_at_1
value: 22.122
- type: map_at_10
value: 33.021
- type: map_at_100
value: 33.919
- type: map_at_1000
value: 33.974
- type: map_at_20
value: 33.57
- type: map_at_3
value: 29.823
- type: map_at_5
value: 31.698999999999998
- type: mrr_at_1
value: 24.632547126855478
- type: mrr_at_10
value: 36.27443108803912
- type: mrr_at_100
value: 37.116518613084615
- type: mrr_at_1000
value: 37.16035774805865
- type: mrr_at_20
value: 36.80159654321279
- type: mrr_at_3
value: 32.990923716721404
- type: mrr_at_5
value: 34.94070060407349
- type: nauc_map_at_1000_diff1
value: 27.10508470711112
- type: nauc_map_at_1000_max
value: 19.122369111175676
- type: nauc_map_at_1000_std
value: 7.556987978069723
- type: nauc_map_at_100_diff1
value: 27.093923762733823
- type: nauc_map_at_100_max
value: 19.125020548640016
- type: nauc_map_at_100_std
value: 7.557312925203464
- type: nauc_map_at_10_diff1
value: 26.99461859425635
- type: nauc_map_at_10_max
value: 18.810433489022632
- type: nauc_map_at_10_std
value: 7.03658830181476
- type: nauc_map_at_1_diff1
value: 29.795588990770582
- type: nauc_map_at_1_max
value: 13.964929294547716
- type: nauc_map_at_1_std
value: 2.014683211424629
- type: nauc_map_at_20_diff1
value: 27.04355894870037
- type: nauc_map_at_20_max
value: 19.053125107329684
- type: nauc_map_at_20_std
value: 7.403819685723809
- type: nauc_map_at_3_diff1
value: 27.04176214403396
- type: nauc_map_at_3_max
value: 17.09452873117263
- type: nauc_map_at_3_std
value: 4.930539918889367
- type: nauc_map_at_5_diff1
value: 26.94702978336792
- type: nauc_map_at_5_max
value: 18.11738254686368
- type: nauc_map_at_5_std
value: 6.07806250068151
- type: nauc_mrr_at_1000_diff1
value: 28.115687821599565
- type: nauc_mrr_at_1000_max
value: 20.022657307983696
- type: nauc_mrr_at_1000_std
value: 6.049841455492619
- type: nauc_mrr_at_100_diff1
value: 28.10698592392563
- type: nauc_mrr_at_100_max
value: 20.039825659015783
- type: nauc_mrr_at_100_std
value: 6.067374534430724
- type: nauc_mrr_at_10_diff1
value: 28.01268183804612
- type: nauc_mrr_at_10_max
value: 19.866652130070182
- type: nauc_mrr_at_10_std
value: 5.7238201723873265
- type: nauc_mrr_at_1_diff1
value: 31.03974019855326
- type: nauc_mrr_at_1_max
value: 14.816836320215055
- type: nauc_mrr_at_1_std
value: 0.9833928793907583
- type: nauc_mrr_at_20_diff1
value: 28.0597179017192
- type: nauc_mrr_at_20_max
value: 20.03479616925749
- type: nauc_mrr_at_20_std
value: 6.003188285805082
- type: nauc_mrr_at_3_diff1
value: 28.102700297200357
- type: nauc_mrr_at_3_max
value: 18.21123845169275
- type: nauc_mrr_at_3_std
value: 3.76508019698659
- type: nauc_mrr_at_5_diff1
value: 27.995019911180925
- type: nauc_mrr_at_5_max
value: 19.247724901440165
- type: nauc_mrr_at_5_std
value: 4.867678189338567
- type: nauc_ndcg_at_1000_diff1
value: 27.001039847228327
- type: nauc_ndcg_at_1000_max
value: 22.747905636602912
- type: nauc_ndcg_at_1000_std
value: 12.131044834687508
- type: nauc_ndcg_at_100_diff1
value: 26.76465860993926
- type: nauc_ndcg_at_100_max
value: 22.999445448847684
- type: nauc_ndcg_at_100_std
value: 12.371530472512905
- type: nauc_ndcg_at_10_diff1
value: 26.271134051292965
- type: nauc_ndcg_at_10_max
value: 21.708152120634573
- type: nauc_ndcg_at_10_std
value: 9.956366430157477
- type: nauc_ndcg_at_1_diff1
value: 31.03974019855326
- type: nauc_ndcg_at_1_max
value: 14.816836320215055
- type: nauc_ndcg_at_1_std
value: 0.9833928793907583
- type: nauc_ndcg_at_20_diff1
value: 26.418892680753135
- type: nauc_ndcg_at_20_max
value: 22.57247401098405
- type: nauc_ndcg_at_20_std
value: 11.318053940484274
- type: nauc_ndcg_at_3_diff1
value: 26.48774759445648
- type: nauc_ndcg_at_3_max
value: 18.362023388227982
- type: nauc_ndcg_at_3_std
value: 5.5666629539029095
- type: nauc_ndcg_at_5_diff1
value: 26.231548007732275
- type: nauc_ndcg_at_5_max
value: 20.108141595384858
- type: nauc_ndcg_at_5_std
value: 7.6451734794663775
- type: nauc_precision_at_1000_diff1
value: 10.312275248214805
- type: nauc_precision_at_1000_max
value: 29.125490605573674
- type: nauc_precision_at_1000_std
value: 27.980820440423077
- type: nauc_precision_at_100_diff1
value: 17.082042984133537
- type: nauc_precision_at_100_max
value: 35.99017212218388
- type: nauc_precision_at_100_std
value: 31.93133992782479
- type: nauc_precision_at_10_diff1
value: 22.490628336580023
- type: nauc_precision_at_10_max
value: 31.505838103581745
- type: nauc_precision_at_10_std
value: 19.04756851410053
- type: nauc_precision_at_1_diff1
value: 31.03974019855326
- type: nauc_precision_at_1_max
value: 14.816836320215055
- type: nauc_precision_at_1_std
value: 0.9833928793907583
- type: nauc_precision_at_20_diff1
value: 20.899414285866147
- type: nauc_precision_at_20_max
value: 34.88025857085343
- type: nauc_precision_at_20_std
value: 24.931946815895103
- type: nauc_precision_at_3_diff1
value: 24.7584008425105
- type: nauc_precision_at_3_max
value: 22.05117420112601
- type: nauc_precision_at_3_std
value: 7.0182363705564725
- type: nauc_precision_at_5_diff1
value: 23.70576059633269
- type: nauc_precision_at_5_max
value: 26.41538125990869
- type: nauc_precision_at_5_std
value: 11.811642655248509
- type: nauc_recall_at_1000_diff1
value: 22.666331359570606
- type: nauc_recall_at_1000_max
value: 42.022586063014
- type: nauc_recall_at_1000_std
value: 51.28078945425114
- type: nauc_recall_at_100_diff1
value: 22.038970758345137
- type: nauc_recall_at_100_max
value: 37.50016128544992
- type: nauc_recall_at_100_std
value: 38.42258858995013
- type: nauc_recall_at_10_diff1
value: 21.83627962615356
- type: nauc_recall_at_10_max
value: 28.17929505850154
- type: nauc_recall_at_10_std
value: 19.711584731579144
- type: nauc_recall_at_1_diff1
value: 29.795588990770582
- type: nauc_recall_at_1_max
value: 13.964929294547716
- type: nauc_recall_at_1_std
value: 2.014683211424629
- type: nauc_recall_at_20_diff1
value: 21.612730102448293
- type: nauc_recall_at_20_max
value: 31.944811405726607
- type: nauc_recall_at_20_std
value: 26.127216044442598
- type: nauc_recall_at_3_diff1
value: 23.365847069042577
- type: nauc_recall_at_3_max
value: 19.98758181260918
- type: nauc_recall_at_3_std
value: 8.28613301698417
- type: nauc_recall_at_5_diff1
value: 22.39829927032666
- type: nauc_recall_at_5_max
value: 23.59853172030037
- type: nauc_recall_at_5_std
value: 12.723055723433257
- type: ndcg_at_1
value: 24.633
- type: ndcg_at_10
value: 39.544000000000004
- type: ndcg_at_100
value: 43.85
- type: ndcg_at_1000
value: 45.283
- type: ndcg_at_20
value: 41.471999999999994
- type: ndcg_at_3
value: 33.168
- type: ndcg_at_5
value: 36.437000000000005
- type: precision_at_1
value: 24.633
- type: precision_at_10
value: 6.429
- type: precision_at_100
value: 0.881
- type: precision_at_1000
value: 0.10200000000000001
- type: precision_at_20
value: 3.649
- type: precision_at_3
value: 14.796999999999999
- type: precision_at_5
value: 10.695
- type: recall_at_1
value: 22.122
- type: recall_at_10
value: 56.74
- type: recall_at_100
value: 76.285
- type: recall_at_1000
value: 87.13300000000001
- type: recall_at_20
value: 64.067
- type: recall_at_3
value: 39.446
- type: recall_at_5
value: 47.379
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018 (default)
type: mteb/fiqa
config: default
split: dev
revision: 27a168819829fe9bcd655c2df245fb19452e8e06
metrics:
- type: main_score
value: 43.452
- type: map_at_1
value: 22.435
- type: map_at_10
value: 35.832
- type: map_at_100
value: 37.632
- type: map_at_1000
value: 37.799
- type: map_at_20
value: 36.8
- type: map_at_3
value: 31.689
- type: map_at_5
value: 34.266999999999996
- type: mrr_at_1
value: 41.4
- type: mrr_at_10
value: 50.4342857142857
- type: mrr_at_100
value: 51.21821035615517
- type: mrr_at_1000
value: 51.266442546162516
- type: mrr_at_20
value: 50.91712414176811
- type: mrr_at_3
value: 48.43333333333335
- type: mrr_at_5
value: 49.64333333333333
- type: nauc_map_at_1000_diff1
value: 42.153292679851276
- type: nauc_map_at_1000_max
value: 29.601619670272207
- type: nauc_map_at_1000_std
value: -0.7383965062341508
- type: nauc_map_at_100_diff1
value: 42.07556710921584
- type: nauc_map_at_100_max
value: 29.464776621441306
- type: nauc_map_at_100_std
value: -0.7388582229371314
- type: nauc_map_at_10_diff1
value: 42.67413345132165
- type: nauc_map_at_10_max
value: 27.785600435680596
- type: nauc_map_at_10_std
value: -2.389554217354724
- type: nauc_map_at_1_diff1
value: 52.10965701450313
- type: nauc_map_at_1_max
value: 21.771380078575188
- type: nauc_map_at_1_std
value: -10.678726101360372
- type: nauc_map_at_20_diff1
value: 42.03673683254875
- type: nauc_map_at_20_max
value: 28.619853272850836
- type: nauc_map_at_20_std
value: -1.352968012843277
- type: nauc_map_at_3_diff1
value: 45.02940789042909
- type: nauc_map_at_3_max
value: 25.73172449783013
- type: nauc_map_at_3_std
value: -5.405882526355802
- type: nauc_map_at_5_diff1
value: 43.70377705353985
- type: nauc_map_at_5_max
value: 27.06264350949003
- type: nauc_map_at_5_std
value: -3.4566488552159638
- type: nauc_mrr_at_1000_diff1
value: 46.960388982124144
- type: nauc_mrr_at_1000_max
value: 38.551457905980335
- type: nauc_mrr_at_1000_std
value: -2.7251142586180452
- type: nauc_mrr_at_100_diff1
value: 46.95671307176507
- type: nauc_mrr_at_100_max
value: 38.557512138487596
- type: nauc_mrr_at_100_std
value: -2.670703447776175
- type: nauc_mrr_at_10_diff1
value: 46.88159177559446
- type: nauc_mrr_at_10_max
value: 38.274948278122025
- type: nauc_mrr_at_10_std
value: -2.7583668412620073
- type: nauc_mrr_at_1_diff1
value: 53.27259228216133
- type: nauc_mrr_at_1_max
value: 39.950693167149126
- type: nauc_mrr_at_1_std
value: -7.918584226469945
- type: nauc_mrr_at_20_diff1
value: 46.81459261048091
- type: nauc_mrr_at_20_max
value: 38.41539751417714
- type: nauc_mrr_at_20_std
value: -2.822290535243399
- type: nauc_mrr_at_3_diff1
value: 47.02719824632526
- type: nauc_mrr_at_3_max
value: 38.18543976624867
- type: nauc_mrr_at_3_std
value: -4.303513153685435
- type: nauc_mrr_at_5_diff1
value: 47.30493874120552
- type: nauc_mrr_at_5_max
value: 38.74873358201773
- type: nauc_mrr_at_5_std
value: -3.018961404554449
- type: nauc_ndcg_at_1000_diff1
value: 41.34144224154281
- type: nauc_ndcg_at_1000_max
value: 34.69321113499929
- type: nauc_ndcg_at_1000_std
value: 3.219617986178771
- type: nauc_ndcg_at_100_diff1
value: 40.527324056051725
- type: nauc_ndcg_at_100_max
value: 33.35505365876672
- type: nauc_ndcg_at_100_std
value: 4.298571598887196
- type: nauc_ndcg_at_10_diff1
value: 40.596904721915195
- type: nauc_ndcg_at_10_max
value: 29.70053338455224
- type: nauc_ndcg_at_10_std
value: 0.6266822992314294
- type: nauc_ndcg_at_1_diff1
value: 53.27259228216133
- type: nauc_ndcg_at_1_max
value: 39.950693167149126
- type: nauc_ndcg_at_1_std
value: -7.918584226469945
- type: nauc_ndcg_at_20_diff1
value: 39.5053199292571
- type: nauc_ndcg_at_20_max
value: 30.481412422904842
- type: nauc_ndcg_at_20_std
value: 1.8476982628200813
- type: nauc_ndcg_at_3_diff1
value: 41.230751065511214
- type: nauc_ndcg_at_3_max
value: 32.91537830913592
- type: nauc_ndcg_at_3_std
value: -1.3527356468708593
- type: nauc_ndcg_at_5_diff1
value: 41.73919141452553
- type: nauc_ndcg_at_5_max
value: 31.475892998444465
- type: nauc_ndcg_at_5_std
value: -0.07736285589192196
- type: nauc_precision_at_1000_diff1
value: -4.581230675366823
- type: nauc_precision_at_1000_max
value: 35.28244987228938
- type: nauc_precision_at_1000_std
value: 19.848724961997043
- type: nauc_precision_at_100_diff1
value: -1.0470568797091013
- type: nauc_precision_at_100_max
value: 38.1171253236869
- type: nauc_precision_at_100_std
value: 25.36360406306609
- type: nauc_precision_at_10_diff1
value: 9.535027734151406
- type: nauc_precision_at_10_max
value: 35.156066985665404
- type: nauc_precision_at_10_std
value: 14.719266832480818
- type: nauc_precision_at_1_diff1
value: 53.27259228216133
- type: nauc_precision_at_1_max
value: 39.950693167149126
- type: nauc_precision_at_1_std
value: -7.918584226469945
- type: nauc_precision_at_20_diff1
value: 3.192256990840671
- type: nauc_precision_at_20_max
value: 36.28479694909197
- type: nauc_precision_at_20_std
value: 20.43233178984753
- type: nauc_precision_at_3_diff1
value: 23.206183980981294
- type: nauc_precision_at_3_max
value: 35.68270813316575
- type: nauc_precision_at_3_std
value: 5.763084791714441
- type: nauc_precision_at_5_diff1
value: 16.311253397261815
- type: nauc_precision_at_5_max
value: 35.362299389447415
- type: nauc_precision_at_5_std
value: 11.070937256666982
- type: nauc_recall_at_1000_diff1
value: 13.246657177207306
- type: nauc_recall_at_1000_max
value: 34.11290480894018
- type: nauc_recall_at_1000_std
value: 28.2944964848988
- type: nauc_recall_at_100_diff1
value: 22.167289428754383
- type: nauc_recall_at_100_max
value: 24.776741594012847
- type: nauc_recall_at_100_std
value: 22.59200251836463
- type: nauc_recall_at_10_diff1
value: 28.175864072418843
- type: nauc_recall_at_10_max
value: 17.931411205973205
- type: nauc_recall_at_10_std
value: 6.352643320621673
- type: nauc_recall_at_1_diff1
value: 52.10965701450313
- type: nauc_recall_at_1_max
value: 21.771380078575188
- type: nauc_recall_at_1_std
value: -10.678726101360372
- type: nauc_recall_at_20_diff1
value: 22.886480231407088
- type: nauc_recall_at_20_max
value: 18.617961694923725
- type: nauc_recall_at_20_std
value: 9.027816465740859
- type: nauc_recall_at_3_diff1
value: 35.80141995616683
- type: nauc_recall_at_3_max
value: 20.28065350985076
- type: nauc_recall_at_3_std
value: -2.0345418109694915
- type: nauc_recall_at_5_diff1
value: 33.96384875377924
- type: nauc_recall_at_5_max
value: 20.532436267734845
- type: nauc_recall_at_5_std
value: 1.626798085762849
- type: ndcg_at_1
value: 41.4
- type: ndcg_at_10
value: 43.452
- type: ndcg_at_100
value: 49.655
- type: ndcg_at_1000
value: 52.596
- type: ndcg_at_20
value: 45.769999999999996
- type: ndcg_at_3
value: 40.157
- type: ndcg_at_5
value: 41.481
- type: precision_at_1
value: 41.4
- type: precision_at_10
value: 11.4
- type: precision_at_100
value: 1.7919999999999998
- type: precision_at_1000
value: 0.22699999999999998
- type: precision_at_20
value: 6.77
- type: precision_at_3
value: 25.733
- type: precision_at_5
value: 19.16
- type: recall_at_1
value: 22.435
- type: recall_at_10
value: 50.638000000000005
- type: recall_at_100
value: 73.819
- type: recall_at_1000
value: 92.134
- type: recall_at_20
value: 57.618
- type: recall_at_3
value: 37.222
- type: recall_at_5
value: 43.602000000000004
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018 (default)
type: mteb/fiqa
config: default
split: test
revision: 27a168819829fe9bcd655c2df245fb19452e8e06
metrics:
- type: main_score
value: 40.965
- type: map_at_1
value: 19.926
- type: map_at_10
value: 32.867000000000004
- type: map_at_100
value: 34.97
- type: map_at_1000
value: 35.158
- type: map_at_20
value: 34.067
- type: map_at_3
value: 28.854000000000003
- type: map_at_5
value: 30.897999999999996
- type: mrr_at_1
value: 39.660493827160494
- type: mrr_at_10
value: 49.50402949245541
- type: mrr_at_100
value: 50.29938288169482
- type: mrr_at_1000
value: 50.33496267350495
- type: mrr_at_20
value: 49.94865952256876
- type: mrr_at_3
value: 46.939300411522616
- type: mrr_at_5
value: 48.39763374485595
- type: nauc_map_at_1000_diff1
value: 26.494855045027233
- type: nauc_map_at_1000_max
value: 27.303620903381308
- type: nauc_map_at_1000_std
value: -1.4561552032645217
- type: nauc_map_at_100_diff1
value: 26.508482155009656
- type: nauc_map_at_100_max
value: 27.197476666972175
- type: nauc_map_at_100_std
value: -1.4452769452596321
- type: nauc_map_at_10_diff1
value: 26.374957272593257
- type: nauc_map_at_10_max
value: 25.630002139583425
- type: nauc_map_at_10_std
value: -2.7457793654796303
- type: nauc_map_at_1_diff1
value: 32.633041679053
- type: nauc_map_at_1_max
value: 18.95078398066585
- type: nauc_map_at_1_std
value: -5.289494036677301
- type: nauc_map_at_20_diff1
value: 26.420832066149863
- type: nauc_map_at_20_max
value: 26.359849203621906
- type: nauc_map_at_20_std
value: -2.1819017372318896
- type: nauc_map_at_3_diff1
value: 27.926274507465116
- type: nauc_map_at_3_max
value: 23.05501535710793
- type: nauc_map_at_3_std
value: -4.325658334680725
- type: nauc_map_at_5_diff1
value: 27.269028553122237
- type: nauc_map_at_5_max
value: 24.606451651258975
- type: nauc_map_at_5_std
value: -3.8403860613657264
- type: nauc_mrr_at_1000_diff1
value: 32.27607008114233
- type: nauc_mrr_at_1000_max
value: 38.46692644890318
- type: nauc_mrr_at_1000_std
value: 0.8468212008159507
- type: nauc_mrr_at_100_diff1
value: 32.27189093598166
- type: nauc_mrr_at_100_max
value: 38.47818433738234
- type: nauc_mrr_at_100_std
value: 0.86792905411205
- type: nauc_mrr_at_10_diff1
value: 32.27220558113546
- type: nauc_mrr_at_10_max
value: 38.5768621159052
- type: nauc_mrr_at_10_std
value: 0.7700057337792469
- type: nauc_mrr_at_1_diff1
value: 36.11482505826484
- type: nauc_mrr_at_1_max
value: 38.21978668740053
- type: nauc_mrr_at_1_std
value: 0.32943153196143976
- type: nauc_mrr_at_20_diff1
value: 32.20429693089343
- type: nauc_mrr_at_20_max
value: 38.45040837784157
- type: nauc_mrr_at_20_std
value: 0.8750958039709447
- type: nauc_mrr_at_3_diff1
value: 32.36086055016881
- type: nauc_mrr_at_3_max
value: 37.587053558645614
- type: nauc_mrr_at_3_std
value: -0.14274663611290778
- type: nauc_mrr_at_5_diff1
value: 32.34528370475561
- type: nauc_mrr_at_5_max
value: 38.39546966320742
- type: nauc_mrr_at_5_std
value: 0.03003840440912135
- type: nauc_ndcg_at_1000_diff1
value: 26.691917929958546
- type: nauc_ndcg_at_1000_max
value: 32.58961283234771
- type: nauc_ndcg_at_1000_std
value: 2.130192967914691
- type: nauc_ndcg_at_100_diff1
value: 26.872799707070357
- type: nauc_ndcg_at_100_max
value: 31.503230695989938
- type: nauc_ndcg_at_100_std
value: 3.3435217735327365
- type: nauc_ndcg_at_10_diff1
value: 26.22445930619204
- type: nauc_ndcg_at_10_max
value: 28.332502188521534
- type: nauc_ndcg_at_10_std
value: -0.6029849879688606
- type: nauc_ndcg_at_1_diff1
value: 36.11482505826484
- type: nauc_ndcg_at_1_max
value: 38.21978668740053
- type: nauc_ndcg_at_1_std
value: 0.32943153196143976
- type: nauc_ndcg_at_20_diff1
value: 26.167580898864678
- type: nauc_ndcg_at_20_max
value: 29.138479770125414
- type: nauc_ndcg_at_20_std
value: 0.7659081670630379
- type: nauc_ndcg_at_3_diff1
value: 28.517059827975217
- type: nauc_ndcg_at_3_max
value: 31.64778875185945
- type: nauc_ndcg_at_3_std
value: -2.4273609628572763
- type: nauc_ndcg_at_5_diff1
value: 27.523926620375715
- type: nauc_ndcg_at_5_max
value: 29.36575905903845
- type: nauc_ndcg_at_5_std
value: -2.786795946093564
- type: nauc_precision_at_1000_diff1
value: -2.197004713716575
- type: nauc_precision_at_1000_max
value: 34.162523162834376
- type: nauc_precision_at_1000_std
value: 8.986097658449319
- type: nauc_precision_at_100_diff1
value: 3.134123713987622
- type: nauc_precision_at_100_max
value: 36.85792724470659
- type: nauc_precision_at_100_std
value: 14.919862433706202
- type: nauc_precision_at_10_diff1
value: 9.854013398949677
- type: nauc_precision_at_10_max
value: 37.525893321985755
- type: nauc_precision_at_10_std
value: 8.696222830370099
- type: nauc_precision_at_1_diff1
value: 36.11482505826484
- type: nauc_precision_at_1_max
value: 38.21978668740053
- type: nauc_precision_at_1_std
value: 0.32943153196143976
- type: nauc_precision_at_20_diff1
value: 6.858037910006135
- type: nauc_precision_at_20_max
value: 37.575334948715025
- type: nauc_precision_at_20_std
value: 11.273137972815993
- type: nauc_precision_at_3_diff1
value: 19.692731738653848
- type: nauc_precision_at_3_max
value: 37.619642815139876
- type: nauc_precision_at_3_std
value: 2.0435791110517734
- type: nauc_precision_at_5_diff1
value: 15.784071911592584
- type: nauc_precision_at_5_max
value: 39.05471012980888
- type: nauc_precision_at_5_std
value: 4.244303875173872
- type: nauc_recall_at_1000_diff1
value: 8.108025620944437
- type: nauc_recall_at_1000_max
value: 15.671603738840128
- type: nauc_recall_at_1000_std
value: 18.119568463795165
- type: nauc_recall_at_100_diff1
value: 18.48107834900102
- type: nauc_recall_at_100_max
value: 20.209949068924594
- type: nauc_recall_at_100_std
value: 19.01259420147078
- type: nauc_recall_at_10_diff1
value: 17.93458808832428
- type: nauc_recall_at_10_max
value: 17.81863413759591
- type: nauc_recall_at_10_std
value: 0.3549693938691897
- type: nauc_recall_at_1_diff1
value: 32.633041679053
- type: nauc_recall_at_1_max
value: 18.95078398066585
- type: nauc_recall_at_1_std
value: -5.289494036677301
- type: nauc_recall_at_20_diff1
value: 16.794175854855236
- type: nauc_recall_at_20_max
value: 17.227732008238515
- type: nauc_recall_at_20_std
value: 4.118397119053081
- type: nauc_recall_at_3_diff1
value: 22.401972357181148
- type: nauc_recall_at_3_max
value: 17.63593806963239
- type: nauc_recall_at_3_std
value: -4.308586374199029
- type: nauc_recall_at_5_diff1
value: 21.149136573276326
- type: nauc_recall_at_5_max
value: 18.586766434816816
- type: nauc_recall_at_5_std
value: -4.198379401256694
- type: ndcg_at_1
value: 39.660000000000004
- type: ndcg_at_10
value: 40.965
- type: ndcg_at_100
value: 48.254999999999995
- type: ndcg_at_1000
value: 51.307
- type: ndcg_at_20
value: 43.943
- type: ndcg_at_3
value: 37.379
- type: ndcg_at_5
value: 38.251000000000005
- type: precision_at_1
value: 39.660000000000004
- type: precision_at_10
value: 11.42
- type: precision_at_100
value: 1.8870000000000002
- type: precision_at_1000
value: 0.242
- type: precision_at_20
value: 6.9750000000000005
- type: precision_at_3
value: 25.308999999999997
- type: precision_at_5
value: 18.21
- type: recall_at_1
value: 19.926
- type: recall_at_10
value: 47.951
- type: recall_at_100
value: 74.703
- type: recall_at_1000
value: 93.003
- type: recall_at_20
value: 57.062000000000005
- type: recall_at_3
value: 33.942
- type: recall_at_5
value: 39.39
- task:
type: Retrieval
dataset:
name: MTEB FiQA2018 (default)
type: mteb/fiqa
config: default
split: train
revision: 27a168819829fe9bcd655c2df245fb19452e8e06
metrics:
- type: main_score
value: 40.109
- type: map_at_1
value: 19.428
- type: map_at_10
value: 32.132
- type: map_at_100
value: 34.095
- type: map_at_1000
value: 34.275
- type: map_at_20
value: 33.223
- type: map_at_3
value: 27.744999999999997
- type: map_at_5
value: 30.139
- type: mrr_at_1
value: 37.43636363636364
- type: mrr_at_10
value: 46.699083694083626
- type: mrr_at_100
value: 47.54203896354004
- type: mrr_at_1000
value: 47.58580275375307
- type: mrr_at_20
value: 47.198759673340916
- type: mrr_at_3
value: 44.00606060606058
- type: mrr_at_5
value: 45.532424242424206
- type: nauc_map_at_1000_diff1
value: 37.79487277315329
- type: nauc_map_at_1000_max
value: 26.0872634860684
- type: nauc_map_at_1000_std
value: -1.0279629281127562
- type: nauc_map_at_100_diff1
value: 37.78657280270208
- type: nauc_map_at_100_max
value: 25.97447841536255
- type: nauc_map_at_100_std
value: -1.0680554938428704
- type: nauc_map_at_10_diff1
value: 37.75308574166357
- type: nauc_map_at_10_max
value: 24.48296458166315
- type: nauc_map_at_10_std
value: -2.4706230738202115
- type: nauc_map_at_1_diff1
value: 42.6444957224609
- type: nauc_map_at_1_max
value: 17.312823253319056
- type: nauc_map_at_1_std
value: -5.949436425909343
- type: nauc_map_at_20_diff1
value: 37.77686951531909
- type: nauc_map_at_20_max
value: 25.40897694790217
- type: nauc_map_at_20_std
value: -1.6373032823813254
- type: nauc_map_at_3_diff1
value: 38.60509752110032
- type: nauc_map_at_3_max
value: 21.348684256478478
- type: nauc_map_at_3_std
value: -4.810470520501061
- type: nauc_map_at_5_diff1
value: 38.09219815429182
- type: nauc_map_at_5_max
value: 22.835685949895517
- type: nauc_map_at_5_std
value: -3.9401553768980317
- type: nauc_mrr_at_1000_diff1
value: 42.085039832710486
- type: nauc_mrr_at_1000_max
value: 35.16928700370471
- type: nauc_mrr_at_1000_std
value: 0.9299074027021959
- type: nauc_mrr_at_100_diff1
value: 42.07215266936291
- type: nauc_mrr_at_100_max
value: 35.17189764710499
- type: nauc_mrr_at_100_std
value: 0.9530063786251255
- type: nauc_mrr_at_10_diff1
value: 41.98881249801537
- type: nauc_mrr_at_10_max
value: 35.09509256566193
- type: nauc_mrr_at_10_std
value: 0.794172569517322
- type: nauc_mrr_at_1_diff1
value: 46.077585624783666
- type: nauc_mrr_at_1_max
value: 35.39409660069189
- type: nauc_mrr_at_1_std
value: -1.533340930893894
- type: nauc_mrr_at_20_diff1
value: 42.00798938048869
- type: nauc_mrr_at_20_max
value: 35.158812601568314
- type: nauc_mrr_at_20_std
value: 0.9056333950877309
- type: nauc_mrr_at_3_diff1
value: 42.59361038737236
- type: nauc_mrr_at_3_max
value: 34.98645026093173
- type: nauc_mrr_at_3_std
value: 0.0023951964897005276
- type: nauc_mrr_at_5_diff1
value: 42.2008327692426
- type: nauc_mrr_at_5_max
value: 34.81803036128217
- type: nauc_mrr_at_5_std
value: 0.21171705701408164
- type: nauc_ndcg_at_1000_diff1
value: 37.59389251562878
- type: nauc_ndcg_at_1000_max
value: 31.01866705444215
- type: nauc_ndcg_at_1000_std
value: 3.218055181162254
- type: nauc_ndcg_at_100_diff1
value: 37.12230330729587
- type: nauc_ndcg_at_100_max
value: 29.935095243810167
- type: nauc_ndcg_at_100_std
value: 3.7200669195773437
- type: nauc_ndcg_at_10_diff1
value: 36.96336573645087
- type: nauc_ndcg_at_10_max
value: 27.254442355026736
- type: nauc_ndcg_at_10_std
value: 0.14665445172204347
- type: nauc_ndcg_at_1_diff1
value: 46.077585624783666
- type: nauc_ndcg_at_1_max
value: 35.39409660069189
- type: nauc_ndcg_at_1_std
value: -1.533340930893894
- type: nauc_ndcg_at_20_diff1
value: 37.074571249220114
- type: nauc_ndcg_at_20_max
value: 28.399045519694383
- type: nauc_ndcg_at_20_std
value: 1.550978185230789
- type: nauc_ndcg_at_3_diff1
value: 37.59928811563115
- type: nauc_ndcg_at_3_max
value: 28.982761358417385
- type: nauc_ndcg_at_3_std
value: -1.238849396182027
- type: nauc_ndcg_at_5_diff1
value: 37.406291033992446
- type: nauc_ndcg_at_5_max
value: 27.126892323381625
- type: nauc_ndcg_at_5_std
value: -1.3896938747693226
- type: nauc_precision_at_1000_diff1
value: -1.2563257534530403
- type: nauc_precision_at_1000_max
value: 35.797345612886375
- type: nauc_precision_at_1000_std
value: 17.41472736776915
- type: nauc_precision_at_100_diff1
value: 4.952971767905249
- type: nauc_precision_at_100_max
value: 38.73911955037188
- type: nauc_precision_at_100_std
value: 21.066156963949332
- type: nauc_precision_at_10_diff1
value: 15.966764461879817
- type: nauc_precision_at_10_max
value: 38.48114461750857
- type: nauc_precision_at_10_std
value: 12.21332681895805
- type: nauc_precision_at_1_diff1
value: 46.077585624783666
- type: nauc_precision_at_1_max
value: 35.39409660069189
- type: nauc_precision_at_1_std
value: -1.533340930893894
- type: nauc_precision_at_20_diff1
value: 12.863561744619767
- type: nauc_precision_at_20_max
value: 39.97799707791258
- type: nauc_precision_at_20_std
value: 16.479326047022596
- type: nauc_precision_at_3_diff1
value: 26.270777531636746
- type: nauc_precision_at_3_max
value: 36.221006349866734
- type: nauc_precision_at_3_std
value: 4.132546766200079
- type: nauc_precision_at_5_diff1
value: 21.514798016502567
- type: nauc_precision_at_5_max
value: 36.78486009359055
- type: nauc_precision_at_5_std
value: 6.550630352857908
- type: nauc_recall_at_1000_diff1
value: 18.296781187173607
- type: nauc_recall_at_1000_max
value: 16.345940832080522
- type: nauc_recall_at_1000_std
value: 25.876663129650872
- type: nauc_recall_at_100_diff1
value: 22.34531148768501
- type: nauc_recall_at_100_max
value: 18.990695799061044
- type: nauc_recall_at_100_std
value: 17.324570191171166
- type: nauc_recall_at_10_diff1
value: 26.9638177104112
- type: nauc_recall_at_10_max
value: 18.015343527525935
- type: nauc_recall_at_10_std
value: 1.9739545303234234
- type: nauc_recall_at_1_diff1
value: 42.6444957224609
- type: nauc_recall_at_1_max
value: 17.312823253319056
- type: nauc_recall_at_1_std
value: -5.949436425909343
- type: nauc_recall_at_20_diff1
value: 25.89370052543927
- type: nauc_recall_at_20_max
value: 19.183543872025176
- type: nauc_recall_at_20_std
value: 5.845168418090508
- type: nauc_recall_at_3_diff1
value: 32.42588221746735
- type: nauc_recall_at_3_max
value: 16.982760583883568
- type: nauc_recall_at_3_std
value: -3.8563202145821722
- type: nauc_recall_at_5_diff1
value: 29.949897843617833
- type: nauc_recall_at_5_max
value: 16.763718302170407
- type: nauc_recall_at_5_std
value: -2.4876499904473515
- type: ndcg_at_1
value: 37.436
- type: ndcg_at_10
value: 40.109
- type: ndcg_at_100
value: 47.099000000000004
- type: ndcg_at_1000
value: 50.09
- type: ndcg_at_20
value: 42.807
- type: ndcg_at_3
value: 35.614000000000004
- type: ndcg_at_5
value: 37.202
- type: precision_at_1
value: 37.436
- type: precision_at_10
value: 11.14
- type: precision_at_100
value: 1.839
- type: precision_at_1000
value: 0.23500000000000001
- type: precision_at_20
value: 6.7379999999999995
- type: precision_at_3
value: 23.660999999999998
- type: precision_at_5
value: 17.589
- type: recall_at_1
value: 19.428
- type: recall_at_10
value: 48.262
- type: recall_at_100
value: 74.533
- type: recall_at_1000
value: 92.707
- type: recall_at_20
value: 56.684
- type: recall_at_3
value: 32.918
- type: recall_at_5
value: 39.565
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA (default)
type: mteb/hotpotqa
config: default
split: dev
revision: ab518f4d6fcca38d87c25209f94beba119d02014
metrics:
- type: main_score
value: 27.319
- type: map_at_1
value: 15.128
- type: map_at_10
value: 21.121000000000002
- type: map_at_100
value: 21.973000000000003
- type: map_at_1000
value: 22.073999999999998
- type: map_at_20
value: 21.582
- type: map_at_3
value: 19.323999999999998
- type: map_at_5
value: 20.358
- type: mrr_at_1
value: 30.255186341105194
- type: mrr_at_10
value: 36.86408857649899
- type: mrr_at_100
value: 37.61006780296905
- type: mrr_at_1000
value: 37.6719518655528
- type: mrr_at_20
value: 37.281873457624904
- type: mrr_at_3
value: 35.11107031393417
- type: mrr_at_5
value: 36.16669726454924
- type: nauc_map_at_1000_diff1
value: 48.847643794509025
- type: nauc_map_at_1000_max
value: 39.770122515103765
- type: nauc_map_at_1000_std
value: 22.807391952064208
- type: nauc_map_at_100_diff1
value: 48.84836642359173
- type: nauc_map_at_100_max
value: 39.74431147433934
- type: nauc_map_at_100_std
value: 22.742541317197745
- type: nauc_map_at_10_diff1
value: 49.4233743176388
- type: nauc_map_at_10_max
value: 39.91993995966069
- type: nauc_map_at_10_std
value: 21.91665772388842
- type: nauc_map_at_1_diff1
value: 58.0289225809443
- type: nauc_map_at_1_max
value: 43.869928323541096
- type: nauc_map_at_1_std
value: 18.414910058766996
- type: nauc_map_at_20_diff1
value: 48.946851303177276
- type: nauc_map_at_20_max
value: 39.637722313894855
- type: nauc_map_at_20_std
value: 22.36990188694816
- type: nauc_map_at_3_diff1
value: 50.9755320771899
- type: nauc_map_at_3_max
value: 41.04341520188789
- type: nauc_map_at_3_std
value: 20.73111078435784
- type: nauc_map_at_5_diff1
value: 50.29129566530908
- type: nauc_map_at_5_max
value: 40.820300878574244
- type: nauc_map_at_5_std
value: 21.632271796529135
- type: nauc_mrr_at_1000_diff1
value: 52.9718341979705
- type: nauc_mrr_at_1000_max
value: 42.0423808089922
- type: nauc_mrr_at_1000_std
value: 20.9799244867451
- type: nauc_mrr_at_100_diff1
value: 52.95863792063511
- type: nauc_mrr_at_100_max
value: 42.02576849188542
- type: nauc_mrr_at_100_std
value: 20.979507647583475
- type: nauc_mrr_at_10_diff1
value: 53.11373095655967
- type: nauc_mrr_at_10_max
value: 42.105421475578744
- type: nauc_mrr_at_10_std
value: 20.722957880316024
- type: nauc_mrr_at_1_diff1
value: 58.0289225809443
- type: nauc_mrr_at_1_max
value: 43.869928323541096
- type: nauc_mrr_at_1_std
value: 18.414910058766996
- type: nauc_mrr_at_20_diff1
value: 53.00870629169023
- type: nauc_mrr_at_20_max
value: 42.041577465122714
- type: nauc_mrr_at_20_std
value: 20.912051250134024
- type: nauc_mrr_at_3_diff1
value: 53.77971068931782
- type: nauc_mrr_at_3_max
value: 42.49181103386521
- type: nauc_mrr_at_3_std
value: 20.140203552794727
- type: nauc_mrr_at_5_diff1
value: 53.475770315841466
- type: nauc_mrr_at_5_max
value: 42.386857011080636
- type: nauc_mrr_at_5_std
value: 20.640622787052557
- type: nauc_ndcg_at_1000_diff1
value: 46.64695418952649
- type: nauc_ndcg_at_1000_max
value: 39.66342903978547
- type: nauc_ndcg_at_1000_std
value: 26.893939887796396
- type: nauc_ndcg_at_100_diff1
value: 46.530602184859525
- type: nauc_ndcg_at_100_max
value: 38.925897286535076
- type: nauc_ndcg_at_100_std
value: 25.713663164332328
- type: nauc_ndcg_at_10_diff1
value: 48.23171706325336
- type: nauc_ndcg_at_10_max
value: 39.30673245947019
- type: nauc_ndcg_at_10_std
value: 22.76919274555658
- type: nauc_ndcg_at_1_diff1
value: 58.0289225809443
- type: nauc_ndcg_at_1_max
value: 43.869928323541096
- type: nauc_ndcg_at_1_std
value: 18.414910058766996
- type: nauc_ndcg_at_20_diff1
value: 47.13236431440934
- type: nauc_ndcg_at_20_max
value: 38.653618449937696
- type: nauc_ndcg_at_20_std
value: 23.92208488492667
- type: nauc_ndcg_at_3_diff1
value: 50.624514000929224
- type: nauc_ndcg_at_3_max
value: 41.105294702595586
- type: nauc_ndcg_at_3_std
value: 21.130087325966326
- type: nauc_ndcg_at_5_diff1
value: 49.69172113431722
- type: nauc_ndcg_at_5_max
value: 40.75186606108917
- type: nauc_ndcg_at_5_std
value: 22.32512469362848
- type: nauc_precision_at_1000_diff1
value: 23.160094544144965
- type: nauc_precision_at_1000_max
value: 28.661321254269932
- type: nauc_precision_at_1000_std
value: 38.94759586932569
- type: nauc_precision_at_100_diff1
value: 28.051281498032537
- type: nauc_precision_at_100_max
value: 27.9218322543234
- type: nauc_precision_at_100_std
value: 32.35825145192534
- type: nauc_precision_at_10_diff1
value: 38.18649250868066
- type: nauc_precision_at_10_max
value: 32.670270264955114
- type: nauc_precision_at_10_std
value: 24.605022379185275
- type: nauc_precision_at_1_diff1
value: 58.0289225809443
- type: nauc_precision_at_1_max
value: 43.869928323541096
- type: nauc_precision_at_1_std
value: 18.414910058766996
- type: nauc_precision_at_20_diff1
value: 34.064945849558676
- type: nauc_precision_at_20_max
value: 29.99092267433295
- type: nauc_precision_at_20_std
value: 27.480111135148235
- type: nauc_precision_at_3_diff1
value: 45.96635358139459
- type: nauc_precision_at_3_max
value: 39.19530950679048
- type: nauc_precision_at_3_std
value: 22.692664071111174
- type: nauc_precision_at_5_diff1
value: 42.80772429338654
- type: nauc_precision_at_5_max
value: 37.30900415179882
- type: nauc_precision_at_5_std
value: 24.39700402843405
- type: nauc_recall_at_1000_diff1
value: 23.160094544144993
- type: nauc_recall_at_1000_max
value: 28.661321254269968
- type: nauc_recall_at_1000_std
value: 38.94759586932564
- type: nauc_recall_at_100_diff1
value: 28.051281498032505
- type: nauc_recall_at_100_max
value: 27.921832254323387
- type: nauc_recall_at_100_std
value: 32.35825145192532
- type: nauc_recall_at_10_diff1
value: 38.186492508680615
- type: nauc_recall_at_10_max
value: 32.6702702649551
- type: nauc_recall_at_10_std
value: 24.60502237918523
- type: nauc_recall_at_1_diff1
value: 58.0289225809443
- type: nauc_recall_at_1_max
value: 43.869928323541096
- type: nauc_recall_at_1_std
value: 18.414910058766996
- type: nauc_recall_at_20_diff1
value: 34.06494584955859
- type: nauc_recall_at_20_max
value: 29.990922674332886
- type: nauc_recall_at_20_std
value: 27.48011113514819
- type: nauc_recall_at_3_diff1
value: 45.96635358139468
- type: nauc_recall_at_3_max
value: 39.19530950679048
- type: nauc_recall_at_3_std
value: 22.69266407111118
- type: nauc_recall_at_5_diff1
value: 42.807724293386535
- type: nauc_recall_at_5_max
value: 37.30900415179882
- type: nauc_recall_at_5_std
value: 24.397004028434065
- type: ndcg_at_1
value: 30.255
- type: ndcg_at_10
value: 27.319
- type: ndcg_at_100
value: 31.384
- type: ndcg_at_1000
value: 33.976
- type: ndcg_at_20
value: 28.811999999999998
- type: ndcg_at_3
value: 23.937
- type: ndcg_at_5
value: 25.665
- type: precision_at_1
value: 30.255
- type: precision_at_10
value: 6.049
- type: precision_at_100
value: 0.931
- type: precision_at_1000
value: 0.128
- type: precision_at_20
value: 3.5060000000000002
- type: precision_at_3
value: 15.097
- type: precision_at_5
value: 10.427999999999999
- type: recall_at_1
value: 15.128
- type: recall_at_10
value: 30.246000000000002
- type: recall_at_100
value: 46.549
- type: recall_at_1000
value: 63.943000000000005
- type: recall_at_20
value: 35.056
- type: recall_at_3
value: 22.645
- type: recall_at_5
value: 26.069
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA (default)
type: mteb/hotpotqa
config: default
split: test
revision: ab518f4d6fcca38d87c25209f94beba119d02014
metrics:
- type: main_score
value: 25.594
- type: map_at_1
value: 13.328999999999999
- type: map_at_10
value: 19.622999999999998
- type: map_at_100
value: 20.443
- type: map_at_1000
value: 20.547
- type: map_at_20
value: 20.061
- type: map_at_3
value: 17.738
- type: map_at_5
value: 18.837
- type: mrr_at_1
value: 26.65766374071573
- type: mrr_at_10
value: 33.638001993505
- type: mrr_at_100
value: 34.36519863954255
- type: mrr_at_1000
value: 34.43147960519581
- type: mrr_at_20
value: 34.05666557215429
- type: mrr_at_3
value: 31.672293495385954
- type: mrr_at_5
value: 32.81003826243525
- type: nauc_map_at_1000_diff1
value: 42.92564602155638
- type: nauc_map_at_1000_max
value: 30.93107038343275
- type: nauc_map_at_1000_std
value: 20.940662128255337
- type: nauc_map_at_100_diff1
value: 42.936496253513425
- type: nauc_map_at_100_max
value: 30.91180851399884
- type: nauc_map_at_100_std
value: 20.868468122916305
- type: nauc_map_at_10_diff1
value: 43.187610636795725
- type: nauc_map_at_10_max
value: 30.845550958099743
- type: nauc_map_at_10_std
value: 19.982030218251648
- type: nauc_map_at_1_diff1
value: 52.309034270608876
- type: nauc_map_at_1_max
value: 31.709473219405616
- type: nauc_map_at_1_std
value: 14.43851822909325
- type: nauc_map_at_20_diff1
value: 43.03978703190173
- type: nauc_map_at_20_max
value: 30.86942775161063
- type: nauc_map_at_20_std
value: 20.45899665222483
- type: nauc_map_at_3_diff1
value: 45.41524124545876
- type: nauc_map_at_3_max
value: 31.643858919734857
- type: nauc_map_at_3_std
value: 18.38563207630924
- type: nauc_map_at_5_diff1
value: 43.901640168008235
- type: nauc_map_at_5_max
value: 30.922718235695463
- type: nauc_map_at_5_std
value: 19.282329560614524
- type: nauc_mrr_at_1000_diff1
value: 46.58287148422368
- type: nauc_mrr_at_1000_max
value: 31.253308486580433
- type: nauc_mrr_at_1000_std
value: 17.55539836580872
- type: nauc_mrr_at_100_diff1
value: 46.573742959015114
- type: nauc_mrr_at_100_max
value: 31.24499189752598
- type: nauc_mrr_at_100_std
value: 17.552024988025664
- type: nauc_mrr_at_10_diff1
value: 46.66525086279671
- type: nauc_mrr_at_10_max
value: 31.230319076841546
- type: nauc_mrr_at_10_std
value: 17.165865128607045
- type: nauc_mrr_at_1_diff1
value: 52.309034270608876
- type: nauc_mrr_at_1_max
value: 31.709473219405616
- type: nauc_mrr_at_1_std
value: 14.43851822909325
- type: nauc_mrr_at_20_diff1
value: 46.588047522930275
- type: nauc_mrr_at_20_max
value: 31.22329428012019
- type: nauc_mrr_at_20_std
value: 17.402623764852763
- type: nauc_mrr_at_3_diff1
value: 47.616857174725034
- type: nauc_mrr_at_3_max
value: 31.683285399772792
- type: nauc_mrr_at_3_std
value: 16.51070486195255
- type: nauc_mrr_at_5_diff1
value: 46.94600877034942
- type: nauc_mrr_at_5_max
value: 31.373228910578714
- type: nauc_mrr_at_5_std
value: 16.878955713897653
- type: nauc_ndcg_at_1000_diff1
value: 40.78231860794336
- type: nauc_ndcg_at_1000_max
value: 31.269823988995526
- type: nauc_ndcg_at_1000_std
value: 25.281682702756076
- type: nauc_ndcg_at_100_diff1
value: 40.85167326240917
- type: nauc_ndcg_at_100_max
value: 30.7489966502373
- type: nauc_ndcg_at_100_std
value: 24.034633639362347
- type: nauc_ndcg_at_10_diff1
value: 41.64071111332904
- type: nauc_ndcg_at_10_max
value: 30.37691893443363
- type: nauc_ndcg_at_10_std
value: 20.727400392560106
- type: nauc_ndcg_at_1_diff1
value: 52.309034270608876
- type: nauc_ndcg_at_1_max
value: 31.709473219405616
- type: nauc_ndcg_at_1_std
value: 14.43851822909325
- type: nauc_ndcg_at_20_diff1
value: 41.20606946396262
- type: nauc_ndcg_at_20_max
value: 30.427872627999093
- type: nauc_ndcg_at_20_std
value: 21.967337950787567
- type: nauc_ndcg_at_3_diff1
value: 44.769112514565826
- type: nauc_ndcg_at_3_max
value: 31.587253638273616
- type: nauc_ndcg_at_3_std
value: 18.44337537736502
- type: nauc_ndcg_at_5_diff1
value: 42.843534154265036
- type: nauc_ndcg_at_5_max
value: 30.656690565149763
- type: nauc_ndcg_at_5_std
value: 19.55710266805375
- type: nauc_precision_at_1000_diff1
value: 20.571918733919397
- type: nauc_precision_at_1000_max
value: 25.92113127229473
- type: nauc_precision_at_1000_std
value: 39.87020670433835
- type: nauc_precision_at_100_diff1
value: 25.698473178831794
- type: nauc_precision_at_100_max
value: 25.352318812011827
- type: nauc_precision_at_100_std
value: 32.846058298512595
- type: nauc_precision_at_10_diff1
value: 31.870926938946866
- type: nauc_precision_at_10_max
value: 26.926394860214852
- type: nauc_precision_at_10_std
value: 24.246520824850055
- type: nauc_precision_at_1_diff1
value: 52.309034270608876
- type: nauc_precision_at_1_max
value: 31.709473219405616
- type: nauc_precision_at_1_std
value: 14.43851822909325
- type: nauc_precision_at_20_diff1
value: 30.028917020847068
- type: nauc_precision_at_20_max
value: 26.597814109731576
- type: nauc_precision_at_20_std
value: 27.346420398015525
- type: nauc_precision_at_3_diff1
value: 40.29446945365145
- type: nauc_precision_at_3_max
value: 31.134170069020573
- type: nauc_precision_at_3_std
value: 20.590305745326894
- type: nauc_precision_at_5_diff1
value: 35.844201172387955
- type: nauc_precision_at_5_max
value: 28.624481394118277
- type: nauc_precision_at_5_std
value: 22.316991114813813
- type: nauc_recall_at_1000_diff1
value: 20.571918733919446
- type: nauc_recall_at_1000_max
value: 25.92113127229482
- type: nauc_recall_at_1000_std
value: 39.87020670433847
- type: nauc_recall_at_100_diff1
value: 25.698473178831783
- type: nauc_recall_at_100_max
value: 25.352318812011802
- type: nauc_recall_at_100_std
value: 32.84605829851259
- type: nauc_recall_at_10_diff1
value: 31.870926938946916
- type: nauc_recall_at_10_max
value: 26.92639486021487
- type: nauc_recall_at_10_std
value: 24.246520824850094
- type: nauc_recall_at_1_diff1
value: 52.309034270608876
- type: nauc_recall_at_1_max
value: 31.709473219405616
- type: nauc_recall_at_1_std
value: 14.43851822909325
- type: nauc_recall_at_20_diff1
value: 30.028917020847096
- type: nauc_recall_at_20_max
value: 26.597814109731615
- type: nauc_recall_at_20_std
value: 27.346420398015525
- type: nauc_recall_at_3_diff1
value: 40.294469453651374
- type: nauc_recall_at_3_max
value: 31.134170069020577
- type: nauc_recall_at_3_std
value: 20.590305745326855
- type: nauc_recall_at_5_diff1
value: 35.844201172387926
- type: nauc_recall_at_5_max
value: 28.624481394118273
- type: nauc_recall_at_5_std
value: 22.31699111481382
- type: ndcg_at_1
value: 26.657999999999998
- type: ndcg_at_10
value: 25.594
- type: ndcg_at_100
value: 29.554000000000002
- type: ndcg_at_1000
value: 32.223
- type: ndcg_at_20
value: 27.044
- type: ndcg_at_3
value: 21.976000000000003
- type: ndcg_at_5
value: 23.813000000000002
- type: precision_at_1
value: 26.657999999999998
- type: precision_at_10
value: 5.864
- type: precision_at_100
value: 0.905
- type: precision_at_1000
value: 0.126
- type: precision_at_20
value: 3.4000000000000004
- type: precision_at_3
value: 14.135
- type: precision_at_5
value: 9.931
- type: recall_at_1
value: 13.328999999999999
- type: recall_at_10
value: 29.317999999999998
- type: recall_at_100
value: 45.253
- type: recall_at_1000
value: 63.187000000000005
- type: recall_at_20
value: 33.997
- type: recall_at_3
value: 21.201999999999998
- type: recall_at_5
value: 24.828
- task:
type: Retrieval
dataset:
name: MTEB HotpotQA (default)
type: mteb/hotpotqa
config: default
split: train
revision: ab518f4d6fcca38d87c25209f94beba119d02014
metrics:
- type: main_score
value: 27.361
- type: map_at_1
value: 15.187999999999999
- type: map_at_10
value: 21.143
- type: map_at_100
value: 21.955
- type: map_at_1000
value: 22.055
- type: map_at_20
value: 21.571
- type: map_at_3
value: 19.337
- type: map_at_5
value: 20.328
- type: mrr_at_1
value: 30.375294117647062
- type: mrr_at_10
value: 36.89110037348242
- type: mrr_at_100
value: 37.59161854635361
- type: mrr_at_1000
value: 37.65278972243383
- type: mrr_at_20
value: 37.28204270783653
- type: mrr_at_3
value: 35.073137254900395
- type: mrr_at_5
value: 36.10778431372484
- type: nauc_map_at_1000_diff1
value: 49.349859863214526
- type: nauc_map_at_1000_max
value: 41.88467513574528
- type: nauc_map_at_1000_std
value: 24.22933543144338
- type: nauc_map_at_100_diff1
value: 49.35940939023821
- type: nauc_map_at_100_max
value: 41.862590382078025
- type: nauc_map_at_100_std
value: 24.15434461176854
- type: nauc_map_at_10_diff1
value: 49.8538165767892
- type: nauc_map_at_10_max
value: 42.07384809374906
- type: nauc_map_at_10_std
value: 23.416303368507712
- type: nauc_map_at_1_diff1
value: 59.92575902483264
- type: nauc_map_at_1_max
value: 45.47700185129611
- type: nauc_map_at_1_std
value: 19.83388552453144
- type: nauc_map_at_20_diff1
value: 49.52738828169819
- type: nauc_map_at_20_max
value: 41.91046133599251
- type: nauc_map_at_20_std
value: 23.79441500917769
- type: nauc_map_at_3_diff1
value: 51.790844119427824
- type: nauc_map_at_3_max
value: 42.999352173314854
- type: nauc_map_at_3_std
value: 22.008776106004134
- type: nauc_map_at_5_diff1
value: 50.670375343301934
- type: nauc_map_at_5_max
value: 42.55434252366988
- type: nauc_map_at_5_std
value: 22.82261757618132
- type: nauc_mrr_at_1000_diff1
value: 54.67988721868572
- type: nauc_mrr_at_1000_max
value: 43.91336195991437
- type: nauc_mrr_at_1000_std
value: 22.158081050617664
- type: nauc_mrr_at_100_diff1
value: 54.66673307398132
- type: nauc_mrr_at_100_max
value: 43.90104349501937
- type: nauc_mrr_at_100_std
value: 22.149879738198813
- type: nauc_mrr_at_10_diff1
value: 54.80299708197367
- type: nauc_mrr_at_10_max
value: 43.97302396926416
- type: nauc_mrr_at_10_std
value: 21.93101088607793
- type: nauc_mrr_at_1_diff1
value: 59.92575902483264
- type: nauc_mrr_at_1_max
value: 45.47700185129611
- type: nauc_mrr_at_1_std
value: 19.83388552453144
- type: nauc_mrr_at_20_diff1
value: 54.69703296749322
- type: nauc_mrr_at_20_max
value: 43.91191360382579
- type: nauc_mrr_at_20_std
value: 22.0679377608011
- type: nauc_mrr_at_3_diff1
value: 55.6492096640659
- type: nauc_mrr_at_3_max
value: 44.39253185930357
- type: nauc_mrr_at_3_std
value: 21.23559827715352
- type: nauc_mrr_at_5_diff1
value: 55.15718817723955
- type: nauc_mrr_at_5_max
value: 44.21359056116808
- type: nauc_mrr_at_5_std
value: 21.689943072561064
- type: nauc_ndcg_at_1000_diff1
value: 47.08149592464777
- type: nauc_ndcg_at_1000_max
value: 41.4824024851649
- type: nauc_ndcg_at_1000_std
value: 28.290841883563324
- type: nauc_ndcg_at_100_diff1
value: 47.203241306354165
- type: nauc_ndcg_at_100_max
value: 41.01082594665584
- type: nauc_ndcg_at_100_std
value: 27.01050126741052
- type: nauc_ndcg_at_10_diff1
value: 48.8574562673927
- type: nauc_ndcg_at_10_max
value: 41.59700363422577
- type: nauc_ndcg_at_10_std
value: 24.302113156303708
- type: nauc_ndcg_at_1_diff1
value: 59.92575902483264
- type: nauc_ndcg_at_1_max
value: 45.47700185129611
- type: nauc_ndcg_at_1_std
value: 19.83388552453144
- type: nauc_ndcg_at_20_diff1
value: 48.01876180994126
- type: nauc_ndcg_at_20_max
value: 41.16103683691111
- type: nauc_ndcg_at_20_std
value: 25.22714091905253
- type: nauc_ndcg_at_3_diff1
value: 51.76534091516943
- type: nauc_ndcg_at_3_max
value: 43.066268249034806
- type: nauc_ndcg_at_3_std
value: 22.189204956774773
- type: nauc_ndcg_at_5_diff1
value: 50.274763573284176
- type: nauc_ndcg_at_5_max
value: 42.46612336859775
- type: nauc_ndcg_at_5_std
value: 23.298594291859136
- type: nauc_precision_at_1000_diff1
value: 21.060533862186702
- type: nauc_precision_at_1000_max
value: 28.16962240505211
- type: nauc_precision_at_1000_std
value: 39.87841510942972
- type: nauc_precision_at_100_diff1
value: 27.25075569238348
- type: nauc_precision_at_100_max
value: 29.229681139793666
- type: nauc_precision_at_100_std
value: 33.06692753126747
- type: nauc_precision_at_10_diff1
value: 37.35908709421307
- type: nauc_precision_at_10_max
value: 35.09768811571614
- type: nauc_precision_at_10_std
value: 26.319669286806963
- type: nauc_precision_at_1_diff1
value: 59.92575902483264
- type: nauc_precision_at_1_max
value: 45.47700185129611
- type: nauc_precision_at_1_std
value: 19.83388552453144
- type: nauc_precision_at_20_diff1
value: 34.02026815173258
- type: nauc_precision_at_20_max
value: 32.97155795749605
- type: nauc_precision_at_20_std
value: 28.446608120331838
- type: nauc_precision_at_3_diff1
value: 46.625368812766844
- type: nauc_precision_at_3_max
value: 41.143493529756945
- type: nauc_precision_at_3_std
value: 23.34446397266084
- type: nauc_precision_at_5_diff1
value: 42.596622234501666
- type: nauc_precision_at_5_max
value: 38.97638871265979
- type: nauc_precision_at_5_std
value: 25.081525286284535
- type: nauc_recall_at_1000_diff1
value: 21.060533862186706
- type: nauc_recall_at_1000_max
value: 28.16962240505217
- type: nauc_recall_at_1000_std
value: 39.878415109429795
- type: nauc_recall_at_100_diff1
value: 27.250755692383482
- type: nauc_recall_at_100_max
value: 29.229681139793694
- type: nauc_recall_at_100_std
value: 33.066927531267474
- type: nauc_recall_at_10_diff1
value: 37.35908709421306
- type: nauc_recall_at_10_max
value: 35.09768811571614
- type: nauc_recall_at_10_std
value: 26.319669286807006
- type: nauc_recall_at_1_diff1
value: 59.92575902483264
- type: nauc_recall_at_1_max
value: 45.47700185129611
- type: nauc_recall_at_1_std
value: 19.83388552453144
- type: nauc_recall_at_20_diff1
value: 34.02026815173259
- type: nauc_recall_at_20_max
value: 32.97155795749605
- type: nauc_recall_at_20_std
value: 28.446608120331906
- type: nauc_recall_at_3_diff1
value: 46.62536881276683
- type: nauc_recall_at_3_max
value: 41.14349352975691
- type: nauc_recall_at_3_std
value: 23.344463972660805
- type: nauc_recall_at_5_diff1
value: 42.59662223450168
- type: nauc_recall_at_5_max
value: 38.97638871265983
- type: nauc_recall_at_5_std
value: 25.081525286284567
- type: ndcg_at_1
value: 30.375000000000004
- type: ndcg_at_10
value: 27.361
- type: ndcg_at_100
value: 31.247000000000003
- type: ndcg_at_1000
value: 33.785
- type: ndcg_at_20
value: 28.755999999999997
- type: ndcg_at_3
value: 23.905
- type: ndcg_at_5
value: 25.570999999999998
- type: precision_at_1
value: 30.375000000000004
- type: precision_at_10
value: 6.072
- type: precision_at_100
value: 0.9199999999999999
- type: precision_at_1000
value: 0.126
- type: precision_at_20
value: 3.486
- type: precision_at_3
value: 15.033
- type: precision_at_5
value: 10.34
- type: recall_at_1
value: 15.187999999999999
- type: recall_at_10
value: 30.359
- type: recall_at_100
value: 45.985
- type: recall_at_1000
value: 63.001
- type: recall_at_20
value: 34.861
- type: recall_at_3
value: 22.55
- type: recall_at_5
value: 25.851000000000003
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO (default)
type: mteb/msmarco
config: default
split: dev
revision: c5a29a104738b98a9e76336939199e264163d4a0
metrics:
- type: main_score
value: 34.803
- type: map_at_1
value: 18.05
- type: map_at_10
value: 28.698
- type: map_at_100
value: 29.87
- type: map_at_1000
value: 29.937
- type: map_at_20
value: 29.407
- type: map_at_3
value: 25.211
- type: map_at_5
value: 27.256999999999998
- type: mrr_at_1
value: 18.595988538681947
- type: mrr_at_10
value: 29.313929822167506
- type: mrr_at_100
value: 30.43456132167563
- type: mrr_at_1000
value: 30.496264584797384
- type: mrr_at_20
value: 29.994260130208367
- type: mrr_at_3
value: 25.847659980897618
- type: mrr_at_5
value: 27.90926456542496
- type: nauc_map_at_1000_diff1
value: 30.967805122512292
- type: nauc_map_at_1000_max
value: 9.508356994054424
- type: nauc_map_at_1000_std
value: -16.7647780708821
- type: nauc_map_at_100_diff1
value: 30.959590208362474
- type: nauc_map_at_100_max
value: 9.519514909485395
- type: nauc_map_at_100_std
value: -16.752225883753628
- type: nauc_map_at_10_diff1
value: 30.961638427898
- type: nauc_map_at_10_max
value: 9.032394524880466
- type: nauc_map_at_10_std
value: -17.58469702660575
- type: nauc_map_at_1_diff1
value: 36.68684868509966
- type: nauc_map_at_1_max
value: 8.006788300165026
- type: nauc_map_at_1_std
value: -16.453929215948566
- type: nauc_map_at_20_diff1
value: 30.914160804266828
- type: nauc_map_at_20_max
value: 9.383047052491237
- type: nauc_map_at_20_std
value: -17.09908974160602
- type: nauc_map_at_3_diff1
value: 32.00864052480307
- type: nauc_map_at_3_max
value: 7.9346719904912755
- type: nauc_map_at_3_std
value: -18.363334919131873
- type: nauc_map_at_5_diff1
value: 31.30004793090547
- type: nauc_map_at_5_max
value: 8.32231303754107
- type: nauc_map_at_5_std
value: -18.244863888195106
- type: nauc_mrr_at_1000_diff1
value: 30.925580995966435
- type: nauc_mrr_at_1000_max
value: 9.570609811089142
- type: nauc_mrr_at_1000_std
value: -16.416887518960856
- type: nauc_mrr_at_100_diff1
value: 30.914394357734505
- type: nauc_mrr_at_100_max
value: 9.586927921368241
- type: nauc_mrr_at_100_std
value: -16.39712616782138
- type: nauc_mrr_at_10_diff1
value: 30.92742800307562
- type: nauc_mrr_at_10_max
value: 9.197220826362587
- type: nauc_mrr_at_10_std
value: -17.151747008284875
- type: nauc_mrr_at_1_diff1
value: 36.60106731450987
- type: nauc_mrr_at_1_max
value: 8.037370217799388
- type: nauc_mrr_at_1_std
value: -16.29920870158755
- type: nauc_mrr_at_20_diff1
value: 30.873870247447528
- type: nauc_mrr_at_20_max
value: 9.489994351963741
- type: nauc_mrr_at_20_std
value: -16.695446595293273
- type: nauc_mrr_at_3_diff1
value: 31.962676684893193
- type: nauc_mrr_at_3_max
value: 7.902980731667526
- type: nauc_mrr_at_3_std
value: -18.087972700882023
- type: nauc_mrr_at_5_diff1
value: 31.266029079987334
- type: nauc_mrr_at_5_max
value: 8.43829728480481
- type: nauc_mrr_at_5_std
value: -17.830313089048254
- type: nauc_ndcg_at_1000_diff1
value: 28.800055526073052
- type: nauc_ndcg_at_1000_max
value: 12.150167860614443
- type: nauc_ndcg_at_1000_std
value: -13.25905395516576
- type: nauc_ndcg_at_100_diff1
value: 28.55273062021815
- type: nauc_ndcg_at_100_max
value: 12.604290774389284
- type: nauc_ndcg_at_100_std
value: -12.640286042753369
- type: nauc_ndcg_at_10_diff1
value: 28.656702169882305
- type: nauc_ndcg_at_10_max
value: 10.390975552193304
- type: nauc_ndcg_at_10_std
value: -16.979914789104274
- type: nauc_ndcg_at_1_diff1
value: 36.60106731450987
- type: nauc_ndcg_at_1_max
value: 7.999297342297694
- type: nauc_ndcg_at_1_std
value: -16.37844144249649
- type: nauc_ndcg_at_20_diff1
value: 28.39512508390008
- type: nauc_ndcg_at_20_max
value: 11.670038381067469
- type: nauc_ndcg_at_20_std
value: -15.19270139716777
- type: nauc_ndcg_at_3_diff1
value: 30.720892075394424
- type: nauc_ndcg_at_3_max
value: 7.944044047190646
- type: nauc_ndcg_at_3_std
value: -18.776183934337002
- type: nauc_ndcg_at_5_diff1
value: 29.499593527403515
- type: nauc_ndcg_at_5_max
value: 8.70632138099665
- type: nauc_ndcg_at_5_std
value: -18.516401671537196
- type: nauc_precision_at_1000_diff1
value: -0.8237646361863551
- type: nauc_precision_at_1000_max
value: 29.65949086980757
- type: nauc_precision_at_1000_std
value: 27.545476721494193
- type: nauc_precision_at_100_diff1
value: 12.284529508091742
- type: nauc_precision_at_100_max
value: 29.413232419844444
- type: nauc_precision_at_100_std
value: 17.196307006886304
- type: nauc_precision_at_10_diff1
value: 21.3879709585264
- type: nauc_precision_at_10_max
value: 14.806399276392005
- type: nauc_precision_at_10_std
value: -13.89189147461678
- type: nauc_precision_at_1_diff1
value: 36.60106731450987
- type: nauc_precision_at_1_max
value: 7.999297342297694
- type: nauc_precision_at_1_std
value: -16.37844144249649
- type: nauc_precision_at_20_diff1
value: 18.793745272557334
- type: nauc_precision_at_20_max
value: 20.52958902353019
- type: nauc_precision_at_20_std
value: -5.831301138585665
- type: nauc_precision_at_3_diff1
value: 27.51302998292356
- type: nauc_precision_at_3_max
value: 8.124801342511773
- type: nauc_precision_at_3_std
value: -19.653828947390405
- type: nauc_precision_at_5_diff1
value: 24.648428234894187
- type: nauc_precision_at_5_max
value: 9.970901593970794
- type: nauc_precision_at_5_std
value: -18.870802823554662
- type: nauc_recall_at_1000_diff1
value: 3.1535887573039503
- type: nauc_recall_at_1000_max
value: 49.483153585320665
- type: nauc_recall_at_1000_std
value: 48.068469198499955
- type: nauc_recall_at_100_diff1
value: 16.09114265175475
- type: nauc_recall_at_100_max
value: 31.63592256685488
- type: nauc_recall_at_100_std
value: 17.262073141805402
- type: nauc_recall_at_10_diff1
value: 21.74285904516543
- type: nauc_recall_at_10_max
value: 14.233113457215413
- type: nauc_recall_at_10_std
value: -15.00141123225369
- type: nauc_recall_at_1_diff1
value: 36.68684868509966
- type: nauc_recall_at_1_max
value: 8.006788300165026
- type: nauc_recall_at_1_std
value: -16.453929215948566
- type: nauc_recall_at_20_diff1
value: 19.721186534513574
- type: nauc_recall_at_20_max
value: 19.839943733839586
- type: nauc_recall_at_20_std
value: -7.528794529976447
- type: nauc_recall_at_3_diff1
value: 27.150549688069308
- type: nauc_recall_at_3_max
value: 7.915443441945287
- type: nauc_recall_at_3_std
value: -19.87191884581972
- type: nauc_recall_at_5_diff1
value: 24.49887160612282
- type: nauc_recall_at_5_max
value: 9.52108917622074
- type: nauc_recall_at_5_std
value: -19.231477971502706
- type: ndcg_at_1
value: 18.596
- type: ndcg_at_10
value: 34.803
- type: ndcg_at_100
value: 40.544000000000004
- type: ndcg_at_1000
value: 42.260999999999996
- type: ndcg_at_20
value: 37.32
- type: ndcg_at_3
value: 27.71
- type: ndcg_at_5
value: 31.385999999999996
- type: precision_at_1
value: 18.596
- type: precision_at_10
value: 5.595
- type: precision_at_100
value: 0.848
- type: precision_at_1000
value: 0.099
- type: precision_at_20
value: 3.32
- type: precision_at_3
value: 11.858
- type: precision_at_5
value: 8.988999999999999
- type: recall_at_1
value: 18.05
- type: recall_at_10
value: 53.516
- type: recall_at_100
value: 80.289
- type: recall_at_1000
value: 93.512
- type: recall_at_20
value: 63.288999999999994
- type: recall_at_3
value: 34.306
- type: recall_at_5
value: 43.145
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO (default)
type: mteb/msmarco
config: default
split: test
revision: c5a29a104738b98a9e76336939199e264163d4a0
metrics:
- type: main_score
value: 59.06400000000001
- type: map_at_1
value: 1.69
- type: map_at_10
value: 10.755
- type: map_at_100
value: 28.62
- type: map_at_1000
value: 34.429
- type: map_at_20
value: 16.674
- type: map_at_3
value: 4.005
- type: map_at_5
value: 6.332
- type: mrr_at_1
value: 88.37209302325581
- type: mrr_at_10
value: 91.47286821705424
- type: mrr_at_100
value: 91.56261235995456
- type: mrr_at_1000
value: 91.56261235995456
- type: mrr_at_20
value: 91.47286821705424
- type: mrr_at_3
value: 91.47286821705424
- type: mrr_at_5
value: 91.47286821705424
- type: nauc_map_at_1000_diff1
value: -58.004048824154644
- type: nauc_map_at_1000_max
value: 34.44700621046442
- type: nauc_map_at_1000_std
value: 66.90595473840271
- type: nauc_map_at_100_diff1
value: -45.770189406894886
- type: nauc_map_at_100_max
value: 26.211654802825322
- type: nauc_map_at_100_std
value: 56.080139344538956
- type: nauc_map_at_10_diff1
value: -10.97871068331086
- type: nauc_map_at_10_max
value: 28.539368064747517
- type: nauc_map_at_10_std
value: 21.028635665204536
- type: nauc_map_at_1_diff1
value: 35.97114363073452
- type: nauc_map_at_1_max
value: -10.574909183137414
- type: nauc_map_at_1_std
value: -16.11433179567385
- type: nauc_map_at_20_diff1
value: -17.439484208262535
- type: nauc_map_at_20_max
value: 25.1778779774871
- type: nauc_map_at_20_std
value: 28.832377812578258
- type: nauc_map_at_3_diff1
value: 7.501651006902326
- type: nauc_map_at_3_max
value: 19.130202694438417
- type: nauc_map_at_3_std
value: 6.940777942691115
- type: nauc_map_at_5_diff1
value: -1.9207947577929207
- type: nauc_map_at_5_max
value: 24.72479488530524
- type: nauc_map_at_5_std
value: 12.662402148436808
- type: nauc_mrr_at_1000_diff1
value: -31.38940176778026
- type: nauc_mrr_at_1000_max
value: 27.91929945678086
- type: nauc_mrr_at_1000_std
value: 63.98020520396501
- type: nauc_mrr_at_100_diff1
value: -31.38940176778026
- type: nauc_mrr_at_100_max
value: 27.91929945678086
- type: nauc_mrr_at_100_std
value: 63.98020520396501
- type: nauc_mrr_at_10_diff1
value: -31.98154638739888
- type: nauc_mrr_at_10_max
value: 28.691575876582025
- type: nauc_mrr_at_10_std
value: 64.15203937521413
- type: nauc_mrr_at_1_diff1
value: -18.33986985937143
- type: nauc_mrr_at_1_max
value: -3.5658506202108935
- type: nauc_mrr_at_1_std
value: 57.265329172858316
- type: nauc_mrr_at_20_diff1
value: -31.98154638739888
- type: nauc_mrr_at_20_max
value: 28.691575876582025
- type: nauc_mrr_at_20_std
value: 64.15203937521413
- type: nauc_mrr_at_3_diff1
value: -31.98154638739888
- type: nauc_mrr_at_3_max
value: 28.691575876582025
- type: nauc_mrr_at_3_std
value: 64.15203937521413
- type: nauc_mrr_at_5_diff1
value: -31.98154638739888
- type: nauc_mrr_at_5_max
value: 28.691575876582025
- type: nauc_mrr_at_5_std
value: 64.15203937521413
- type: nauc_ndcg_at_1000_diff1
value: -64.4009360298104
- type: nauc_ndcg_at_1000_max
value: 47.73173186354194
- type: nauc_ndcg_at_1000_std
value: 67.76224269862708
- type: nauc_ndcg_at_100_diff1
value: -55.81170448144629
- type: nauc_ndcg_at_100_max
value: 45.76842611746515
- type: nauc_ndcg_at_100_std
value: 65.75760535333559
- type: nauc_ndcg_at_10_diff1
value: -48.26548478088951
- type: nauc_ndcg_at_10_max
value: 47.221554778624736
- type: nauc_ndcg_at_10_std
value: 48.66818342066006
- type: nauc_ndcg_at_1_diff1
value: 22.361573521914167
- type: nauc_ndcg_at_1_max
value: -4.095691143820581
- type: nauc_ndcg_at_1_std
value: -22.86231879814137
- type: nauc_ndcg_at_20_diff1
value: -52.771674683458514
- type: nauc_ndcg_at_20_max
value: 39.94057642569563
- type: nauc_ndcg_at_20_std
value: 55.44432532491158
- type: nauc_ndcg_at_3_diff1
value: -15.072453465017949
- type: nauc_ndcg_at_3_max
value: 36.22918826732502
- type: nauc_ndcg_at_3_std
value: 16.823145415489098
- type: nauc_ndcg_at_5_diff1
value: -32.572693237589036
- type: nauc_ndcg_at_5_max
value: 44.30557738022862
- type: nauc_ndcg_at_5_std
value: 30.75861792737064
- type: nauc_precision_at_1000_diff1
value: -61.90365106233373
- type: nauc_precision_at_1000_max
value: 29.37450084944317
- type: nauc_precision_at_1000_std
value: 51.15294977255612
- type: nauc_precision_at_100_diff1
value: -64.87811325128291
- type: nauc_precision_at_100_max
value: 27.082351018049334
- type: nauc_precision_at_100_std
value: 59.21347841222029
- type: nauc_precision_at_10_diff1
value: -73.17039435952445
- type: nauc_precision_at_10_max
value: 50.03963485425603
- type: nauc_precision_at_10_std
value: 74.7760482349427
- type: nauc_precision_at_1_diff1
value: -18.33986985937143
- type: nauc_precision_at_1_max
value: -3.5658506202108935
- type: nauc_precision_at_1_std
value: 57.265329172858316
- type: nauc_precision_at_20_diff1
value: -69.19781045685487
- type: nauc_precision_at_20_max
value: 35.34563871429891
- type: nauc_precision_at_20_std
value: 69.73611685460025
- type: nauc_precision_at_3_diff1
value: -65.76154675431599
- type: nauc_precision_at_3_max
value: 58.774704910679354
- type: nauc_precision_at_3_std
value: 81.26148034443878
- type: nauc_precision_at_5_diff1
value: -78.31465612075088
- type: nauc_precision_at_5_max
value: 59.02550474452744
- type: nauc_precision_at_5_std
value: 83.37770095414346
- type: nauc_recall_at_1000_diff1
value: -71.70183001436035
- type: nauc_recall_at_1000_max
value: 39.784139396042576
- type: nauc_recall_at_1000_std
value: 69.52333709786
- type: nauc_recall_at_100_diff1
value: -39.867533213723476
- type: nauc_recall_at_100_max
value: 25.88219867077632
- type: nauc_recall_at_100_std
value: 50.413431837035105
- type: nauc_recall_at_10_diff1
value: -4.726447649175431
- type: nauc_recall_at_10_max
value: 27.630825828715384
- type: nauc_recall_at_10_std
value: 14.42376069507755
- type: nauc_recall_at_1_diff1
value: 35.97114363073452
- type: nauc_recall_at_1_max
value: -10.574909183137414
- type: nauc_recall_at_1_std
value: -16.11433179567385
- type: nauc_recall_at_20_diff1
value: -8.787671293237674
- type: nauc_recall_at_20_max
value: 25.207984756226242
- type: nauc_recall_at_20_std
value: 20.728749504224318
- type: nauc_recall_at_3_diff1
value: 9.441972498951067
- type: nauc_recall_at_3_max
value: 22.118880183280638
- type: nauc_recall_at_3_std
value: 4.297339270917102
- type: nauc_recall_at_5_diff1
value: 4.072468434627269
- type: nauc_recall_at_5_max
value: 24.12093962060067
- type: nauc_recall_at_5_std
value: 6.723457577572802
- type: ndcg_at_1
value: 68.217
- type: ndcg_at_10
value: 59.06400000000001
- type: ndcg_at_100
value: 51.28
- type: ndcg_at_1000
value: 57.745999999999995
- type: ndcg_at_20
value: 55.969
- type: ndcg_at_3
value: 63.207
- type: ndcg_at_5
value: 62.099000000000004
- type: precision_at_1
value: 88.372
- type: precision_at_10
value: 68.372
- type: precision_at_100
value: 31.349
- type: precision_at_1000
value: 5.858
- type: precision_at_20
value: 59.767
- type: precision_at_3
value: 77.519
- type: precision_at_5
value: 75.81400000000001
- type: recall_at_1
value: 1.69
- type: recall_at_10
value: 11.779
- type: recall_at_100
value: 38.83
- type: recall_at_1000
value: 62.572
- type: recall_at_20
value: 19.155
- type: recall_at_3
value: 4.143
- type: recall_at_5
value: 6.801
- task:
type: Retrieval
dataset:
name: MTEB MSMARCO (default)
type: mteb/msmarco
config: default
split: train
revision: c5a29a104738b98a9e76336939199e264163d4a0
metrics:
- type: main_score
value: 33.715
- type: map_at_1
value: 16.719
- type: map_at_10
value: 27.433000000000003
- type: map_at_100
value: 28.703
- type: map_at_1000
value: 28.767
- type: map_at_20
value: 28.191
- type: map_at_3
value: 23.826
- type: map_at_5
value: 25.849
- type: mrr_at_1
value: 17.20904523212557
- type: mrr_at_10
value: 27.92367929563414
- type: mrr_at_100
value: 29.152633780765758
- type: mrr_at_1000
value: 29.21099712739743
- type: mrr_at_20
value: 28.66210180739731
- type: mrr_at_3
value: 24.356466026020737
- type: mrr_at_5
value: 26.365397526652018
- type: nauc_map_at_1000_diff1
value: 29.63353838791719
- type: nauc_map_at_1000_max
value: 5.7454487966009555
- type: nauc_map_at_1000_std
value: -18.063069395275722
- type: nauc_map_at_100_diff1
value: 29.616549284785748
- type: nauc_map_at_100_max
value: 5.760919702571223
- type: nauc_map_at_100_std
value: -18.044574507278362
- type: nauc_map_at_10_diff1
value: 29.700954720537165
- type: nauc_map_at_10_max
value: 5.424404826187661
- type: nauc_map_at_10_std
value: -18.842507893783537
- type: nauc_map_at_1_diff1
value: 35.211396613278865
- type: nauc_map_at_1_max
value: 3.1343744942229086
- type: nauc_map_at_1_std
value: -18.170256518161427
- type: nauc_map_at_20_diff1
value: 29.613406086113486
- type: nauc_map_at_20_max
value: 5.641627873928245
- type: nauc_map_at_20_std
value: -18.396499430312748
- type: nauc_map_at_3_diff1
value: 30.603238239884483
- type: nauc_map_at_3_max
value: 4.453242887818738
- type: nauc_map_at_3_std
value: -19.268024232660473
- type: nauc_map_at_5_diff1
value: 30.003625205143397
- type: nauc_map_at_5_max
value: 4.969401253866233
- type: nauc_map_at_5_std
value: -19.276106426526184
- type: nauc_mrr_at_1000_diff1
value: 29.486572453576716
- type: nauc_mrr_at_1000_max
value: 5.693116182861802
- type: nauc_mrr_at_1000_std
value: -17.917227560489565
- type: nauc_mrr_at_100_diff1
value: 29.468797895964215
- type: nauc_mrr_at_100_max
value: 5.710921844344999
- type: nauc_mrr_at_100_std
value: -17.89511910086844
- type: nauc_mrr_at_10_diff1
value: 29.53746647566686
- type: nauc_mrr_at_10_max
value: 5.406885859896506
- type: nauc_mrr_at_10_std
value: -18.644478337030773
- type: nauc_mrr_at_1_diff1
value: 34.91427458267835
- type: nauc_mrr_at_1_max
value: 3.2309227517846852
- type: nauc_mrr_at_1_std
value: -18.080446645652778
- type: nauc_mrr_at_20_diff1
value: 29.45773578143679
- type: nauc_mrr_at_20_max
value: 5.60850730413169
- type: nauc_mrr_at_20_std
value: -18.21575909250342
- type: nauc_mrr_at_3_diff1
value: 30.40915710544333
- type: nauc_mrr_at_3_max
value: 4.474773651459147
- type: nauc_mrr_at_3_std
value: -19.11200967355029
- type: nauc_mrr_at_5_diff1
value: 29.817993784330003
- type: nauc_mrr_at_5_max
value: 4.973386662596821
- type: nauc_mrr_at_5_std
value: -19.08901246396465
- type: nauc_ndcg_at_1000_diff1
value: 27.742972640187336
- type: nauc_ndcg_at_1000_max
value: 7.916917003519648
- type: nauc_ndcg_at_1000_std
value: -15.17896601259802
- type: nauc_ndcg_at_100_diff1
value: 27.238853230669203
- type: nauc_ndcg_at_100_max
value: 8.460940793652366
- type: nauc_ndcg_at_100_std
value: -14.261896738662886
- type: nauc_ndcg_at_10_diff1
value: 27.66904868796311
- type: nauc_ndcg_at_10_max
value: 6.735176820125588
- type: nauc_ndcg_at_10_std
value: -18.448078157577832
- type: nauc_ndcg_at_1_diff1
value: 34.95550568195721
- type: nauc_ndcg_at_1_max
value: 3.2125833396044503
- type: nauc_ndcg_at_1_std
value: -18.10007817354091
- type: nauc_ndcg_at_20_diff1
value: 27.301933503569487
- type: nauc_ndcg_at_20_max
value: 7.536489988333836
- type: nauc_ndcg_at_20_std
value: -16.842402929106534
- type: nauc_ndcg_at_3_diff1
value: 29.372635951851155
- type: nauc_ndcg_at_3_max
value: 4.820361216791387
- type: nauc_ndcg_at_3_std
value: -19.513367064010172
- type: nauc_ndcg_at_5_diff1
value: 28.37372030928328
- type: nauc_ndcg_at_5_max
value: 5.678798799104905
- type: nauc_ndcg_at_5_std
value: -19.501286961437117
- type: nauc_precision_at_1000_diff1
value: -3.510350547767326
- type: nauc_precision_at_1000_max
value: 21.377749750431178
- type: nauc_precision_at_1000_std
value: 22.27774613662246
- type: nauc_precision_at_100_diff1
value: 8.87592055675654
- type: nauc_precision_at_100_max
value: 23.073116570870873
- type: nauc_precision_at_100_std
value: 15.055227218558281
- type: nauc_precision_at_10_diff1
value: 21.052874039568565
- type: nauc_precision_at_10_max
value: 10.660428704804987
- type: nauc_precision_at_10_std
value: -16.430957222899334
- type: nauc_precision_at_1_diff1
value: 34.95550568195721
- type: nauc_precision_at_1_max
value: 3.2125833396044503
- type: nauc_precision_at_1_std
value: -18.10007817354091
- type: nauc_precision_at_20_diff1
value: 17.989882712589928
- type: nauc_precision_at_20_max
value: 14.07438846286306
- type: nauc_precision_at_20_std
value: -9.440997023749803
- type: nauc_precision_at_3_diff1
value: 26.086905823987642
- type: nauc_precision_at_3_max
value: 5.8656016048393225
- type: nauc_precision_at_3_std
value: -20.040710978552877
- type: nauc_precision_at_5_diff1
value: 23.81277688113566
- type: nauc_precision_at_5_max
value: 7.673783195436691
- type: nauc_precision_at_5_std
value: -19.793547751043462
- type: nauc_recall_at_1000_diff1
value: 2.1874443524235154
- type: nauc_recall_at_1000_max
value: 51.05037398799104
- type: nauc_recall_at_1000_std
value: 53.35202253619663
- type: nauc_recall_at_100_diff1
value: 13.542586793678119
- type: nauc_recall_at_100_max
value: 27.18200902894963
- type: nauc_recall_at_100_std
value: 16.971437674746966
- type: nauc_recall_at_10_diff1
value: 21.973446928460568
- type: nauc_recall_at_10_max
value: 10.41568392262844
- type: nauc_recall_at_10_std
value: -16.967676205416033
- type: nauc_recall_at_1_diff1
value: 35.211396613278865
- type: nauc_recall_at_1_max
value: 3.1343744942229086
- type: nauc_recall_at_1_std
value: -18.170256518161427
- type: nauc_recall_at_20_diff1
value: 19.711023045178955
- type: nauc_recall_at_20_max
value: 14.095426841286946
- type: nauc_recall_at_20_std
value: -10.417724099405264
- type: nauc_recall_at_3_diff1
value: 26.246688921271716
- type: nauc_recall_at_3_max
value: 5.649635052724743
- type: nauc_recall_at_3_std
value: -20.04554246302986
- type: nauc_recall_at_5_diff1
value: 24.168809764476798
- type: nauc_recall_at_5_max
value: 7.399478058484081
- type: nauc_recall_at_5_std
value: -19.9514394901081
- type: ndcg_at_1
value: 17.201
- type: ndcg_at_10
value: 33.715
- type: ndcg_at_100
value: 40.025
- type: ndcg_at_1000
value: 41.685
- type: ndcg_at_20
value: 36.424
- type: ndcg_at_3
value: 26.31
- type: ndcg_at_5
value: 29.93
- type: precision_at_1
value: 17.201
- type: precision_at_10
value: 5.541
- type: precision_at_100
value: 0.8699999999999999
- type: precision_at_1000
value: 0.101
- type: precision_at_20
value: 3.329
- type: precision_at_3
value: 11.37
- type: precision_at_5
value: 8.649
- type: recall_at_1
value: 16.719
- type: recall_at_10
value: 53.248
- type: recall_at_100
value: 82.863
- type: recall_at_1000
value: 95.721
- type: recall_at_20
value: 63.797000000000004
- type: recall_at_3
value: 32.984
- type: recall_at_5
value: 41.702
- task:
type: Reranking
dataset:
name: MTEB MindSmallReranking (default)
type: mteb/mind_small
config: default
split: test
revision: 59042f120c80e8afa9cdbb224f67076cec0fc9a7
metrics:
- type: main_score
value: 31.370055361148335
- type: map
value: 31.370055361148335
- type: mrr
value: 32.37440631143583
- type: nAUC_map_diff1
value: 14.695683399887827
- type: nAUC_map_max
value: -23.4492856197865
- type: nAUC_map_std
value: -8.885861431808289
- type: nAUC_mrr_diff1
value: 13.34849840135066
- type: nAUC_mrr_max
value: -18.199354674002286
- type: nAUC_mrr_std
value: -6.5654110123418565
- task:
type: Retrieval
dataset:
name: MTEB NFCorpus (default)
type: mteb/nfcorpus
config: default
split: test
revision: ec0fa4fe99da2ff19ca1214b7966684033a58814
metrics:
- type: main_score
value: 29.754
- type: map_at_1
value: 4.567
- type: map_at_10
value: 10.027999999999999
- type: map_at_100
value: 13.431000000000001
- type: map_at_1000
value: 14.92
- type: map_at_20
value: 11.431
- type: map_at_3
value: 7.224
- type: map_at_5
value: 8.386000000000001
- type: mrr_at_1
value: 37.77089783281733
- type: mrr_at_10
value: 47.909725293626224
- type: mrr_at_100
value: 48.65139591019661
- type: mrr_at_1000
value: 48.69916406344342
- type: mrr_at_20
value: 48.367288102536705
- type: mrr_at_3
value: 46.49122807017543
- type: mrr_at_5
value: 47.0639834881321
- type: nauc_map_at_1000_diff1
value: 14.085748354172512
- type: nauc_map_at_1000_max
value: 28.172950420215592
- type: nauc_map_at_1000_std
value: 18.144213868106675
- type: nauc_map_at_100_diff1
value: 14.589976235435408
- type: nauc_map_at_100_max
value: 26.576946199479153
- type: nauc_map_at_100_std
value: 13.843137874037378
- type: nauc_map_at_10_diff1
value: 17.8495149344614
- type: nauc_map_at_10_max
value: 18.82526312062093
- type: nauc_map_at_10_std
value: 2.86785487558237
- type: nauc_map_at_1_diff1
value: 28.714904063353853
- type: nauc_map_at_1_max
value: -3.169789138203512
- type: nauc_map_at_1_std
value: -15.236185778974088
- type: nauc_map_at_20_diff1
value: 15.794494377362259
- type: nauc_map_at_20_max
value: 22.383385559577366
- type: nauc_map_at_20_std
value: 7.50366878545232
- type: nauc_map_at_3_diff1
value: 20.770182594071223
- type: nauc_map_at_3_max
value: 9.50930632250425
- type: nauc_map_at_3_std
value: -7.066408370212426
- type: nauc_map_at_5_diff1
value: 20.197174149206827
- type: nauc_map_at_5_max
value: 14.021867267435159
- type: nauc_map_at_5_std
value: -3.409836969018499
- type: nauc_mrr_at_1000_diff1
value: 22.986158985525027
- type: nauc_mrr_at_1000_max
value: 37.185817395515556
- type: nauc_mrr_at_1000_std
value: 23.706769752670958
- type: nauc_mrr_at_100_diff1
value: 22.94515989033249
- type: nauc_mrr_at_100_max
value: 37.209904708113726
- type: nauc_mrr_at_100_std
value: 23.746222614427094
- type: nauc_mrr_at_10_diff1
value: 22.81681858431028
- type: nauc_mrr_at_10_max
value: 37.10239557383066
- type: nauc_mrr_at_10_std
value: 23.663887285644442
- type: nauc_mrr_at_1_diff1
value: 27.148124088268162
- type: nauc_mrr_at_1_max
value: 26.428295034510924
- type: nauc_mrr_at_1_std
value: 11.84295790231629
- type: nauc_mrr_at_20_diff1
value: 22.946866772785427
- type: nauc_mrr_at_20_max
value: 37.38173901985591
- type: nauc_mrr_at_20_std
value: 23.776573546658362
- type: nauc_mrr_at_3_diff1
value: 23.563086894697765
- type: nauc_mrr_at_3_max
value: 36.719834088530625
- type: nauc_mrr_at_3_std
value: 22.817102267109433
- type: nauc_mrr_at_5_diff1
value: 23.131059142837216
- type: nauc_mrr_at_5_max
value: 36.731326253130234
- type: nauc_mrr_at_5_std
value: 23.069191250755768
- type: nauc_ndcg_at_1000_diff1
value: 15.95138362626174
- type: nauc_ndcg_at_1000_max
value: 44.33022365744542
- type: nauc_ndcg_at_1000_std
value: 35.72521592020635
- type: nauc_ndcg_at_100_diff1
value: 14.575407179245719
- type: nauc_ndcg_at_100_max
value: 36.95692885691867
- type: nauc_ndcg_at_100_std
value: 26.80789282599495
- type: nauc_ndcg_at_10_diff1
value: 15.594400206553201
- type: nauc_ndcg_at_10_max
value: 35.348349357904375
- type: nauc_ndcg_at_10_std
value: 25.492218422553346
- type: nauc_ndcg_at_1_diff1
value: 28.078477674957707
- type: nauc_ndcg_at_1_max
value: 22.554521470143378
- type: nauc_ndcg_at_1_std
value: 11.86138448259433
- type: nauc_ndcg_at_20_diff1
value: 13.880875656746857
- type: nauc_ndcg_at_20_max
value: 34.45274753229036
- type: nauc_ndcg_at_20_std
value: 25.804360762917444
- type: nauc_ndcg_at_3_diff1
value: 16.85604207727626
- type: nauc_ndcg_at_3_max
value: 32.72858990186108
- type: nauc_ndcg_at_3_std
value: 19.852261713002537
- type: nauc_ndcg_at_5_diff1
value: 16.70613670641336
- type: nauc_ndcg_at_5_max
value: 35.398452792018126
- type: nauc_ndcg_at_5_std
value: 22.547276511653237
- type: nauc_precision_at_1000_diff1
value: -3.9205652143854004
- type: nauc_precision_at_1000_max
value: 11.56760575704433
- type: nauc_precision_at_1000_std
value: 29.969843364807815
- type: nauc_precision_at_100_diff1
value: -3.2791504425228504
- type: nauc_precision_at_100_max
value: 23.105292891322655
- type: nauc_precision_at_100_std
value: 38.24772180179586
- type: nauc_precision_at_10_diff1
value: 4.077038954732221
- type: nauc_precision_at_10_max
value: 37.97114306278218
- type: nauc_precision_at_10_std
value: 35.33785209505845
- type: nauc_precision_at_1_diff1
value: 27.148124088268162
- type: nauc_precision_at_1_max
value: 26.428295034510924
- type: nauc_precision_at_1_std
value: 11.84295790231629
- type: nauc_precision_at_20_diff1
value: -0.07715823284655204
- type: nauc_precision_at_20_max
value: 32.32150505807481
- type: nauc_precision_at_20_std
value: 35.93434300391347
- type: nauc_precision_at_3_diff1
value: 7.47311429058567
- type: nauc_precision_at_3_max
value: 38.53868770942289
- type: nauc_precision_at_3_std
value: 25.65319420988019
- type: nauc_precision_at_5_diff1
value: 6.8117959185149655
- type: nauc_precision_at_5_max
value: 40.78936637790633
- type: nauc_precision_at_5_std
value: 29.29130912546785
- type: nauc_recall_at_1000_diff1
value: 4.8295758031310285
- type: nauc_recall_at_1000_max
value: 21.69239465610227
- type: nauc_recall_at_1000_std
value: 18.948672947229195
- type: nauc_recall_at_100_diff1
value: 7.988051122946883
- type: nauc_recall_at_100_max
value: 22.98766591265696
- type: nauc_recall_at_100_std
value: 15.290463328578074
- type: nauc_recall_at_10_diff1
value: 13.365133585158528
- type: nauc_recall_at_10_max
value: 20.732053918735378
- type: nauc_recall_at_10_std
value: 6.370291758470967
- type: nauc_recall_at_1_diff1
value: 28.714904063353853
- type: nauc_recall_at_1_max
value: -3.169789138203512
- type: nauc_recall_at_1_std
value: -15.236185778974088
- type: nauc_recall_at_20_diff1
value: 8.20907091382773
- type: nauc_recall_at_20_max
value: 21.274336782803303
- type: nauc_recall_at_20_std
value: 11.112201780365108
- type: nauc_recall_at_3_diff1
value: 18.540225578507606
- type: nauc_recall_at_3_max
value: 15.389377575957791
- type: nauc_recall_at_3_std
value: -3.7988750765195216
- type: nauc_recall_at_5_diff1
value: 16.698902011414855
- type: nauc_recall_at_5_max
value: 17.822073045346276
- type: nauc_recall_at_5_std
value: -0.07358971200090962
- type: ndcg_at_1
value: 35.759
- type: ndcg_at_10
value: 29.754
- type: ndcg_at_100
value: 28.652
- type: ndcg_at_1000
value: 37.912
- type: ndcg_at_20
value: 28.366000000000003
- type: ndcg_at_3
value: 34.504000000000005
- type: ndcg_at_5
value: 31.996000000000002
- type: precision_at_1
value: 37.771
- type: precision_at_10
value: 23.064999999999998
- type: precision_at_100
value: 7.95
- type: precision_at_1000
value: 2.0820000000000003
- type: precision_at_20
value: 17.601
- type: precision_at_3
value: 33.745999999999995
- type: precision_at_5
value: 28.607
- type: recall_at_1
value: 4.567
- type: recall_at_10
value: 13.782
- type: recall_at_100
value: 31.269000000000002
- type: recall_at_1000
value: 65.342
- type: recall_at_20
value: 18.67
- type: recall_at_3
value: 8.280999999999999
- type: recall_at_5
value: 10.023
- task:
type: Retrieval
dataset:
name: MTEB NQ (default)
type: mteb/nq
config: default
split: test
revision: b774495ed302d8c44a3a7ea25c90dbce03968f31
metrics:
- type: main_score
value: 43.91
- type: map_at_1
value: 22.675
- type: map_at_10
value: 36.403
- type: map_at_100
value: 37.714
- type: map_at_1000
value: 37.759
- type: map_at_20
value: 37.252
- type: map_at_3
value: 32.115
- type: map_at_5
value: 34.537
- type: mrr_at_1
value: 25.869061413673233
- type: mrr_at_10
value: 38.71765160293547
- type: mrr_at_100
value: 39.729686512365404
- type: mrr_at_1000
value: 39.764837578266224
- type: mrr_at_20
value: 39.37441098811741
- type: mrr_at_3
value: 35.04731556585558
- type: mrr_at_5
value: 37.1330629586714
- type: nauc_map_at_1000_diff1
value: 29.08399936916601
- type: nauc_map_at_1000_max
value: 16.29226971170329
- type: nauc_map_at_1000_std
value: -7.18828364236571
- type: nauc_map_at_100_diff1
value: 29.07598191479223
- type: nauc_map_at_100_max
value: 16.313043083658002
- type: nauc_map_at_100_std
value: -7.159235211023741
- type: nauc_map_at_10_diff1
value: 28.90215829867663
- type: nauc_map_at_10_max
value: 15.780381654305788
- type: nauc_map_at_10_std
value: -8.11712873780433
- type: nauc_map_at_1_diff1
value: 33.88307764782601
- type: nauc_map_at_1_max
value: 13.654560605994789
- type: nauc_map_at_1_std
value: -7.876960683195404
- type: nauc_map_at_20_diff1
value: 29.05304473680581
- type: nauc_map_at_20_max
value: 16.24806438199264
- type: nauc_map_at_20_std
value: -7.506230537647335
- type: nauc_map_at_3_diff1
value: 28.67483907357866
- type: nauc_map_at_3_max
value: 14.10844684276125
- type: nauc_map_at_3_std
value: -9.093065914867637
- type: nauc_map_at_5_diff1
value: 28.53223723787074
- type: nauc_map_at_5_max
value: 15.10465507767692
- type: nauc_map_at_5_std
value: -8.732885670313973
- type: nauc_mrr_at_1000_diff1
value: 29.291300363671773
- type: nauc_mrr_at_1000_max
value: 16.988142920972564
- type: nauc_mrr_at_1000_std
value: -5.75719276458362
- type: nauc_mrr_at_100_diff1
value: 29.283110074292846
- type: nauc_mrr_at_100_max
value: 17.014791337479164
- type: nauc_mrr_at_100_std
value: -5.723942677946949
- type: nauc_mrr_at_10_diff1
value: 29.034300179010796
- type: nauc_mrr_at_10_max
value: 16.698355482721286
- type: nauc_mrr_at_10_std
value: -6.298274321653915
- type: nauc_mrr_at_1_diff1
value: 34.02416083157839
- type: nauc_mrr_at_1_max
value: 15.09009259703498
- type: nauc_mrr_at_1_std
value: -6.412150592020908
- type: nauc_mrr_at_20_diff1
value: 29.246190262288547
- type: nauc_mrr_at_20_max
value: 16.993567089527037
- type: nauc_mrr_at_20_std
value: -5.952301904693604
- type: nauc_mrr_at_3_diff1
value: 29.252167190201583
- type: nauc_mrr_at_3_max
value: 15.472766023941444
- type: nauc_mrr_at_3_std
value: -7.080025242602122
- type: nauc_mrr_at_5_diff1
value: 28.753793317757832
- type: nauc_mrr_at_5_max
value: 16.217731120360888
- type: nauc_mrr_at_5_std
value: -6.784186528364365
- type: nauc_ndcg_at_1000_diff1
value: 28.37769622443293
- type: nauc_ndcg_at_1000_max
value: 18.86280647857455
- type: nauc_ndcg_at_1000_std
value: -4.117301819438475
- type: nauc_ndcg_at_100_diff1
value: 28.160781181149385
- type: nauc_ndcg_at_100_max
value: 19.666949578583665
- type: nauc_ndcg_at_100_std
value: -3.0099241628896127
- type: nauc_ndcg_at_10_diff1
value: 27.487943653391138
- type: nauc_ndcg_at_10_max
value: 17.72148461204397
- type: nauc_ndcg_at_10_std
value: -7.13147681057827
- type: nauc_ndcg_at_1_diff1
value: 34.02416083157839
- type: nauc_ndcg_at_1_max
value: 15.09009259703498
- type: nauc_ndcg_at_1_std
value: -6.412150592020908
- type: nauc_ndcg_at_20_diff1
value: 27.98913738838462
- type: nauc_ndcg_at_20_max
value: 19.358472910765496
- type: nauc_ndcg_at_20_std
value: -5.168140450720515
- type: nauc_ndcg_at_3_diff1
value: 27.421114192997827
- type: nauc_ndcg_at_3_max
value: 14.573815202096219
- type: nauc_ndcg_at_3_std
value: -9.02108020108124
- type: nauc_ndcg_at_5_diff1
value: 26.863276500871386
- type: nauc_ndcg_at_5_max
value: 16.215661987458027
- type: nauc_ndcg_at_5_std
value: -8.51151065053834
- type: nauc_precision_at_1000_diff1
value: -1.9585068692545722
- type: nauc_precision_at_1000_max
value: 12.525664210853856
- type: nauc_precision_at_1000_std
value: 18.708210958723804
- type: nauc_precision_at_100_diff1
value: 4.218996145429994
- type: nauc_precision_at_100_max
value: 21.29629036792087
- type: nauc_precision_at_100_std
value: 22.876761226300427
- type: nauc_precision_at_10_diff1
value: 14.912990806791418
- type: nauc_precision_at_10_max
value: 20.15924358583
- type: nauc_precision_at_10_std
value: 1.139621390169978
- type: nauc_precision_at_1_diff1
value: 34.02416083157839
- type: nauc_precision_at_1_max
value: 15.09009259703498
- type: nauc_precision_at_1_std
value: -6.412150592020908
- type: nauc_precision_at_20_diff1
value: 11.657378156434522
- type: nauc_precision_at_20_max
value: 23.38664754164927
- type: nauc_precision_at_20_std
value: 9.617305551800612
- type: nauc_precision_at_3_diff1
value: 20.65465729065584
- type: nauc_precision_at_3_max
value: 15.503084569371712
- type: nauc_precision_at_3_std
value: -7.20044261830195
- type: nauc_precision_at_5_diff1
value: 17.40716535145957
- type: nauc_precision_at_5_max
value: 17.712445714665233
- type: nauc_precision_at_5_std
value: -4.721781521873729
- type: nauc_recall_at_1000_diff1
value: 21.620801919444965
- type: nauc_recall_at_1000_max
value: 70.1515764766629
- type: nauc_recall_at_1000_std
value: 68.09307840072094
- type: nauc_recall_at_100_diff1
value: 21.56265782889045
- type: nauc_recall_at_100_max
value: 49.39463091037063
- type: nauc_recall_at_100_std
value: 39.090999181732265
- type: nauc_recall_at_10_diff1
value: 21.766222223998323
- type: nauc_recall_at_10_max
value: 22.538222732899847
- type: nauc_recall_at_10_std
value: -5.681237338134916
- type: nauc_recall_at_1_diff1
value: 33.88307764782601
- type: nauc_recall_at_1_max
value: 13.654560605994789
- type: nauc_recall_at_1_std
value: -7.876960683195404
- type: nauc_recall_at_20_diff1
value: 22.984671066381193
- type: nauc_recall_at_20_max
value: 32.19082601358247
- type: nauc_recall_at_20_std
value: 3.881886772570749
- type: nauc_recall_at_3_diff1
value: 22.784351240811095
- type: nauc_recall_at_3_max
value: 14.199364117910083
- type: nauc_recall_at_3_std
value: -10.359618653345851
- type: nauc_recall_at_5_diff1
value: 20.994385793302932
- type: nauc_recall_at_5_max
value: 17.75611443293454
- type: nauc_recall_at_5_std
value: -9.551812400313235
- type: ndcg_at_1
value: 25.868999999999996
- type: ndcg_at_10
value: 43.91
- type: ndcg_at_100
value: 49.533
- type: ndcg_at_1000
value: 50.589
- type: ndcg_at_20
value: 46.691
- type: ndcg_at_3
value: 35.681000000000004
- type: ndcg_at_5
value: 39.75
- type: precision_at_1
value: 25.868999999999996
- type: precision_at_10
value: 7.722999999999999
- type: precision_at_100
value: 1.083
- type: precision_at_1000
value: 0.11800000000000001
- type: precision_at_20
value: 4.521
- type: precision_at_3
value: 16.773
- type: precision_at_5
value: 12.422
- type: recall_at_1
value: 22.675
- type: recall_at_10
value: 64.63600000000001
- type: recall_at_100
value: 89.253
- type: recall_at_1000
value: 97.103
- type: recall_at_20
value: 75.027
- type: recall_at_3
value: 43.079
- type: recall_at_5
value: 52.518
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval (default)
type: mteb/quora
config: default
split: dev
revision: e4e08e0b7dbe3c8700f0daef558ff32256715259
metrics:
- type: main_score
value: 85.386
- type: map_at_1
value: 66.987
- type: map_at_10
value: 81.444
- type: map_at_100
value: 82.14
- type: map_at_1000
value: 82.157
- type: map_at_20
value: 81.932
- type: map_at_3
value: 78.484
- type: map_at_5
value: 80.274
- type: mrr_at_1
value: 77.44
- type: mrr_at_10
value: 84.2241825396826
- type: mrr_at_100
value: 84.40693808056355
- type: mrr_at_1000
value: 84.40988912757359
- type: mrr_at_20
value: 84.34978828558788
- type: mrr_at_3
value: 83.10333333333354
- type: mrr_at_5
value: 83.82533333333342
- type: nauc_map_at_1000_diff1
value: 71.91679334130635
- type: nauc_map_at_1000_max
value: 31.103584943909812
- type: nauc_map_at_1000_std
value: -40.456705674932074
- type: nauc_map_at_100_diff1
value: 71.91963348224603
- type: nauc_map_at_100_max
value: 31.113011110459922
- type: nauc_map_at_100_std
value: -40.502282566735715
- type: nauc_map_at_10_diff1
value: 72.12325712296365
- type: nauc_map_at_10_max
value: 30.467237037792295
- type: nauc_map_at_10_std
value: -41.94971797229005
- type: nauc_map_at_1_diff1
value: 76.18084680744592
- type: nauc_map_at_1_max
value: 19.221792285260147
- type: nauc_map_at_1_std
value: -36.62117815127342
- type: nauc_map_at_20_diff1
value: 72.00403582380048
- type: nauc_map_at_20_max
value: 31.004347688742417
- type: nauc_map_at_20_std
value: -41.060042626586416
- type: nauc_map_at_3_diff1
value: 72.93189642332118
- type: nauc_map_at_3_max
value: 27.73892205168097
- type: nauc_map_at_3_std
value: -43.30991939814638
- type: nauc_map_at_5_diff1
value: 72.5160537522699
- type: nauc_map_at_5_max
value: 29.63685791438478
- type: nauc_map_at_5_std
value: -42.96659156810372
- type: nauc_mrr_at_1000_diff1
value: 71.90147173095302
- type: nauc_mrr_at_1000_max
value: 35.940317491851694
- type: nauc_mrr_at_1000_std
value: -36.68579617983882
- type: nauc_mrr_at_100_diff1
value: 71.90072229115142
- type: nauc_mrr_at_100_max
value: 35.94516255768936
- type: nauc_mrr_at_100_std
value: -36.67856828426162
- type: nauc_mrr_at_10_diff1
value: 71.85801945991824
- type: nauc_mrr_at_10_max
value: 36.07450549726053
- type: nauc_mrr_at_10_std
value: -36.6574239418474
- type: nauc_mrr_at_1_diff1
value: 72.7487436455008
- type: nauc_mrr_at_1_max
value: 33.4319386768096
- type: nauc_mrr_at_1_std
value: -36.45664395168034
- type: nauc_mrr_at_20_diff1
value: 71.90639260981615
- type: nauc_mrr_at_20_max
value: 36.016630372090525
- type: nauc_mrr_at_20_std
value: -36.66806949361515
- type: nauc_mrr_at_3_diff1
value: 71.44580989174713
- type: nauc_mrr_at_3_max
value: 36.14773805963233
- type: nauc_mrr_at_3_std
value: -36.77858114551201
- type: nauc_mrr_at_5_diff1
value: 71.76720756731626
- type: nauc_mrr_at_5_max
value: 36.2542204305357
- type: nauc_mrr_at_5_std
value: -36.6104713553764
- type: nauc_ndcg_at_1000_diff1
value: 71.53171713574864
- type: nauc_ndcg_at_1000_max
value: 33.46348829708273
- type: nauc_ndcg_at_1000_std
value: -38.07080231429607
- type: nauc_ndcg_at_100_diff1
value: 71.51771411127783
- type: nauc_ndcg_at_100_max
value: 33.606977743408365
- type: nauc_ndcg_at_100_std
value: -38.095941435745075
- type: nauc_ndcg_at_10_diff1
value: 71.69089621879667
- type: nauc_ndcg_at_10_max
value: 33.00069266017414
- type: nauc_ndcg_at_10_std
value: -40.7188348379717
- type: nauc_ndcg_at_1_diff1
value: 72.79491888263016
- type: nauc_ndcg_at_1_max
value: 33.29638988300579
- type: nauc_ndcg_at_1_std
value: -36.447696477457235
- type: nauc_ndcg_at_20_diff1
value: 71.76982125411759
- type: nauc_ndcg_at_20_max
value: 33.64375758413667
- type: nauc_ndcg_at_20_std
value: -39.49026957812086
- type: nauc_ndcg_at_3_diff1
value: 71.16497890966858
- type: nauc_ndcg_at_3_max
value: 32.07287318566269
- type: nauc_ndcg_at_3_std
value: -41.48214270415671
- type: nauc_ndcg_at_5_diff1
value: 71.62531294064848
- type: nauc_ndcg_at_5_max
value: 32.669489846127426
- type: nauc_ndcg_at_5_std
value: -41.48868720768341
- type: nauc_precision_at_1000_diff1
value: -41.78128711551645
- type: nauc_precision_at_1000_max
value: 2.438112824671157
- type: nauc_precision_at_1000_std
value: 36.31304884788196
- type: nauc_precision_at_100_diff1
value: -40.88585036122386
- type: nauc_precision_at_100_max
value: 4.549436618614055
- type: nauc_precision_at_100_std
value: 34.22863218610554
- type: nauc_precision_at_10_diff1
value: -32.498791014783215
- type: nauc_precision_at_10_max
value: 10.335105655353592
- type: nauc_precision_at_10_std
value: 19.615568390029996
- type: nauc_precision_at_1_diff1
value: 72.79491888263016
- type: nauc_precision_at_1_max
value: 33.29638988300579
- type: nauc_precision_at_1_std
value: -36.447696477457235
- type: nauc_precision_at_20_diff1
value: -37.02913687382553
- type: nauc_precision_at_20_max
value: 7.702625378474581
- type: nauc_precision_at_20_std
value: 26.924503385144305
- type: nauc_precision_at_3_diff1
value: -10.70889509336384
- type: nauc_precision_at_3_max
value: 19.01391790730139
- type: nauc_precision_at_3_std
value: -0.3082223867032308
- type: nauc_precision_at_5_diff1
value: -23.455129536841024
- type: nauc_precision_at_5_max
value: 15.200131768852115
- type: nauc_precision_at_5_std
value: 10.043071356667568
- type: nauc_recall_at_1000_diff1
value: 53.062595262077814
- type: nauc_recall_at_1000_max
value: -22.60327055724744
- type: nauc_recall_at_1000_std
value: 30.926362639911186
- type: nauc_recall_at_100_diff1
value: 64.86049883288433
- type: nauc_recall_at_100_max
value: 43.630795280243426
- type: nauc_recall_at_100_std
value: -19.004039027440363
- type: nauc_recall_at_10_diff1
value: 67.6088540053325
- type: nauc_recall_at_10_max
value: 32.59063448235274
- type: nauc_recall_at_10_std
value: -48.53346063375749
- type: nauc_recall_at_1_diff1
value: 76.18084680744592
- type: nauc_recall_at_1_max
value: 19.221792285260147
- type: nauc_recall_at_1_std
value: -36.62117815127342
- type: nauc_recall_at_20_diff1
value: 69.9198659134458
- type: nauc_recall_at_20_max
value: 39.37372830048965
- type: nauc_recall_at_20_std
value: -45.5107664415436
- type: nauc_recall_at_3_diff1
value: 69.4918403464634
- type: nauc_recall_at_3_max
value: 27.888269294676952
- type: nauc_recall_at_3_std
value: -46.75391426769085
- type: nauc_recall_at_5_diff1
value: 68.69308264633088
- type: nauc_recall_at_5_max
value: 30.788769684713763
- type: nauc_recall_at_5_std
value: -48.20245507137716
- type: ndcg_at_1
value: 77.42
- type: ndcg_at_10
value: 85.386
- type: ndcg_at_100
value: 86.964
- type: ndcg_at_1000
value: 87.10799999999999
- type: ndcg_at_20
value: 86.259
- type: ndcg_at_3
value: 82.356
- type: ndcg_at_5
value: 83.91
- type: precision_at_1
value: 77.42
- type: precision_at_10
value: 13.064
- type: precision_at_100
value: 1.488
- type: precision_at_1000
value: 0.152
- type: precision_at_20
value: 6.978
- type: precision_at_3
value: 36.353
- type: precision_at_5
value: 23.888
- type: recall_at_1
value: 66.987
- type: recall_at_10
value: 93.487
- type: recall_at_100
value: 99.125
- type: recall_at_1000
value: 99.94
- type: recall_at_20
value: 96.263
- type: recall_at_3
value: 84.761
- type: recall_at_5
value: 89.242
- task:
type: Retrieval
dataset:
name: MTEB QuoraRetrieval (default)
type: mteb/quora
config: default
split: test
revision: e4e08e0b7dbe3c8700f0daef558ff32256715259
metrics:
- type: main_score
value: 85.56099999999999
- type: map_at_1
value: 67.497
- type: map_at_10
value: 81.523
- type: map_at_100
value: 82.253
- type: map_at_1000
value: 82.269
- type: map_at_20
value: 82.0
- type: map_at_3
value: 78.39399999999999
- type: map_at_5
value: 80.352
- type: mrr_at_1
value: 77.97
- type: mrr_at_10
value: 84.6137420634918
- type: mrr_at_100
value: 84.7867074667826
- type: mrr_at_1000
value: 84.7886101354869
- type: mrr_at_20
value: 84.74068012792132
- type: mrr_at_3
value: 83.48833333333296
- type: mrr_at_5
value: 84.26183333333283
- type: nauc_map_at_1000_diff1
value: 72.75256203555736
- type: nauc_map_at_1000_max
value: 32.109499410706924
- type: nauc_map_at_1000_std
value: -43.48607664191692
- type: nauc_map_at_100_diff1
value: 72.7621590254899
- type: nauc_map_at_100_max
value: 32.08666240896605
- type: nauc_map_at_100_std
value: -43.52175837960735
- type: nauc_map_at_10_diff1
value: 73.0914502532422
- type: nauc_map_at_10_max
value: 31.21935764927552
- type: nauc_map_at_10_std
value: -45.596143090602155
- type: nauc_map_at_1_diff1
value: 77.6224306824682
- type: nauc_map_at_1_max
value: 22.507118034491427
- type: nauc_map_at_1_std
value: -40.33167676556522
- type: nauc_map_at_20_diff1
value: 72.8820293672954
- type: nauc_map_at_20_max
value: 31.73360303879727
- type: nauc_map_at_20_std
value: -44.454775928554376
- type: nauc_map_at_3_diff1
value: 74.10546474965095
- type: nauc_map_at_3_max
value: 27.647987662942402
- type: nauc_map_at_3_std
value: -47.234626583510355
- type: nauc_map_at_5_diff1
value: 73.52017296999442
- type: nauc_map_at_5_max
value: 29.763316409372663
- type: nauc_map_at_5_std
value: -46.85804426702457
- type: nauc_mrr_at_1000_diff1
value: 72.46174562552228
- type: nauc_mrr_at_1000_max
value: 36.29693186997953
- type: nauc_mrr_at_1000_std
value: -39.138708209808
- type: nauc_mrr_at_100_diff1
value: 72.46035839735728
- type: nauc_mrr_at_100_max
value: 36.302723895121645
- type: nauc_mrr_at_100_std
value: -39.1324046718197
- type: nauc_mrr_at_10_diff1
value: 72.36917674578362
- type: nauc_mrr_at_10_max
value: 36.369854321798634
- type: nauc_mrr_at_10_std
value: -39.236853924584764
- type: nauc_mrr_at_1_diff1
value: 74.17185246735727
- type: nauc_mrr_at_1_max
value: 34.95837580050766
- type: nauc_mrr_at_1_std
value: -37.80684342737821
- type: nauc_mrr_at_20_diff1
value: 72.44437909450644
- type: nauc_mrr_at_20_max
value: 36.33224715875864
- type: nauc_mrr_at_20_std
value: -39.169673628719565
- type: nauc_mrr_at_3_diff1
value: 72.22379259192769
- type: nauc_mrr_at_3_max
value: 35.824553778653325
- type: nauc_mrr_at_3_std
value: -39.79306459480282
- type: nauc_mrr_at_5_diff1
value: 72.2369691369586
- type: nauc_mrr_at_5_max
value: 36.33784223941021
- type: nauc_mrr_at_5_std
value: -39.36670062046302
- type: nauc_ndcg_at_1000_diff1
value: 72.01179014018193
- type: nauc_ndcg_at_1000_max
value: 34.66610828967116
- type: nauc_ndcg_at_1000_std
value: -40.79246584671935
- type: nauc_ndcg_at_100_diff1
value: 72.02571416492458
- type: nauc_ndcg_at_100_max
value: 34.724662063977476
- type: nauc_ndcg_at_100_std
value: -40.71853456489787
- type: nauc_ndcg_at_10_diff1
value: 72.05449454021301
- type: nauc_ndcg_at_10_max
value: 33.65231664572072
- type: nauc_ndcg_at_10_std
value: -44.387726473312185
- type: nauc_ndcg_at_1_diff1
value: 74.17185246735727
- type: nauc_ndcg_at_1_max
value: 34.97004776008378
- type: nauc_ndcg_at_1_std
value: -37.71249508747126
- type: nauc_ndcg_at_20_diff1
value: 72.09392270433015
- type: nauc_ndcg_at_20_max
value: 34.04000107921998
- type: nauc_ndcg_at_20_std
value: -42.86390076582997
- type: nauc_ndcg_at_3_diff1
value: 72.04472679866134
- type: nauc_ndcg_at_3_max
value: 31.342820752545002
- type: nauc_ndcg_at_3_std
value: -44.8034473982025
- type: nauc_ndcg_at_5_diff1
value: 72.01402290251924
- type: nauc_ndcg_at_5_max
value: 32.582295878461906
- type: nauc_ndcg_at_5_std
value: -45.23466775961186
- type: nauc_precision_at_1000_diff1
value: -43.708171172526
- type: nauc_precision_at_1000_max
value: 4.162817192837958
- type: nauc_precision_at_1000_std
value: 41.4385681190868
- type: nauc_precision_at_100_diff1
value: -43.108346163549854
- type: nauc_precision_at_100_max
value: 4.67035667258918
- type: nauc_precision_at_100_std
value: 41.46618060427868
- type: nauc_precision_at_10_diff1
value: -34.98729419918756
- type: nauc_precision_at_10_max
value: 8.60311602976671
- type: nauc_precision_at_10_std
value: 24.27173552773047
- type: nauc_precision_at_1_diff1
value: 74.17185246735727
- type: nauc_precision_at_1_max
value: 34.97004776008378
- type: nauc_precision_at_1_std
value: -37.71249508747126
- type: nauc_precision_at_20_diff1
value: -39.693045353877146
- type: nauc_precision_at_20_max
value: 6.566610093825938
- type: nauc_precision_at_20_std
value: 32.591669383633864
- type: nauc_precision_at_3_diff1
value: -10.84059740293422
- type: nauc_precision_at_3_max
value: 14.778921667316514
- type: nauc_precision_at_3_std
value: 0.20610132151386867
- type: nauc_precision_at_5_diff1
value: -25.776681106529015
- type: nauc_precision_at_5_max
value: 11.422241065581462
- type: nauc_precision_at_5_std
value: 13.113277134240322
- type: nauc_recall_at_1000_diff1
value: 63.99277654005502
- type: nauc_recall_at_1000_max
value: 81.89809119086901
- type: nauc_recall_at_1000_std
value: 49.009493075982604
- type: nauc_recall_at_100_diff1
value: 65.00747551591195
- type: nauc_recall_at_100_max
value: 58.34661374064065
- type: nauc_recall_at_100_std
value: -2.2251834665513215
- type: nauc_recall_at_10_diff1
value: 65.53778699047477
- type: nauc_recall_at_10_max
value: 32.37899366515905
- type: nauc_recall_at_10_std
value: -56.58653525784876
- type: nauc_recall_at_1_diff1
value: 77.6224306824682
- type: nauc_recall_at_1_max
value: 22.507118034491427
- type: nauc_recall_at_1_std
value: -40.33167676556522
- type: nauc_recall_at_20_diff1
value: 65.50582600247682
- type: nauc_recall_at_20_max
value: 35.57972574345174
- type: nauc_recall_at_20_std
value: -51.94704680030098
- type: nauc_recall_at_3_diff1
value: 70.09083256856663
- type: nauc_recall_at_3_max
value: 24.1544101793869
- type: nauc_recall_at_3_std
value: -53.465675345124055
- type: nauc_recall_at_5_diff1
value: 67.77056112320096
- type: nauc_recall_at_5_max
value: 27.958593788155987
- type: nauc_recall_at_5_std
value: -56.15384175590208
- type: ndcg_at_1
value: 77.97
- type: ndcg_at_10
value: 85.56099999999999
- type: ndcg_at_100
value: 87.14699999999999
- type: ndcg_at_1000
value: 87.274
- type: ndcg_at_20
value: 86.416
- type: ndcg_at_3
value: 82.349
- type: ndcg_at_5
value: 84.148
- type: precision_at_1
value: 77.97
- type: precision_at_10
value: 13.175
- type: precision_at_100
value: 1.5270000000000001
- type: precision_at_1000
value: 0.156
- type: precision_at_20
value: 7.037
- type: precision_at_3
value: 36.183
- type: precision_at_5
value: 24.01
- type: recall_at_1
value: 67.497
- type: recall_at_10
value: 93.452
- type: recall_at_100
value: 99.181
- type: recall_at_1000
value: 99.855
- type: recall_at_20
value: 96.31
- type: recall_at_3
value: 84.396
- type: recall_at_5
value: 89.298
- task:
type: Retrieval
dataset:
name: MTEB SCIDOCS (default)
type: mteb/scidocs
config: default
split: test
revision: f8c2fcf00f625baaa80f62ec5bd9e1fff3b8ae88
metrics:
- type: main_score
value: 21.383
- type: map_at_1
value: 4.547
- type: map_at_10
value: 12.629000000000001
- type: map_at_100
value: 15.085999999999999
- type: map_at_1000
value: 15.478
- type: map_at_20
value: 13.806
- type: map_at_3
value: 8.620999999999999
- type: map_at_5
value: 10.465
- type: mrr_at_1
value: 22.3
- type: mrr_at_10
value: 34.701309523809535
- type: mrr_at_100
value: 35.731488574144606
- type: mrr_at_1000
value: 35.78105399278444
- type: mrr_at_20
value: 35.27448919286768
- type: mrr_at_3
value: 30.800000000000015
- type: mrr_at_5
value: 33.14999999999999
- type: nauc_map_at_1000_diff1
value: 8.343120963945573
- type: nauc_map_at_1000_max
value: 33.75996610404545
- type: nauc_map_at_1000_std
value: 21.75562874917409
- type: nauc_map_at_100_diff1
value: 8.384427363803946
- type: nauc_map_at_100_max
value: 33.79077477489537
- type: nauc_map_at_100_std
value: 21.686104124370104
- type: nauc_map_at_10_diff1
value: 9.234064591129155
- type: nauc_map_at_10_max
value: 32.0489964068748
- type: nauc_map_at_10_std
value: 17.763385123837153
- type: nauc_map_at_1_diff1
value: 15.665356237945588
- type: nauc_map_at_1_max
value: 26.57885889682774
- type: nauc_map_at_1_std
value: 13.193930049941363
- type: nauc_map_at_20_diff1
value: 8.780121965824657
- type: nauc_map_at_20_max
value: 32.869986461841606
- type: nauc_map_at_20_std
value: 19.467991991140142
- type: nauc_map_at_3_diff1
value: 11.035561869808728
- type: nauc_map_at_3_max
value: 27.582026721813047
- type: nauc_map_at_3_std
value: 13.036167718922828
- type: nauc_map_at_5_diff1
value: 10.080603205194704
- type: nauc_map_at_5_max
value: 30.044299131641584
- type: nauc_map_at_5_std
value: 15.350543387048896
- type: nauc_mrr_at_1000_diff1
value: 11.840089363851067
- type: nauc_mrr_at_1000_max
value: 29.12864266402923
- type: nauc_mrr_at_1000_std
value: 17.228534279486695
- type: nauc_mrr_at_100_diff1
value: 11.809506621783779
- type: nauc_mrr_at_100_max
value: 29.160886200309044
- type: nauc_mrr_at_100_std
value: 17.282962681247795
- type: nauc_mrr_at_10_diff1
value: 11.935445560209875
- type: nauc_mrr_at_10_max
value: 29.30219333248834
- type: nauc_mrr_at_10_std
value: 16.987513421652007
- type: nauc_mrr_at_1_diff1
value: 16.006372448277954
- type: nauc_mrr_at_1_max
value: 26.835362452310946
- type: nauc_mrr_at_1_std
value: 13.253888908633249
- type: nauc_mrr_at_20_diff1
value: 11.859112690530099
- type: nauc_mrr_at_20_max
value: 29.23212198991369
- type: nauc_mrr_at_20_std
value: 17.371690243627583
- type: nauc_mrr_at_3_diff1
value: 11.942072558497351
- type: nauc_mrr_at_3_max
value: 27.518198076910377
- type: nauc_mrr_at_3_std
value: 15.553451577229513
- type: nauc_mrr_at_5_diff1
value: 11.844643930313167
- type: nauc_mrr_at_5_max
value: 28.507660302183385
- type: nauc_mrr_at_5_std
value: 16.407092796978915
- type: nauc_ndcg_at_1000_diff1
value: 7.566941588567799
- type: nauc_ndcg_at_1000_max
value: 35.66190450829144
- type: nauc_ndcg_at_1000_std
value: 27.644208320073655
- type: nauc_ndcg_at_100_diff1
value: 7.789806046298824
- type: nauc_ndcg_at_100_max
value: 36.44803836644228
- type: nauc_ndcg_at_100_std
value: 28.570220616023008
- type: nauc_ndcg_at_10_diff1
value: 9.112031867544205
- type: nauc_ndcg_at_10_max
value: 33.29203357424967
- type: nauc_ndcg_at_10_std
value: 19.717835519047455
- type: nauc_ndcg_at_1_diff1
value: 16.006372448277954
- type: nauc_ndcg_at_1_max
value: 26.835362452310946
- type: nauc_ndcg_at_1_std
value: 13.253888908633249
- type: nauc_ndcg_at_20_diff1
value: 8.512820930810134
- type: nauc_ndcg_at_20_max
value: 34.04402394206212
- type: nauc_ndcg_at_20_std
value: 22.450723493452905
- type: nauc_ndcg_at_3_diff1
value: 10.666228630298292
- type: nauc_ndcg_at_3_max
value: 27.790265132571
- type: nauc_ndcg_at_3_std
value: 14.549017414623137
- type: nauc_ndcg_at_5_diff1
value: 9.689838800282292
- type: nauc_ndcg_at_5_max
value: 30.200901003139375
- type: nauc_ndcg_at_5_std
value: 16.756955996011495
- type: nauc_precision_at_1000_diff1
value: -1.7688373117252636
- type: nauc_precision_at_1000_max
value: 28.14939872063652
- type: nauc_precision_at_1000_std
value: 36.11781984236825
- type: nauc_precision_at_100_diff1
value: 2.6982654390147434
- type: nauc_precision_at_100_max
value: 34.95302652738801
- type: nauc_precision_at_100_std
value: 38.612855789126435
- type: nauc_precision_at_10_diff1
value: 6.3582141835613575
- type: nauc_precision_at_10_max
value: 34.35036042979775
- type: nauc_precision_at_10_std
value: 22.37284626453628
- type: nauc_precision_at_1_diff1
value: 16.006372448277954
- type: nauc_precision_at_1_max
value: 26.835362452310946
- type: nauc_precision_at_1_std
value: 13.253888908633249
- type: nauc_precision_at_20_diff1
value: 4.929659714506811
- type: nauc_precision_at_20_max
value: 33.63912155282006
- type: nauc_precision_at_20_std
value: 26.48159430953102
- type: nauc_precision_at_3_diff1
value: 8.684533707455072
- type: nauc_precision_at_3_max
value: 27.201795527705
- type: nauc_precision_at_3_std
value: 14.702968668907863
- type: nauc_precision_at_5_diff1
value: 6.996882444408692
- type: nauc_precision_at_5_max
value: 30.190181104191716
- type: nauc_precision_at_5_std
value: 17.9773294217075
- type: nauc_recall_at_1000_diff1
value: -2.081286806432597
- type: nauc_recall_at_1000_max
value: 28.92540441778861
- type: nauc_recall_at_1000_std
value: 38.288436893164594
- type: nauc_recall_at_100_diff1
value: 2.587296034481601
- type: nauc_recall_at_100_max
value: 35.29377052645708
- type: nauc_recall_at_100_std
value: 39.05821703702423
- type: nauc_recall_at_10_diff1
value: 6.316338189514334
- type: nauc_recall_at_10_max
value: 34.11763962208226
- type: nauc_recall_at_10_std
value: 22.195626530130237
- type: nauc_recall_at_1_diff1
value: 15.665356237945588
- type: nauc_recall_at_1_max
value: 26.57885889682774
- type: nauc_recall_at_1_std
value: 13.193930049941363
- type: nauc_recall_at_20_diff1
value: 4.830004620366135
- type: nauc_recall_at_20_max
value: 33.408749981759904
- type: nauc_recall_at_20_std
value: 26.32129089751909
- type: nauc_recall_at_3_diff1
value: 8.682897494798821
- type: nauc_recall_at_3_max
value: 26.920725011561657
- type: nauc_recall_at_3_std
value: 14.373627083919555
- type: nauc_recall_at_5_diff1
value: 7.078349615166568
- type: nauc_recall_at_5_max
value: 29.938684723317483
- type: nauc_recall_at_5_std
value: 17.610677580599983
- type: ndcg_at_1
value: 22.3
- type: ndcg_at_10
value: 21.383
- type: ndcg_at_100
value: 30.686999999999998
- type: ndcg_at_1000
value: 36.806
- type: ndcg_at_20
value: 24.38
- type: ndcg_at_3
value: 19.262999999999998
- type: ndcg_at_5
value: 17.226
- type: precision_at_1
value: 22.3
- type: precision_at_10
value: 11.53
- type: precision_at_100
value: 2.5170000000000003
- type: precision_at_1000
value: 0.39699999999999996
- type: precision_at_20
value: 7.495
- type: precision_at_3
value: 18.3
- type: precision_at_5
value: 15.479999999999999
- type: recall_at_1
value: 4.547
- type: recall_at_10
value: 23.357
- type: recall_at_100
value: 51.047
- type: recall_at_1000
value: 80.63
- type: recall_at_20
value: 30.392999999999997
- type: recall_at_3
value: 11.161999999999999
- type: recall_at_5
value: 15.702
- task:
type: STS
dataset:
name: MTEB SICK-R (default)
type: mteb/sickr-sts
config: default
split: test
revision: 20a6d6f312dd54037fe07a32d58e5e168867909d
metrics:
- type: cosine_pearson
value: 82.89761522599564
- type: cosine_spearman
value: 79.61239240879378
- type: euclidean_pearson
value: 80.47994478833725
- type: euclidean_spearman
value: 79.61239321441792
- type: main_score
value: 79.61239240879378
- type: manhattan_pearson
value: 80.28255784608595
- type: manhattan_spearman
value: 79.40840813688955
- type: pearson
value: 82.89761555899388
- type: spearman
value: 79.6123884921359
- task:
type: STS
dataset:
name: MTEB STS12 (default)
type: mteb/sts12-sts
config: default
split: test
revision: a0d554a64d88156834ff5ae9920b964011b16384
metrics:
- type: cosine_pearson
value: 80.60566114110073
- type: cosine_spearman
value: 70.13320748146175
- type: euclidean_pearson
value: 77.57067368654582
- type: euclidean_spearman
value: 70.13314252571948
- type: main_score
value: 70.13320748146175
- type: manhattan_pearson
value: 77.29676392658962
- type: manhattan_spearman
value: 69.88321444641842
- type: pearson
value: 80.60566179532725
- type: spearman
value: 70.13415482313484
- task:
type: STS
dataset:
name: MTEB STS13 (default)
type: mteb/sts13-sts
config: default
split: test
revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
metrics:
- type: cosine_pearson
value: 78.320437342309
- type: cosine_spearman
value: 80.74355074851618
- type: euclidean_pearson
value: 80.17187362623575
- type: euclidean_spearman
value: 80.7435506767441
- type: main_score
value: 80.74355074851618
- type: manhattan_pearson
value: 79.88874001476627
- type: manhattan_spearman
value: 80.45868266096211
- type: pearson
value: 78.32043820850613
- type: spearman
value: 80.7435506767441
- task:
type: STS
dataset:
name: MTEB STS14 (default)
type: mteb/sts14-sts
config: default
split: test
revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
metrics:
- type: cosine_pearson
value: 79.70921389242768
- type: cosine_spearman
value: 76.24023668566399
- type: euclidean_pearson
value: 79.46251371995523
- type: euclidean_spearman
value: 76.24023301664893
- type: main_score
value: 76.24023668566399
- type: manhattan_pearson
value: 79.20065381375537
- type: manhattan_spearman
value: 76.00256203342937
- type: pearson
value: 79.70921366493712
- type: spearman
value: 76.24024123334894
- task:
type: STS
dataset:
name: MTEB STS15 (default)
type: mteb/sts15-sts
config: default
split: test
revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
metrics:
- type: cosine_pearson
value: 82.52746024419504
- type: cosine_spearman
value: 84.11623590647757
- type: euclidean_pearson
value: 83.48562697051246
- type: euclidean_spearman
value: 84.11623814887781
- type: main_score
value: 84.11623590647757
- type: manhattan_pearson
value: 83.31205342795175
- type: manhattan_spearman
value: 83.94265353931766
- type: pearson
value: 82.5274599058833
- type: spearman
value: 84.11622788808319
- task:
type: STS
dataset:
name: MTEB STS16 (default)
type: mteb/sts16-sts
config: default
split: test
revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
metrics:
- type: cosine_pearson
value: 74.82265172874041
- type: cosine_spearman
value: 78.4960329393101
- type: euclidean_pearson
value: 78.22716476239002
- type: euclidean_spearman
value: 78.4960329393101
- type: main_score
value: 78.4960329393101
- type: manhattan_pearson
value: 78.05885015944702
- type: manhattan_spearman
value: 78.3249847702297
- type: pearson
value: 74.8226511407601
- type: spearman
value: 78.49603251582666
- task:
type: STS
dataset:
name: MTEB STS17 (en-ar)
type: mteb/sts17-crosslingual-sts
config: en-ar
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 9.087713281584145
- type: cosine_spearman
value: 9.1279366990081
- type: euclidean_pearson
value: 9.117969623657645
- type: euclidean_spearman
value: 9.1279366990081
- type: main_score
value: 9.1279366990081
- type: manhattan_pearson
value: 6.440301458166155
- type: manhattan_spearman
value: 6.5264401438882
- type: pearson
value: 9.087716717716269
- type: spearman
value: 9.1279366990081
- task:
type: STS
dataset:
name: MTEB STS17 (ko-ko)
type: mteb/sts17-crosslingual-sts
config: ko-ko
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 21.931422026594692
- type: cosine_spearman
value: 32.79923712532921
- type: euclidean_pearson
value: 28.93227440763988
- type: euclidean_spearman
value: 32.79905675768862
- type: main_score
value: 32.79923712532921
- type: manhattan_pearson
value: 29.05113424705813
- type: manhattan_spearman
value: 32.79564881007963
- type: pearson
value: 21.931423583103467
- type: spearman
value: 32.798301050425714
- task:
type: STS
dataset:
name: MTEB STS17 (en-de)
type: mteb/sts17-crosslingual-sts
config: en-de
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 30.28151988475531
- type: cosine_spearman
value: 28.038768547398902
- type: euclidean_pearson
value: 30.498915267658234
- type: euclidean_spearman
value: 28.038768547398902
- type: main_score
value: 28.038768547398902
- type: manhattan_pearson
value: 30.794292054714152
- type: manhattan_spearman
value: 28.56988926201773
- type: pearson
value: 30.281515394951132
- type: spearman
value: 28.038768547398902
- task:
type: STS
dataset:
name: MTEB STS17 (en-tr)
type: mteb/sts17-crosslingual-sts
config: en-tr
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 1.103529546877178
- type: cosine_spearman
value: -4.344212733215535
- type: euclidean_pearson
value: 1.4727741337049667
- type: euclidean_spearman
value: -4.344212733215535
- type: main_score
value: -4.344212733215535
- type: manhattan_pearson
value: 3.066202915700401
- type: manhattan_spearman
value: -1.8750642080759847
- type: pearson
value: 1.1035283349143992
- type: spearman
value: -4.344212733215535
- task:
type: STS
dataset:
name: MTEB STS17 (ar-ar)
type: mteb/sts17-crosslingual-sts
config: ar-ar
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 34.20639223082103
- type: cosine_spearman
value: 50.099745536749865
- type: euclidean_pearson
value: 46.672397550772246
- type: euclidean_spearman
value: 50.099745536749865
- type: main_score
value: 50.099745536749865
- type: manhattan_pearson
value: 46.27511641947692
- type: manhattan_spearman
value: 49.773911789667046
- type: pearson
value: 34.20638672669267
- type: spearman
value: 50.092555073567134
- task:
type: STS
dataset:
name: MTEB STS17 (it-en)
type: mteb/sts17-crosslingual-sts
config: it-en
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 28.724311230307375
- type: cosine_spearman
value: 28.250724805661058
- type: euclidean_pearson
value: 28.315938847964397
- type: euclidean_spearman
value: 28.250724805661058
- type: main_score
value: 28.250724805661058
- type: manhattan_pearson
value: 27.674391866744763
- type: manhattan_spearman
value: 26.340658343374766
- type: pearson
value: 28.724300976547458
- type: spearman
value: 28.250724805661058
- task:
type: STS
dataset:
name: MTEB STS17 (es-es)
type: mteb/sts17-crosslingual-sts
config: es-es
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 68.60948138294124
- type: cosine_spearman
value: 71.2644385447136
- type: euclidean_pearson
value: 73.24322173392738
- type: euclidean_spearman
value: 71.2644385447136
- type: main_score
value: 71.2644385447136
- type: manhattan_pearson
value: 72.77369507880161
- type: manhattan_spearman
value: 70.79274454462727
- type: pearson
value: 68.6094780579963
- type: spearman
value: 71.26563420541223
- task:
type: STS
dataset:
name: MTEB STS17 (en-en)
type: mteb/sts17-crosslingual-sts
config: en-en
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 89.55726457989037
- type: cosine_spearman
value: 89.80253006361272
- type: euclidean_pearson
value: 89.52662858022953
- type: euclidean_spearman
value: 89.80253006361272
- type: main_score
value: 89.80253006361272
- type: manhattan_pearson
value: 89.515204742883
- type: manhattan_spearman
value: 89.85327040835446
- type: pearson
value: 89.5572652518414
- type: spearman
value: 89.80253006361272
- task:
type: STS
dataset:
name: MTEB STS17 (fr-en)
type: mteb/sts17-crosslingual-sts
config: fr-en
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 35.25506119586307
- type: cosine_spearman
value: 34.7509473636312
- type: euclidean_pearson
value: 34.966513615613614
- type: euclidean_spearman
value: 34.7509473636312
- type: main_score
value: 34.7509473636312
- type: manhattan_pearson
value: 34.205440386931265
- type: manhattan_spearman
value: 34.19580186457024
- type: pearson
value: 35.255050728265005
- type: spearman
value: 34.7509473636312
- task:
type: STS
dataset:
name: MTEB STS17 (nl-en)
type: mteb/sts17-crosslingual-sts
config: nl-en
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 29.15956073846505
- type: cosine_spearman
value: 23.4150546328063
- type: euclidean_pearson
value: 29.36385636868059
- type: euclidean_spearman
value: 23.4150546328063
- type: main_score
value: 23.4150546328063
- type: manhattan_pearson
value: 27.709164754459888
- type: manhattan_spearman
value: 21.483654161935416
- type: pearson
value: 29.159552605316065
- type: spearman
value: 23.4150546328063
- task:
type: STS
dataset:
name: MTEB STS17 (es-en)
type: mteb/sts17-crosslingual-sts
config: es-en
split: test
revision: faeb762787bd10488a50c8b5be4a3b82e411949c
metrics:
- type: cosine_pearson
value: 16.614254117804347
- type: cosine_spearman
value: 17.838933806777177
- type: euclidean_pearson
value: 16.374166189982727
- type: euclidean_spearman
value: 17.838933806777177
- type: main_score
value: 17.838933806777177
- type: manhattan_pearson
value: 17.061405031678706
- type: manhattan_spearman
value: 17.720413541179628
- type: pearson
value: 16.61424897880095
- type: spearman
value: 17.838933806777177
- task:
type: STS
dataset:
name: MTEB STS22 (es-it)
type: mteb/sts22-crosslingual-sts
config: es-it
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 33.923866543364184
- type: cosine_spearman
value: 28.270091354677568
- type: euclidean_pearson
value: 32.54215682097072
- type: euclidean_spearman
value: 28.270091354677568
- type: main_score
value: 28.270091354677568
- type: manhattan_pearson
value: 32.330367232207976
- type: manhattan_spearman
value: 28.255815608210906
- type: pearson
value: 33.923868062160935
- type: spearman
value: 28.270091354677568
- task:
type: STS
dataset:
name: MTEB STS22 (ar)
type: mteb/sts22-crosslingual-sts
config: ar
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 33.485734155602124
- type: cosine_spearman
value: 35.184256236066545
- type: euclidean_pearson
value: 35.72218023358308
- type: euclidean_spearman
value: 35.184256236066545
- type: main_score
value: 35.184256236066545
- type: manhattan_pearson
value: 35.6410009964212
- type: manhattan_spearman
value: 35.234893627175516
- type: pearson
value: 33.485735809605345
- type: spearman
value: 35.184256236066545
- task:
type: STS
dataset:
name: MTEB STS22 (de)
type: mteb/sts22-crosslingual-sts
config: de
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 9.234564624165541
- type: cosine_spearman
value: 9.26135671318589
- type: euclidean_pearson
value: 5.846102863095134
- type: euclidean_spearman
value: 9.26135671318589
- type: main_score
value: 9.26135671318589
- type: manhattan_pearson
value: 5.7129856416500635
- type: manhattan_spearman
value: 9.101317893245328
- type: pearson
value: 9.23456334729063
- type: spearman
value: 9.253696154102018
- task:
type: STS
dataset:
name: MTEB STS22 (de-pl)
type: mteb/sts22-crosslingual-sts
config: de-pl
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 7.541720515534752
- type: cosine_spearman
value: 16.228492544303737
- type: euclidean_pearson
value: 8.688753048966332
- type: euclidean_spearman
value: 16.228492544303737
- type: main_score
value: 16.228492544303737
- type: manhattan_pearson
value: 8.82151603264942
- type: manhattan_spearman
value: 14.072905299329804
- type: pearson
value: 7.5417136414683394
- type: spearman
value: 16.228492544303737
- task:
type: STS
dataset:
name: MTEB STS22 (zh)
type: mteb/sts22-crosslingual-sts
config: zh
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 14.272880349051686
- type: cosine_spearman
value: 26.923117280473786
- type: euclidean_pearson
value: 21.452060315735473
- type: euclidean_spearman
value: 26.923117280473786
- type: main_score
value: 26.923117280473786
- type: manhattan_pearson
value: 21.100522438025678
- type: manhattan_spearman
value: 26.877576505273343
- type: pearson
value: 14.272881506036967
- type: spearman
value: 26.92266584563422
- task:
type: STS
dataset:
name: MTEB STS22 (de-en)
type: mteb/sts22-crosslingual-sts
config: de-en
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 43.299601499947705
- type: cosine_spearman
value: 44.906191184092506
- type: euclidean_pearson
value: 44.27229004756392
- type: euclidean_spearman
value: 44.906191184092506
- type: main_score
value: 44.906191184092506
- type: manhattan_pearson
value: 43.30226127948028
- type: manhattan_spearman
value: 44.15123742556422
- type: pearson
value: 43.29959849491429
- type: spearman
value: 44.906191184092506
- task:
type: STS
dataset:
name: MTEB STS22 (zh-en)
type: mteb/sts22-crosslingual-sts
config: zh-en
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 38.208833040701016
- type: cosine_spearman
value: 40.35513412565045
- type: euclidean_pearson
value: 37.69713615937953
- type: euclidean_spearman
value: 40.35513412565045
- type: main_score
value: 40.35513412565045
- type: manhattan_pearson
value: 38.07295736531383
- type: manhattan_spearman
value: 41.18949838024728
- type: pearson
value: 38.20882899305228
- type: spearman
value: 40.35513412565045
- task:
type: STS
dataset:
name: MTEB STS22 (fr)
type: mteb/sts22-crosslingual-sts
config: fr
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 57.64495357984024
- type: cosine_spearman
value: 58.654277042743985
- type: euclidean_pearson
value: 57.6918916487343
- type: euclidean_spearman
value: 58.654277042743985
- type: main_score
value: 58.654277042743985
- type: manhattan_pearson
value: 57.44807489792652
- type: manhattan_spearman
value: 58.55625686584207
- type: pearson
value: 57.64495326981972
- type: spearman
value: 58.654277042743985
- task:
type: STS
dataset:
name: MTEB STS22 (tr)
type: mteb/sts22-crosslingual-sts
config: tr
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 11.450692016022549
- type: cosine_spearman
value: 17.51115938663503
- type: euclidean_pearson
value: 15.02321459776712
- type: euclidean_spearman
value: 17.51115938663503
- type: main_score
value: 17.51115938663503
- type: manhattan_pearson
value: 14.927559119791256
- type: manhattan_spearman
value: 17.573688652181342
- type: pearson
value: 11.450690790822087
- type: spearman
value: 17.51115938663503
- task:
type: STS
dataset:
name: MTEB STS22 (pl-en)
type: mteb/sts22-crosslingual-sts
config: pl-en
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 43.813735649943744
- type: cosine_spearman
value: 43.21656383381947
- type: euclidean_pearson
value: 44.23609795164775
- type: euclidean_spearman
value: 43.21656383381947
- type: main_score
value: 43.21656383381947
- type: manhattan_pearson
value: 45.15245284513481
- type: manhattan_spearman
value: 44.617872376647036
- type: pearson
value: 43.81373603921556
- type: spearman
value: 43.21656383381947
- task:
type: STS
dataset:
name: MTEB STS22 (pl)
type: mteb/sts22-crosslingual-sts
config: pl
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 2.064593579711082
- type: cosine_spearman
value: 13.671807943753796
- type: euclidean_pearson
value: -0.2014491570823139
- type: euclidean_spearman
value: 13.671807943753796
- type: main_score
value: 13.671807943753796
- type: manhattan_pearson
value: -0.47928238790519784
- type: manhattan_spearman
value: 13.539652287241891
- type: pearson
value: 2.064597029930494
- type: spearman
value: 13.877866791327195
- task:
type: STS
dataset:
name: MTEB STS22 (it)
type: mteb/sts22-crosslingual-sts
config: it
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 34.34600586953367
- type: cosine_spearman
value: 35.333432813362556
- type: euclidean_pearson
value: 35.55800799947818
- type: euclidean_spearman
value: 35.333432813362556
- type: main_score
value: 35.333432813362556
- type: manhattan_pearson
value: 35.52036355368523
- type: manhattan_spearman
value: 35.32333999151806
- type: pearson
value: 34.34600393440248
- type: spearman
value: 35.333432813362556
- task:
type: STS
dataset:
name: MTEB STS22 (de-fr)
type: mteb/sts22-crosslingual-sts
config: de-fr
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 34.38468852241422
- type: cosine_spearman
value: 37.517923206663426
- type: euclidean_pearson
value: 36.15747644065988
- type: euclidean_spearman
value: 37.517923206663426
- type: main_score
value: 37.517923206663426
- type: manhattan_pearson
value: 36.367891064808674
- type: manhattan_spearman
value: 36.98092141912518
- type: pearson
value: 34.3846838424612
- type: spearman
value: 37.517923206663426
- task:
type: STS
dataset:
name: MTEB STS22 (en)
type: mteb/sts22-crosslingual-sts
config: en
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 62.10106220976315
- type: cosine_spearman
value: 62.162497286103665
- type: euclidean_pearson
value: 63.47223854069964
- type: euclidean_spearman
value: 62.162497286103665
- type: main_score
value: 62.162497286103665
- type: manhattan_pearson
value: 63.20260003333065
- type: manhattan_spearman
value: 61.91516380449414
- type: pearson
value: 62.10106168936663
- type: spearman
value: 62.162497286103665
- task:
type: STS
dataset:
name: MTEB STS22 (es-en)
type: mteb/sts22-crosslingual-sts
config: es-en
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 43.060498443703956
- type: cosine_spearman
value: 46.96962646942495
- type: euclidean_pearson
value: 44.32895696900264
- type: euclidean_spearman
value: 46.96962646942495
- type: main_score
value: 46.96962646942495
- type: manhattan_pearson
value: 44.35969054771841
- type: manhattan_spearman
value: 47.06500736586804
- type: pearson
value: 43.060495996175085
- type: spearman
value: 46.96962646942495
- task:
type: STS
dataset:
name: MTEB STS22 (es)
type: mteb/sts22-crosslingual-sts
config: es
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 31.552876706984677
- type: cosine_spearman
value: 36.385581860846955
- type: euclidean_pearson
value: 35.09094631898299
- type: euclidean_spearman
value: 36.385581860846955
- type: main_score
value: 36.385581860846955
- type: manhattan_pearson
value: 34.90394958111092
- type: manhattan_spearman
value: 36.31134774911119
- type: pearson
value: 31.552876690362574
- type: spearman
value: 36.385581860846955
- task:
type: STS
dataset:
name: MTEB STS22 (fr-pl)
type: mteb/sts22-crosslingual-sts
config: fr-pl
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 69.17422308603082
- type: cosine_spearman
value: 61.97797868009122
- type: euclidean_pearson
value: 68.87185772300731
- type: euclidean_spearman
value: 61.97797868009122
- type: main_score
value: 61.97797868009122
- type: manhattan_pearson
value: 69.95446576863942
- type: manhattan_spearman
value: 61.97797868009122
- type: pearson
value: 69.17420713169005
- type: spearman
value: 61.97797868009122
- task:
type: STS
dataset:
name: MTEB STS22 (ru)
type: mteb/sts22-crosslingual-sts
config: ru
split: test
revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3
metrics:
- type: cosine_pearson
value: 7.810349558361704
- type: cosine_spearman
value: 13.082192851727692
- type: euclidean_pearson
value: 9.256112073506458
- type: euclidean_spearman
value: 13.082192851727692
- type: main_score
value: 13.082192851727692
- type: manhattan_pearson
value: 8.72571445455125
- type: manhattan_spearman
value: 12.72837879957086
- type: pearson
value: 7.8103471976464665
- type: spearman
value: 13.082192851727692
- task:
type: STS
dataset:
name: MTEB STSBenchmark (default)
type: mteb/stsbenchmark-sts
config: default
split: test
revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831
metrics:
- type: cosine_pearson
value: 81.60094971873578
- type: cosine_spearman
value: 82.21497157084686
- type: euclidean_pearson
value: 82.7649580865935
- type: euclidean_spearman
value: 82.21487652663633
- type: main_score
value: 82.21497157084686
- type: manhattan_pearson
value: 82.60879748854263
- type: manhattan_spearman
value: 82.02680374744259
- type: pearson
value: 81.60094958106916
- type: spearman
value: 82.21497157084686
- task:
type: Reranking
dataset:
name: MTEB SciDocsRR (default)
type: mteb/scidocs-reranking
config: default
split: test
revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab
metrics:
- type: main_score
value: 88.0620858216092
- type: map
value: 88.0620858216092
- type: mrr
value: 96.56171442936149
- type: nAUC_map_diff1
value: -10.48187303124293
- type: nAUC_map_max
value: 56.61425009792597
- type: nAUC_map_std
value: 71.46629366886685
- type: nAUC_mrr_diff1
value: 32.581743266040704
- type: nAUC_mrr_max
value: 88.42416536858423
- type: nAUC_mrr_std
value: 83.97838942252417
- task:
type: Retrieval
dataset:
name: MTEB SciFact (default)
type: mteb/scifact
config: default
split: test
revision: 0228b52cf27578f30900b9e5271d331663a030d7
metrics:
- type: main_score
value: 60.521
- type: map_at_1
value: 44.622
- type: map_at_10
value: 55.033
- type: map_at_100
value: 55.659000000000006
- type: map_at_1000
value: 55.701
- type: map_at_20
value: 55.379999999999995
- type: map_at_3
value: 51.918
- type: map_at_5
value: 53.563
- type: mrr_at_1
value: 46.666666666666664
- type: mrr_at_10
value: 56.386640211640184
- type: mrr_at_100
value: 56.86324391737727
- type: mrr_at_1000
value: 56.899770911998424
- type: mrr_at_20
value: 56.63502866958745
- type: mrr_at_3
value: 53.88888888888887
- type: mrr_at_5
value: 55.388888888888886
- type: nauc_map_at_1000_diff1
value: 60.51522831114532
- type: nauc_map_at_1000_max
value: 38.729857044583156
- type: nauc_map_at_1000_std
value: 11.893619390095981
- type: nauc_map_at_100_diff1
value: 60.49224806809106
- type: nauc_map_at_100_max
value: 38.757391171919394
- type: nauc_map_at_100_std
value: 11.904869622843274
- type: nauc_map_at_10_diff1
value: 60.59184266027688
- type: nauc_map_at_10_max
value: 38.814372515682855
- type: nauc_map_at_10_std
value: 12.062834601122749
- type: nauc_map_at_1_diff1
value: 65.53217534037644
- type: nauc_map_at_1_max
value: 28.820516516628203
- type: nauc_map_at_1_std
value: 0.31747544825532187
- type: nauc_map_at_20_diff1
value: 60.49567580134993
- type: nauc_map_at_20_max
value: 38.75172270044811
- type: nauc_map_at_20_std
value: 12.03257481376082
- type: nauc_map_at_3_diff1
value: 61.95672662726325
- type: nauc_map_at_3_max
value: 36.21518511557246
- type: nauc_map_at_3_std
value: 8.725184216964175
- type: nauc_map_at_5_diff1
value: 61.59116859538641
- type: nauc_map_at_5_max
value: 37.099329579199555
- type: nauc_map_at_5_std
value: 9.77500772774504
- type: nauc_mrr_at_1000_diff1
value: 59.42105746719931
- type: nauc_mrr_at_1000_max
value: 40.45270689627016
- type: nauc_mrr_at_1000_std
value: 14.875759703157328
- type: nauc_mrr_at_100_diff1
value: 59.39676827842953
- type: nauc_mrr_at_100_max
value: 40.471355257224204
- type: nauc_mrr_at_100_std
value: 14.869956496055684
- type: nauc_mrr_at_10_diff1
value: 59.259253351613026
- type: nauc_mrr_at_10_max
value: 40.51223230352826
- type: nauc_mrr_at_10_std
value: 15.184710455010213
- type: nauc_mrr_at_1_diff1
value: 64.15225217314949
- type: nauc_mrr_at_1_max
value: 32.78250943893229
- type: nauc_mrr_at_1_std
value: 6.2994995170778845
- type: nauc_mrr_at_20_diff1
value: 59.353790297025796
- type: nauc_mrr_at_20_max
value: 40.50201824596607
- type: nauc_mrr_at_20_std
value: 15.006822621780044
- type: nauc_mrr_at_3_diff1
value: 60.38367084336967
- type: nauc_mrr_at_3_max
value: 39.80345696060504
- type: nauc_mrr_at_3_std
value: 14.692239492205003
- type: nauc_mrr_at_5_diff1
value: 59.88907677461462
- type: nauc_mrr_at_5_max
value: 40.45929072557247
- type: nauc_mrr_at_5_std
value: 14.716015778108465
- type: nauc_ndcg_at_1000_diff1
value: 57.912136336340005
- type: nauc_ndcg_at_1000_max
value: 41.805732022231986
- type: nauc_ndcg_at_1000_std
value: 16.26204553755279
- type: nauc_ndcg_at_100_diff1
value: 57.20651790458488
- type: nauc_ndcg_at_100_max
value: 42.35739596941831
- type: nauc_ndcg_at_100_std
value: 16.634877514141273
- type: nauc_ndcg_at_10_diff1
value: 57.21191994351011
- type: nauc_ndcg_at_10_max
value: 42.90550640696327
- type: nauc_ndcg_at_10_std
value: 17.95798580533489
- type: nauc_ndcg_at_1_diff1
value: 64.15225217314949
- type: nauc_ndcg_at_1_max
value: 32.78250943893229
- type: nauc_ndcg_at_1_std
value: 6.2994995170778845
- type: nauc_ndcg_at_20_diff1
value: 57.194338622478966
- type: nauc_ndcg_at_20_max
value: 42.87305570893914
- type: nauc_ndcg_at_20_std
value: 17.722159014890266
- type: nauc_ndcg_at_3_diff1
value: 59.87720937097173
- type: nauc_ndcg_at_3_max
value: 39.382390067554226
- type: nauc_ndcg_at_3_std
value: 13.309064283277102
- type: nauc_ndcg_at_5_diff1
value: 59.43934315091936
- type: nauc_ndcg_at_5_max
value: 40.3233098967969
- type: nauc_ndcg_at_5_std
value: 13.866407342104736
- type: nauc_precision_at_1000_diff1
value: -25.868335387629987
- type: nauc_precision_at_1000_max
value: 30.021175057678224
- type: nauc_precision_at_1000_std
value: 48.10593849751903
- type: nauc_precision_at_100_diff1
value: -9.246187406942138
- type: nauc_precision_at_100_max
value: 38.74029847793067
- type: nauc_precision_at_100_std
value: 46.21609723039785
- type: nauc_precision_at_10_diff1
value: 14.57423342836458
- type: nauc_precision_at_10_max
value: 53.79340709160492
- type: nauc_precision_at_10_std
value: 51.46731880835919
- type: nauc_precision_at_1_diff1
value: 64.15225217314949
- type: nauc_precision_at_1_max
value: 32.78250943893229
- type: nauc_precision_at_1_std
value: 6.2994995170778845
- type: nauc_precision_at_20_diff1
value: 7.661113308500396
- type: nauc_precision_at_20_max
value: 48.52849772461544
- type: nauc_precision_at_20_std
value: 50.06541458940931
- type: nauc_precision_at_3_diff1
value: 45.45337537808087
- type: nauc_precision_at_3_max
value: 49.37622387908755
- type: nauc_precision_at_3_std
value: 31.44254548463311
- type: nauc_precision_at_5_diff1
value: 37.18964707686113
- type: nauc_precision_at_5_max
value: 52.78467184760005
- type: nauc_precision_at_5_std
value: 37.360215414394794
- type: nauc_recall_at_1000_diff1
value: 2.4976657329602574
- type: nauc_recall_at_1000_max
value: 89.30127606598155
- type: nauc_recall_at_1000_std
value: 88.21195144724587
- type: nauc_recall_at_100_diff1
value: 32.06651881995142
- type: nauc_recall_at_100_max
value: 61.723309041324924
- type: nauc_recall_at_100_std
value: 40.02293962790224
- type: nauc_recall_at_10_diff1
value: 43.70825842063566
- type: nauc_recall_at_10_max
value: 56.410600790424525
- type: nauc_recall_at_10_std
value: 37.41756199503754
- type: nauc_recall_at_1_diff1
value: 65.53217534037644
- type: nauc_recall_at_1_max
value: 28.820516516628203
- type: nauc_recall_at_1_std
value: 0.31747544825532187
- type: nauc_recall_at_20_diff1
value: 42.28136019198356
- type: nauc_recall_at_20_max
value: 58.728571008292086
- type: nauc_recall_at_20_std
value: 38.78438410167401
- type: nauc_recall_at_3_diff1
value: 55.25292186933368
- type: nauc_recall_at_3_max
value: 42.11001146172373
- type: nauc_recall_at_3_std
value: 18.099532745235926
- type: nauc_recall_at_5_diff1
value: 53.15377657121759
- type: nauc_recall_at_5_max
value: 45.75866806524819
- type: nauc_recall_at_5_std
value: 20.94017969011607
- type: ndcg_at_1
value: 46.666999999999994
- type: ndcg_at_10
value: 60.521
- type: ndcg_at_100
value: 63.466
- type: ndcg_at_1000
value: 64.517
- type: ndcg_at_20
value: 61.632
- type: ndcg_at_3
value: 54.832
- type: ndcg_at_5
value: 57.459
- type: precision_at_1
value: 46.666999999999994
- type: precision_at_10
value: 8.6
- type: precision_at_100
value: 1.023
- type: precision_at_1000
value: 0.11100000000000002
- type: precision_at_20
value: 4.567
- type: precision_at_3
value: 21.778
- type: precision_at_5
value: 14.667
- type: recall_at_1
value: 44.622
- type: recall_at_10
value: 75.98899999999999
- type: recall_at_100
value: 89.767
- type: recall_at_1000
value: 98.0
- type: recall_at_20
value: 80.167
- type: recall_at_3
value: 60.633
- type: recall_at_5
value: 67.161
- task:
type: Retrieval
dataset:
name: MTEB SciFact (default)
type: mteb/scifact
config: default
split: train
revision: 0228b52cf27578f30900b9e5271d331663a030d7
metrics:
- type: main_score
value: 60.968
- type: map_at_1
value: 45.56
- type: map_at_10
value: 55.931
- type: map_at_100
value: 56.742000000000004
- type: map_at_1000
value: 56.776
- type: map_at_20
value: 56.424
- type: map_at_3
value: 53.077
- type: map_at_5
value: 54.869
- type: mrr_at_1
value: 48.084054388133495
- type: mrr_at_10
value: 57.47807404791333
- type: mrr_at_100
value: 58.099160087560875
- type: mrr_at_1000
value: 58.12765167354059
- type: mrr_at_20
value: 57.819082752169095
- type: mrr_at_3
value: 55.356407086938574
- type: mrr_at_5
value: 56.74701277297069
- type: nauc_map_at_1000_diff1
value: 55.74211749136413
- type: nauc_map_at_1000_max
value: 38.56607968837275
- type: nauc_map_at_1000_std
value: 8.843116088465356
- type: nauc_map_at_100_diff1
value: 55.73174781950032
- type: nauc_map_at_100_max
value: 38.58911494708111
- type: nauc_map_at_100_std
value: 8.88289997368927
- type: nauc_map_at_10_diff1
value: 55.684855899450326
- type: nauc_map_at_10_max
value: 38.61699467700117
- type: nauc_map_at_10_std
value: 8.843324283472004
- type: nauc_map_at_1_diff1
value: 60.51537312610602
- type: nauc_map_at_1_max
value: 33.79559472004539
- type: nauc_map_at_1_std
value: 2.231571936720321
- type: nauc_map_at_20_diff1
value: 55.64541576881228
- type: nauc_map_at_20_max
value: 38.46528450367595
- type: nauc_map_at_20_std
value: 8.941393850487122
- type: nauc_map_at_3_diff1
value: 56.72487646415579
- type: nauc_map_at_3_max
value: 35.519310437502725
- type: nauc_map_at_3_std
value: 5.488759577412841
- type: nauc_map_at_5_diff1
value: 55.803777316483064
- type: nauc_map_at_5_max
value: 37.41665211733391
- type: nauc_map_at_5_std
value: 7.806708616433203
- type: nauc_mrr_at_1000_diff1
value: 54.81790389145318
- type: nauc_mrr_at_1000_max
value: 39.86866604814943
- type: nauc_mrr_at_1000_std
value: 9.635106150864706
- type: nauc_mrr_at_100_diff1
value: 54.81237221566927
- type: nauc_mrr_at_100_max
value: 39.8961644304281
- type: nauc_mrr_at_100_std
value: 9.675552350620915
- type: nauc_mrr_at_10_diff1
value: 54.61930534460625
- type: nauc_mrr_at_10_max
value: 39.92147432792213
- type: nauc_mrr_at_10_std
value: 9.757897616300196
- type: nauc_mrr_at_1_diff1
value: 59.09832131205337
- type: nauc_mrr_at_1_max
value: 37.460196328629216
- type: nauc_mrr_at_1_std
value: 4.083199168195615
- type: nauc_mrr_at_20_diff1
value: 54.71252617109308
- type: nauc_mrr_at_20_max
value: 39.816810748191386
- type: nauc_mrr_at_20_std
value: 9.74015716919154
- type: nauc_mrr_at_3_diff1
value: 54.84574115652329
- type: nauc_mrr_at_3_max
value: 38.93977163525027
- type: nauc_mrr_at_3_std
value: 7.5157914615030155
- type: nauc_mrr_at_5_diff1
value: 54.28866742264963
- type: nauc_mrr_at_5_max
value: 39.857609786127355
- type: nauc_mrr_at_5_std
value: 9.748147844235504
- type: nauc_ndcg_at_1000_diff1
value: 54.828301594419926
- type: nauc_ndcg_at_1000_max
value: 40.792175720018
- type: nauc_ndcg_at_1000_std
value: 11.993991378627298
- type: nauc_ndcg_at_100_diff1
value: 54.56732091443972
- type: nauc_ndcg_at_100_max
value: 41.520966863404
- type: nauc_ndcg_at_100_std
value: 13.093045429472133
- type: nauc_ndcg_at_10_diff1
value: 53.867885134870754
- type: nauc_ndcg_at_10_max
value: 41.26263077190897
- type: nauc_ndcg_at_10_std
value: 13.180315525575079
- type: nauc_ndcg_at_1_diff1
value: 59.09832131205337
- type: nauc_ndcg_at_1_max
value: 37.460196328629216
- type: nauc_ndcg_at_1_std
value: 4.083199168195615
- type: nauc_ndcg_at_20_diff1
value: 54.04981222269078
- type: nauc_ndcg_at_20_max
value: 40.804574285983215
- type: nauc_ndcg_at_20_std
value: 13.25303328684406
- type: nauc_ndcg_at_3_diff1
value: 54.98672360900976
- type: nauc_ndcg_at_3_max
value: 37.021872876909285
- type: nauc_ndcg_at_3_std
value: 7.369741800312488
- type: nauc_ndcg_at_5_diff1
value: 53.738484591338775
- type: nauc_ndcg_at_5_max
value: 39.476934276971114
- type: nauc_ndcg_at_5_std
value: 11.468428891779542
- type: nauc_precision_at_1000_diff1
value: -19.823126106144006
- type: nauc_precision_at_1000_max
value: 30.52146999323513
- type: nauc_precision_at_1000_std
value: 29.933270069329826
- type: nauc_precision_at_100_diff1
value: -6.440029104435929
- type: nauc_precision_at_100_max
value: 39.331835639782156
- type: nauc_precision_at_100_std
value: 34.94900154413054
- type: nauc_precision_at_10_diff1
value: 16.817759554481484
- type: nauc_precision_at_10_max
value: 49.249021274893906
- type: nauc_precision_at_10_std
value: 33.75171756578377
- type: nauc_precision_at_1_diff1
value: 59.09832131205337
- type: nauc_precision_at_1_max
value: 37.460196328629216
- type: nauc_precision_at_1_std
value: 4.083199168195615
- type: nauc_precision_at_20_diff1
value: 9.157889901211199
- type: nauc_precision_at_20_max
value: 41.571198827265164
- type: nauc_precision_at_20_std
value: 32.65877960233835
- type: nauc_precision_at_3_diff1
value: 39.681670852445436
- type: nauc_precision_at_3_max
value: 42.86111363688067
- type: nauc_precision_at_3_std
value: 14.510088215612296
- type: nauc_precision_at_5_diff1
value: 27.384565276080302
- type: nauc_precision_at_5_max
value: 47.40454163560573
- type: nauc_precision_at_5_std
value: 25.867913617850697
- type: nauc_recall_at_1000_diff1
value: 71.51075746149512
- type: nauc_recall_at_1000_max
value: 76.34440349098203
- type: nauc_recall_at_1000_std
value: 80.44096320223078
- type: nauc_recall_at_100_diff1
value: 51.71053059895534
- type: nauc_recall_at_100_max
value: 66.88452980621679
- type: nauc_recall_at_100_std
value: 53.499016266851065
- type: nauc_recall_at_10_diff1
value: 47.702505621629825
- type: nauc_recall_at_10_max
value: 48.18585866434868
- type: nauc_recall_at_10_std
value: 28.809966003226883
- type: nauc_recall_at_1_diff1
value: 60.51537312610602
- type: nauc_recall_at_1_max
value: 33.79559472004539
- type: nauc_recall_at_1_std
value: 2.231571936720321
- type: nauc_recall_at_20_diff1
value: 48.95315412443835
- type: nauc_recall_at_20_max
value: 47.846213097193406
- type: nauc_recall_at_20_std
value: 32.44656771521295
- type: nauc_recall_at_3_diff1
value: 51.09488539609923
- type: nauc_recall_at_3_max
value: 36.21610788720492
- type: nauc_recall_at_3_std
value: 10.351917684505048
- type: nauc_recall_at_5_diff1
value: 47.52872180072941
- type: nauc_recall_at_5_max
value: 42.41341732768043
- type: nauc_recall_at_5_std
value: 22.031877672306795
- type: ndcg_at_1
value: 48.083999999999996
- type: ndcg_at_10
value: 60.968
- type: ndcg_at_100
value: 64.68199999999999
- type: ndcg_at_1000
value: 65.522
- type: ndcg_at_20
value: 62.50599999999999
- type: ndcg_at_3
value: 55.95
- type: ndcg_at_5
value: 58.75899999999999
- type: precision_at_1
value: 48.083999999999996
- type: precision_at_10
value: 8.455
- type: precision_at_100
value: 1.051
- type: precision_at_1000
value: 0.11199999999999999
- type: precision_at_20
value: 4.604
- type: precision_at_3
value: 22.415
- type: precision_at_5
value: 15.129999999999999
- type: recall_at_1
value: 45.56
- type: recall_at_10
value: 74.714
- type: recall_at_100
value: 91.862
- type: recall_at_1000
value: 98.393
- type: recall_at_20
value: 80.449
- type: recall_at_3
value: 61.747
- type: recall_at_5
value: 68.381
- task:
type: PairClassification
dataset:
name: MTEB SprintDuplicateQuestions (default)
type: mteb/sprintduplicatequestions-pairclassification
config: default
split: test
revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
metrics:
- type: cosine_accuracy
value: 99.52376237623763
- type: cosine_accuracy_threshold
value: 83.40442776679993
- type: cosine_ap
value: 82.29539182575489
- type: cosine_f1
value: 73.40139709833423
- type: cosine_f1_threshold
value: 81.47163987159729
- type: cosine_precision
value: 79.32636469221835
- type: cosine_recall
value: 68.30000000000001
- type: dot_accuracy
value: 99.52376237623763
- type: dot_accuracy_threshold
value: 83.40442776679993
- type: dot_ap
value: 82.29539339884745
- type: dot_f1
value: 73.40139709833423
- type: dot_f1_threshold
value: 81.47163391113281
- type: dot_precision
value: 79.32636469221835
- type: dot_recall
value: 68.30000000000001
- type: euclidean_accuracy
value: 99.52376237623763
- type: euclidean_accuracy_threshold
value: 57.61163830757141
- type: euclidean_ap
value: 82.29539339884744
- type: euclidean_f1
value: 73.40139709833423
- type: euclidean_f1_threshold
value: 60.8742356300354
- type: euclidean_precision
value: 79.32636469221835
- type: euclidean_recall
value: 68.30000000000001
- type: main_score
value: 82.29539339884745
- type: manhattan_accuracy
value: 99.50990099009901
- type: manhattan_accuracy_threshold
value: 1265.035057067871
- type: manhattan_ap
value: 81.78663637659726
- type: manhattan_f1
value: 73.2620320855615
- type: manhattan_f1_threshold
value: 1333.2223892211914
- type: manhattan_precision
value: 78.73563218390804
- type: manhattan_recall
value: 68.5
- type: max_accuracy
value: 99.52376237623763
- type: max_ap
value: 82.29539339884745
- type: max_f1
value: 73.40139709833423
- type: max_precision
value: 79.32636469221835
- type: max_recall
value: 68.5
- type: similarity_accuracy
value: 99.52376237623763
- type: similarity_accuracy_threshold
value: 83.40443968772888
- type: similarity_ap
value: 82.2953820339594
- type: similarity_f1
value: 73.40139709833423
- type: similarity_f1_threshold
value: 81.47163987159729
- type: similarity_precision
value: 79.32636469221835
- type: similarity_recall
value: 68.30000000000001
- task:
type: PairClassification
dataset:
name: MTEB SprintDuplicateQuestions (default)
type: mteb/sprintduplicatequestions-pairclassification
config: default
split: validation
revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
metrics:
- type: cosine_accuracy
value: 99.4990099009901
- type: cosine_accuracy_threshold
value: 80.80509901046753
- type: cosine_ap
value: 82.11932803714892
- type: cosine_f1
value: 73.94724664507173
- type: cosine_f1_threshold
value: 75.20748376846313
- type: cosine_precision
value: 68.81998277347114
- type: cosine_recall
value: 79.9
- type: dot_accuracy
value: 99.4990099009901
- type: dot_accuracy_threshold
value: 80.80509901046753
- type: dot_ap
value: 82.11932803714892
- type: dot_f1
value: 73.94724664507173
- type: dot_f1_threshold
value: 75.20748376846313
- type: dot_precision
value: 68.81998277347114
- type: dot_recall
value: 79.9
- type: euclidean_accuracy
value: 99.4990099009901
- type: euclidean_accuracy_threshold
value: 61.959511041641235
- type: euclidean_ap
value: 82.11932803714893
- type: euclidean_f1
value: 73.94724664507173
- type: euclidean_f1_threshold
value: 70.41663527488708
- type: euclidean_precision
value: 68.81998277347114
- type: euclidean_recall
value: 79.9
- type: main_score
value: 82.11932803714893
- type: manhattan_accuracy
value: 99.48910891089109
- type: manhattan_accuracy_threshold
value: 1331.6292762756348
- type: manhattan_ap
value: 81.61037880327734
- type: manhattan_f1
value: 73.582295988935
- type: manhattan_f1_threshold
value: 1543.6763763427734
- type: manhattan_precision
value: 68.26347305389223
- type: manhattan_recall
value: 79.80000000000001
- type: max_accuracy
value: 99.4990099009901
- type: max_ap
value: 82.11932803714893
- type: max_f1
value: 73.94724664507173
- type: max_precision
value: 68.81998277347114
- type: max_recall
value: 79.9
- type: similarity_accuracy
value: 99.4990099009901
- type: similarity_accuracy_threshold
value: 80.80509901046753
- type: similarity_ap
value: 82.11932803714892
- type: similarity_f1
value: 73.94724664507173
- type: similarity_f1_threshold
value: 75.20748376846313
- type: similarity_precision
value: 68.81998277347114
- type: similarity_recall
value: 79.9
- task:
type: Reranking
dataset:
name: MTEB StackOverflowDupQuestions (default)
type: mteb/stackoverflowdupquestions-reranking
config: default
split: test
revision: e185fbe320c72810689fc5848eb6114e1ef5ec69
metrics:
- type: main_score
value: 50.48960206352392
- type: map
value: 50.48960206352392
- type: mrr
value: 51.35440539852305
- type: nAUC_map_diff1
value: 34.53252917797536
- type: nAUC_map_max
value: 12.790792149348837
- type: nAUC_map_std
value: 6.669172097020934
- type: nAUC_mrr_diff1
value: 35.200389562175985
- type: nAUC_mrr_max
value: 13.668483379551915
- type: nAUC_mrr_std
value: 6.28436956387114
- task:
type: Retrieval
dataset:
name: MTEB TRECCOVID (default)
type: mteb/trec-covid
config: default
split: test
revision: bb9466bac8153a0349341eb1b22e06409e78ef4e
metrics:
- type: main_score
value: 56.266000000000005
- type: map_at_1
value: 0.193
- type: map_at_10
value: 1.3050000000000002
- type: map_at_100
value: 7.3340000000000005
- type: map_at_1000
value: 19.07
- type: map_at_20
value: 2.246
- type: map_at_3
value: 0.46699999999999997
- type: map_at_5
value: 0.733
- type: mrr_at_1
value: 64.0
- type: mrr_at_10
value: 77.06666666666666
- type: mrr_at_100
value: 77.24651162790698
- type: mrr_at_1000
value: 77.24651162790698
- type: mrr_at_20
value: 77.2
- type: mrr_at_3
value: 75.33333333333334
- type: mrr_at_5
value: 76.73333333333332
- type: nauc_map_at_1000_diff1
value: -24.73401101431591
- type: nauc_map_at_1000_max
value: 50.802272967189666
- type: nauc_map_at_1000_std
value: 86.72287095417501
- type: nauc_map_at_100_diff1
value: -25.085120642600828
- type: nauc_map_at_100_max
value: 49.31639831109537
- type: nauc_map_at_100_std
value: 71.17329215601556
- type: nauc_map_at_10_diff1
value: -14.614132651074147
- type: nauc_map_at_10_max
value: 32.721906387813256
- type: nauc_map_at_10_std
value: 25.60590793742716
- type: nauc_map_at_1_diff1
value: -4.218583653029754
- type: nauc_map_at_1_max
value: 18.589859656136742
- type: nauc_map_at_1_std
value: -0.7303444219812425
- type: nauc_map_at_20_diff1
value: -17.63292260337728
- type: nauc_map_at_20_max
value: 36.89990317740711
- type: nauc_map_at_20_std
value: 36.25627402017833
- type: nauc_map_at_3_diff1
value: -10.619502082953264
- type: nauc_map_at_3_max
value: 26.141633359773707
- type: nauc_map_at_3_std
value: 9.472414121311855
- type: nauc_map_at_5_diff1
value: -10.032105512858019
- type: nauc_map_at_5_max
value: 30.19477111697671
- type: nauc_map_at_5_std
value: 13.615465098322698
- type: nauc_mrr_at_1000_diff1
value: 18.351107936542924
- type: nauc_mrr_at_1000_max
value: 32.724989617966784
- type: nauc_mrr_at_1000_std
value: 11.379920959476115
- type: nauc_mrr_at_100_diff1
value: 18.351107936542924
- type: nauc_mrr_at_100_max
value: 32.724989617966784
- type: nauc_mrr_at_100_std
value: 11.379920959476115
- type: nauc_mrr_at_10_diff1
value: 18.810199118309463
- type: nauc_mrr_at_10_max
value: 32.90746100091763
- type: nauc_mrr_at_10_std
value: 12.170087976539595
- type: nauc_mrr_at_1_diff1
value: 22.953123004215087
- type: nauc_mrr_at_1_max
value: 33.759100779154394
- type: nauc_mrr_at_1_std
value: 5.521139353684961
- type: nauc_mrr_at_20_diff1
value: 18.440581264618835
- type: nauc_mrr_at_20_max
value: 32.887391287492
- type: nauc_mrr_at_20_std
value: 11.562294043911466
- type: nauc_mrr_at_3_diff1
value: 13.326882737022542
- type: nauc_mrr_at_3_max
value: 31.03893993559256
- type: nauc_mrr_at_3_std
value: 12.094868082843375
- type: nauc_mrr_at_5_diff1
value: 15.553466717555938
- type: nauc_mrr_at_5_max
value: 31.399808850294487
- type: nauc_mrr_at_5_std
value: 13.184440540139914
- type: nauc_ndcg_at_1000_diff1
value: -21.89727935740918
- type: nauc_ndcg_at_1000_max
value: 46.822006422768894
- type: nauc_ndcg_at_1000_std
value: 81.91290246912763
- type: nauc_ndcg_at_100_diff1
value: -17.48350814577877
- type: nauc_ndcg_at_100_max
value: 46.79297448451001
- type: nauc_ndcg_at_100_std
value: 78.8401319727116
- type: nauc_ndcg_at_10_diff1
value: -1.3821765420279954
- type: nauc_ndcg_at_10_max
value: 39.99897786678473
- type: nauc_ndcg_at_10_std
value: 46.364065597834106
- type: nauc_ndcg_at_1_diff1
value: 13.291418162587759
- type: nauc_ndcg_at_1_max
value: 39.50553809090106
- type: nauc_ndcg_at_1_std
value: 17.240062285160228
- type: nauc_ndcg_at_20_diff1
value: -5.006533721396937
- type: nauc_ndcg_at_20_max
value: 40.08638499376968
- type: nauc_ndcg_at_20_std
value: 54.21211699915291
- type: nauc_ndcg_at_3_diff1
value: 1.1177340010993853
- type: nauc_ndcg_at_3_max
value: 31.249409630478247
- type: nauc_ndcg_at_3_std
value: 23.5436216578103
- type: nauc_ndcg_at_5_diff1
value: -1.1160397524543895
- type: nauc_ndcg_at_5_max
value: 34.660234833569106
- type: nauc_ndcg_at_5_std
value: 32.63370377000179
- type: nauc_precision_at_1000_diff1
value: -13.0177453890889
- type: nauc_precision_at_1000_max
value: 30.72811980155358
- type: nauc_precision_at_1000_std
value: 63.304048454768505
- type: nauc_precision_at_100_diff1
value: -17.976796177452165
- type: nauc_precision_at_100_max
value: 46.981157094069914
- type: nauc_precision_at_100_std
value: 82.25965581862002
- type: nauc_precision_at_10_diff1
value: -0.615847761785869
- type: nauc_precision_at_10_max
value: 43.227090827429116
- type: nauc_precision_at_10_std
value: 54.13755012361465
- type: nauc_precision_at_1_diff1
value: 22.953123004215087
- type: nauc_precision_at_1_max
value: 33.759100779154394
- type: nauc_precision_at_1_std
value: 5.521139353684961
- type: nauc_precision_at_20_diff1
value: -3.9540692673731157
- type: nauc_precision_at_20_max
value: 44.35467403164495
- type: nauc_precision_at_20_std
value: 60.82047452956477
- type: nauc_precision_at_3_diff1
value: 3.020914020139455
- type: nauc_precision_at_3_max
value: 30.52731931120773
- type: nauc_precision_at_3_std
value: 23.58338795209433
- type: nauc_precision_at_5_diff1
value: -1.4227357063785067
- type: nauc_precision_at_5_max
value: 37.08155978608297
- type: nauc_precision_at_5_std
value: 37.53293730607916
- type: nauc_recall_at_1000_diff1
value: -20.649212891265538
- type: nauc_recall_at_1000_max
value: 43.07097184316376
- type: nauc_recall_at_1000_std
value: 77.81281303329497
- type: nauc_recall_at_100_diff1
value: -25.20899327705024
- type: nauc_recall_at_100_max
value: 44.06686144213573
- type: nauc_recall_at_100_std
value: 64.51149000534564
- type: nauc_recall_at_10_diff1
value: -16.705615179862683
- type: nauc_recall_at_10_max
value: 27.971302554687405
- type: nauc_recall_at_10_std
value: 24.823568423744703
- type: nauc_recall_at_1_diff1
value: -4.218583653029754
- type: nauc_recall_at_1_max
value: 18.589859656136742
- type: nauc_recall_at_1_std
value: -0.7303444219812425
- type: nauc_recall_at_20_diff1
value: -18.61953310144389
- type: nauc_recall_at_20_max
value: 32.59370606161681
- type: nauc_recall_at_20_std
value: 33.65262269593963
- type: nauc_recall_at_3_diff1
value: -12.73760452556664
- type: nauc_recall_at_3_max
value: 23.76654241253409
- type: nauc_recall_at_3_std
value: 9.977574788163437
- type: nauc_recall_at_5_diff1
value: -13.251480959967921
- type: nauc_recall_at_5_max
value: 28.794884382741404
- type: nauc_recall_at_5_std
value: 15.784579094515388
- type: ndcg_at_1
value: 57.99999999999999
- type: ndcg_at_10
value: 56.266000000000005
- type: ndcg_at_100
value: 43.847
- type: ndcg_at_1000
value: 42.278
- type: ndcg_at_20
value: 52.928
- type: ndcg_at_3
value: 58.531
- type: ndcg_at_5
value: 58.792
- type: precision_at_1
value: 64.0
- type: precision_at_10
value: 60.6
- type: precision_at_100
value: 45.28
- type: precision_at_1000
value: 19.122
- type: precision_at_20
value: 56.00000000000001
- type: precision_at_3
value: 63.333
- type: precision_at_5
value: 64.4
- type: recall_at_1
value: 0.193
- type: recall_at_10
value: 1.5890000000000002
- type: recall_at_100
value: 10.881
- type: recall_at_1000
value: 40.844
- type: recall_at_20
value: 2.861
- type: recall_at_3
value: 0.5
- type: recall_at_5
value: 0.835
- task:
type: Retrieval
dataset:
name: MTEB Touche2020 (default)
type: mteb/touche2020
config: default
split: test
revision: a34f9a33db75fa0cbb21bb5cfc3dae8dc8bec93f
metrics:
- type: main_score
value: 18.801000000000002
- type: map_at_1
value: 1.522
- type: map_at_10
value: 7.117999999999999
- type: map_at_100
value: 12.491
- type: map_at_1000
value: 13.913
- type: map_at_20
value: 9.014999999999999
- type: map_at_3
value: 4.144
- type: map_at_5
value: 5.192
- type: mrr_at_1
value: 20.408163265306122
- type: mrr_at_10
value: 34.63637836086815
- type: mrr_at_100
value: 35.855625090003045
- type: mrr_at_1000
value: 35.9025357255321
- type: mrr_at_20
value: 35.528481245452234
- type: mrr_at_3
value: 29.931972789115648
- type: mrr_at_5
value: 31.87074829931973
- type: nauc_map_at_1000_diff1
value: 2.2039398587625088
- type: nauc_map_at_1000_max
value: -7.1185216957557005
- type: nauc_map_at_1000_std
value: 13.743946451796427
- type: nauc_map_at_100_diff1
value: 2.352298138176251
- type: nauc_map_at_100_max
value: -7.6894882871790236
- type: nauc_map_at_100_std
value: 9.423702448674415
- type: nauc_map_at_10_diff1
value: -0.7996006244796997
- type: nauc_map_at_10_max
value: -3.883617316378952
- type: nauc_map_at_10_std
value: -9.010162215584359
- type: nauc_map_at_1_diff1
value: -14.540804131417854
- type: nauc_map_at_1_max
value: -5.1573008851326065
- type: nauc_map_at_1_std
value: -7.656042570304873
- type: nauc_map_at_20_diff1
value: 2.4262621068257135
- type: nauc_map_at_20_max
value: -2.797719479935332
- type: nauc_map_at_20_std
value: -6.492723413671281
- type: nauc_map_at_3_diff1
value: -12.76977998910709
- type: nauc_map_at_3_max
value: -5.4884334672841995
- type: nauc_map_at_3_std
value: -8.027896734726257
- type: nauc_map_at_5_diff1
value: -3.649569377955254
- type: nauc_map_at_5_max
value: -11.595839414332854
- type: nauc_map_at_5_std
value: -8.830419718907086
- type: nauc_mrr_at_1000_diff1
value: -0.6197732098443051
- type: nauc_mrr_at_1000_max
value: -24.253402425330417
- type: nauc_mrr_at_1000_std
value: 10.37771989076477
- type: nauc_mrr_at_100_diff1
value: -0.6376238919106149
- type: nauc_mrr_at_100_max
value: -24.342213789915842
- type: nauc_mrr_at_100_std
value: 10.51412622043574
- type: nauc_mrr_at_10_diff1
value: 0.16993164166959274
- type: nauc_mrr_at_10_max
value: -25.6197772706814
- type: nauc_mrr_at_10_std
value: 9.042003512118251
- type: nauc_mrr_at_1_diff1
value: -4.3668409896483125
- type: nauc_mrr_at_1_max
value: -10.908434972927616
- type: nauc_mrr_at_1_std
value: 1.8363187629423776
- type: nauc_mrr_at_20_diff1
value: -0.4751612608893524
- type: nauc_mrr_at_20_max
value: -25.330726501014965
- type: nauc_mrr_at_20_std
value: 11.110113131699444
- type: nauc_mrr_at_3_diff1
value: 1.1477476210783253
- type: nauc_mrr_at_3_max
value: -18.50801052487565
- type: nauc_mrr_at_3_std
value: 7.812899024458835
- type: nauc_mrr_at_5_diff1
value: 0.6521469927673968
- type: nauc_mrr_at_5_max
value: -21.416210744034153
- type: nauc_mrr_at_5_std
value: 11.504256088841963
- type: nauc_ndcg_at_1000_diff1
value: 3.4056842894504022
- type: nauc_ndcg_at_1000_max
value: -20.75643760341645
- type: nauc_ndcg_at_1000_std
value: 40.59382673351137
- type: nauc_ndcg_at_100_diff1
value: 2.6307496691299748
- type: nauc_ndcg_at_100_max
value: -25.295593466700737
- type: nauc_ndcg_at_100_std
value: 31.39618142617403
- type: nauc_ndcg_at_10_diff1
value: 5.002963427746491
- type: nauc_ndcg_at_10_max
value: -21.675189721633814
- type: nauc_ndcg_at_10_std
value: -1.469843197474695
- type: nauc_ndcg_at_1_diff1
value: -9.16604671836183
- type: nauc_ndcg_at_1_max
value: -14.823569484107907
- type: nauc_ndcg_at_1_std
value: 2.7125738258252983
- type: nauc_ndcg_at_20_diff1
value: 4.383118964859878
- type: nauc_ndcg_at_20_max
value: -22.34000929195479
- type: nauc_ndcg_at_20_std
value: 1.7471883582083971
- type: nauc_ndcg_at_3_diff1
value: -4.534899565092995
- type: nauc_ndcg_at_3_max
value: -23.884347314346275
- type: nauc_ndcg_at_3_std
value: 5.406857104296577
- type: nauc_ndcg_at_5_diff1
value: 5.404901375866572
- type: nauc_ndcg_at_5_max
value: -25.33284917518358
- type: nauc_ndcg_at_5_std
value: 3.3375579303514655
- type: nauc_precision_at_1000_diff1
value: -0.4809491820541609
- type: nauc_precision_at_1000_max
value: 14.095928757973825
- type: nauc_precision_at_1000_std
value: 40.99850413556261
- type: nauc_precision_at_100_diff1
value: 2.4829117900025968
- type: nauc_precision_at_100_max
value: -19.358122626022247
- type: nauc_precision_at_100_std
value: 69.16347293290907
- type: nauc_precision_at_10_diff1
value: 17.1708621551976
- type: nauc_precision_at_10_max
value: -18.120442425032554
- type: nauc_precision_at_10_std
value: 3.345789829578046
- type: nauc_precision_at_1_diff1
value: -4.3668409896483125
- type: nauc_precision_at_1_max
value: -10.908434972927616
- type: nauc_precision_at_1_std
value: 1.8363187629423776
- type: nauc_precision_at_20_diff1
value: 14.726770245618805
- type: nauc_precision_at_20_max
value: -21.815069886538808
- type: nauc_precision_at_20_std
value: 18.892811452623157
- type: nauc_precision_at_3_diff1
value: 0.9886305353262563
- type: nauc_precision_at_3_max
value: -20.231039993615312
- type: nauc_precision_at_3_std
value: 6.358708074233125
- type: nauc_precision_at_5_diff1
value: 15.348665084122503
- type: nauc_precision_at_5_max
value: -23.255200414646325
- type: nauc_precision_at_5_std
value: 5.968161709012153
- type: nauc_recall_at_1000_diff1
value: 4.060744763807052
- type: nauc_recall_at_1000_max
value: -11.386644024112234
- type: nauc_recall_at_1000_std
value: 77.38341256255143
- type: nauc_recall_at_100_diff1
value: 3.0602490159843883
- type: nauc_recall_at_100_max
value: -23.133334130489928
- type: nauc_recall_at_100_std
value: 42.72840616542226
- type: nauc_recall_at_10_diff1
value: 8.794442377965375
- type: nauc_recall_at_10_max
value: -11.666849404762925
- type: nauc_recall_at_10_std
value: -7.829821703241252
- type: nauc_recall_at_1_diff1
value: -14.540804131417854
- type: nauc_recall_at_1_max
value: -5.1573008851326065
- type: nauc_recall_at_1_std
value: -7.656042570304873
- type: nauc_recall_at_20_diff1
value: 9.614735799856538
- type: nauc_recall_at_20_max
value: -14.765691684991822
- type: nauc_recall_at_20_std
value: 0.612424582124656
- type: nauc_recall_at_3_diff1
value: -6.910547196179005
- type: nauc_recall_at_3_max
value: -9.544659311113964
- type: nauc_recall_at_3_std
value: -5.046214274501399
- type: nauc_recall_at_5_diff1
value: 6.384770222182381
- type: nauc_recall_at_5_max
value: -19.233502105079932
- type: nauc_recall_at_5_std
value: -5.566077739772719
- type: ndcg_at_1
value: 18.367
- type: ndcg_at_10
value: 18.801000000000002
- type: ndcg_at_100
value: 30.296
- type: ndcg_at_1000
value: 42.018
- type: ndcg_at_20
value: 20.14
- type: ndcg_at_3
value: 20.96
- type: ndcg_at_5
value: 19.683999999999997
- type: precision_at_1
value: 20.408
- type: precision_at_10
value: 16.531000000000002
- type: precision_at_100
value: 6.49
- type: precision_at_1000
value: 1.443
- type: precision_at_20
value: 13.776
- type: precision_at_3
value: 22.448999999999998
- type: precision_at_5
value: 19.592000000000002
- type: recall_at_1
value: 1.522
- type: recall_at_10
value: 12.255
- type: recall_at_100
value: 41.64
- type: recall_at_1000
value: 77.46499999999999
- type: recall_at_20
value: 19.136
- type: recall_at_3
value: 5.113
- type: recall_at_5
value: 7.348000000000001
- task:
type: PairClassification
dataset:
name: MTEB TwitterSemEval2015 (default)
type: mteb/twittersemeval2015-pairclassification
config: default
split: test
revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1
metrics:
- type: cosine_accuracy
value: 86.35036061274364
- type: cosine_accuracy_threshold
value: 78.45808267593384
- type: cosine_ap
value: 74.3562958182802
- type: cosine_f1
value: 69.23173016864459
- type: cosine_f1_threshold
value: 75.69116353988647
- type: cosine_precision
value: 65.74139976275207
- type: cosine_recall
value: 73.11345646437995
- type: dot_accuracy
value: 86.35036061274364
- type: dot_accuracy_threshold
value: 78.45808267593384
- type: dot_ap
value: 74.35629143935617
- type: dot_f1
value: 69.23173016864459
- type: dot_f1_threshold
value: 75.69116353988647
- type: dot_precision
value: 65.74139976275207
- type: dot_recall
value: 73.11345646437995
- type: euclidean_accuracy
value: 86.35036061274364
- type: euclidean_accuracy_threshold
value: 65.63827991485596
- type: euclidean_ap
value: 74.35629336635377
- type: euclidean_f1
value: 69.23173016864459
- type: euclidean_f1_threshold
value: 69.72637176513672
- type: euclidean_precision
value: 65.74139976275207
- type: euclidean_recall
value: 73.11345646437995
- type: main_score
value: 74.3562958182802
- type: manhattan_accuracy
value: 86.30267628300649
- type: manhattan_accuracy_threshold
value: 1388.2848739624023
- type: manhattan_ap
value: 74.34214111115519
- type: manhattan_f1
value: 69.18905284399041
- type: manhattan_f1_threshold
value: 1506.6452026367188
- type: manhattan_precision
value: 66.27204638801642
- type: manhattan_recall
value: 72.37467018469657
- type: max_accuracy
value: 86.35036061274364
- type: max_ap
value: 74.3562958182802
- type: max_f1
value: 69.23173016864459
- type: max_precision
value: 66.27204638801642
- type: max_recall
value: 73.11345646437995
- type: similarity_accuracy
value: 86.35036061274364
- type: similarity_accuracy_threshold
value: 78.4580945968628
- type: similarity_ap
value: 74.3562875678525
- type: similarity_f1
value: 69.23173016864459
- type: similarity_f1_threshold
value: 75.69118142127991
- type: similarity_precision
value: 65.74139976275207
- type: similarity_recall
value: 73.11345646437995
- task:
type: PairClassification
dataset:
name: MTEB TwitterURLCorpus (default)
type: mteb/twitterurlcorpus-pairclassification
config: default
split: test
revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf
metrics:
- type: cosine_accuracy
value: 88.15539255637054
- type: cosine_accuracy_threshold
value: 78.5060167312622
- type: cosine_ap
value: 84.15130253353578
- type: cosine_f1
value: 76.15275390479039
- type: cosine_f1_threshold
value: 75.17023086547852
- type: cosine_precision
value: 73.97996224771308
- type: cosine_recall
value: 78.45703726516786
- type: dot_accuracy
value: 88.15539255637054
- type: dot_accuracy_threshold
value: 78.50601077079773
- type: dot_ap
value: 84.15130163532636
- type: dot_f1
value: 76.15275390479039
- type: dot_f1_threshold
value: 75.17023086547852
- type: dot_precision
value: 73.97996224771308
- type: dot_recall
value: 78.45703726516786
- type: euclidean_accuracy
value: 88.15539255637054
- type: euclidean_accuracy_threshold
value: 65.56521654129028
- type: euclidean_ap
value: 84.1513046482876
- type: euclidean_f1
value: 76.15275390479039
- type: euclidean_f1_threshold
value: 70.46952247619629
- type: euclidean_precision
value: 73.97996224771308
- type: euclidean_recall
value: 78.45703726516786
- type: main_score
value: 84.1513046482876
- type: manhattan_accuracy
value: 88.13404742500097
- type: manhattan_accuracy_threshold
value: 1414.7026062011719
- type: manhattan_ap
value: 84.12376985849109
- type: manhattan_f1
value: 76.16659806159488
- type: manhattan_f1_threshold
value: 1529.8437118530273
- type: manhattan_precision
value: 74.09537677466326
- type: manhattan_recall
value: 78.35694487218971
- type: max_accuracy
value: 88.15539255637054
- type: max_ap
value: 84.1513046482876
- type: max_f1
value: 76.16659806159488
- type: max_precision
value: 74.09537677466326
- type: max_recall
value: 78.45703726516786
- type: similarity_accuracy
value: 88.15539255637054
- type: similarity_accuracy_threshold
value: 78.5060167312622
- type: similarity_ap
value: 84.15130432409686
- type: similarity_f1
value: 76.15275390479039
- type: similarity_f1_threshold
value: 75.17023086547852
- type: similarity_precision
value: 73.97996224771308
- type: similarity_recall
value: 78.45703726516786
---
# FUE
This is a sentence transformer model based on all-mpnet-base-v2.
## Usage
```python
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is encoded"]
model = SentenceTransformer('FAE')
embeddings = model.encode(sentences)
print(embeddings)
```
| [
"BIOSSES",
"SCIFACT"
] |
bartowski/Einstein-v7-Qwen2-7B-GGUF | bartowski | text-generation | [
"gguf",
"axolotl",
"instruct",
"finetune",
"chatml",
"gpt4",
"synthetic data",
"science",
"physics",
"chemistry",
"biology",
"math",
"qwen",
"qwen2",
"text-generation",
"en",
"dataset:allenai/ai2_arc",
"dataset:camel-ai/physics",
"dataset:camel-ai/chemistry",
"dataset:camel-ai/biology",
"dataset:camel-ai/math",
"dataset:metaeval/reclor",
"dataset:openbookqa",
"dataset:mandyyyyii/scibench",
"dataset:derek-thomas/ScienceQA",
"dataset:TIGER-Lab/ScienceEval",
"dataset:jondurbin/airoboros-3.2",
"dataset:LDJnr/Capybara",
"dataset:Cot-Alpaca-GPT4-From-OpenHermes-2.5",
"dataset:STEM-AI-mtl/Electrical-engineering",
"dataset:knowrohit07/saraswati-stem",
"dataset:sablo/oasst2_curated",
"dataset:lmsys/lmsys-chat-1m",
"dataset:TIGER-Lab/MathInstruct",
"dataset:bigbio/med_qa",
"dataset:meta-math/MetaMathQA-40K",
"dataset:piqa",
"dataset:scibench",
"dataset:sciq",
"dataset:Open-Orca/SlimOrca",
"dataset:migtissera/Synthia-v1.3",
"dataset:allenai/WildChat",
"dataset:microsoft/orca-math-word-problems-200k",
"dataset:openchat/openchat_sharegpt4_dataset",
"dataset:teknium/GPTeacher-General-Instruct",
"dataset:m-a-p/CodeFeedback-Filtered-Instruction",
"dataset:totally-not-an-llm/EverythingLM-data-V3",
"dataset:HuggingFaceH4/no_robots",
"dataset:OpenAssistant/oasst_top1_2023-08-25",
"dataset:WizardLM/WizardLM_evol_instruct_70k",
"dataset:abacusai/SystemChat-1.1",
"dataset:H-D-T/Buzz-V1.2",
"base_model:Qwen/Qwen2-7B",
"base_model:quantized:Qwen/Qwen2-7B",
"license:other",
"endpoints_compatible",
"region:us",
"conversational"
] | "2024-06-25T23:11:04Z" | 2024-06-25T23:33:55+00:00 | 432 | 7 | ---
base_model: Qwen/Qwen2-7B
datasets:
- allenai/ai2_arc
- camel-ai/physics
- camel-ai/chemistry
- camel-ai/biology
- camel-ai/math
- metaeval/reclor
- openbookqa
- mandyyyyii/scibench
- derek-thomas/ScienceQA
- TIGER-Lab/ScienceEval
- jondurbin/airoboros-3.2
- LDJnr/Capybara
- Cot-Alpaca-GPT4-From-OpenHermes-2.5
- STEM-AI-mtl/Electrical-engineering
- knowrohit07/saraswati-stem
- sablo/oasst2_curated
- lmsys/lmsys-chat-1m
- TIGER-Lab/MathInstruct
- bigbio/med_qa
- meta-math/MetaMathQA-40K
- openbookqa
- piqa
- metaeval/reclor
- derek-thomas/ScienceQA
- scibench
- sciq
- Open-Orca/SlimOrca
- migtissera/Synthia-v1.3
- TIGER-Lab/ScienceEval
- allenai/WildChat
- microsoft/orca-math-word-problems-200k
- openchat/openchat_sharegpt4_dataset
- teknium/GPTeacher-General-Instruct
- m-a-p/CodeFeedback-Filtered-Instruction
- totally-not-an-llm/EverythingLM-data-V3
- HuggingFaceH4/no_robots
- OpenAssistant/oasst_top1_2023-08-25
- WizardLM/WizardLM_evol_instruct_70k
- abacusai/SystemChat-1.1
- H-D-T/Buzz-V1.2
language:
- en
license: other
pipeline_tag: text-generation
tags:
- axolotl
- instruct
- finetune
- chatml
- gpt4
- synthetic data
- science
- physics
- chemistry
- biology
- math
- qwen
- qwen2
quantized_by: bartowski
---
## Llamacpp imatrix Quantizations of Einstein-v7-Qwen2-7B
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3197">b3197</a> for quantization.
Original model: https://huggingface.co/Weyaxi/Einstein-v7-Qwen2-7B
All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
## Prompt format
```
<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
## Download a file (not the whole branch) from below:
| Filename | Quant type | File Size | Description |
| -------- | ---------- | --------- | ----------- |
| [Einstein-v7-Qwen2-7B-Q8_0_L.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q8_1.gguf) | Q8_0_L | 9.12GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Extremely high quality, generally unneeded but max available quant. |
| [Einstein-v7-Qwen2-7B-Q8_0.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q8_0.gguf) | Q8_0 | 8.09GB | Extremely high quality, generally unneeded but max available quant. |
| [Einstein-v7-Qwen2-7B-Q6_K_L.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q6_K_L.gguf) | Q6_K_L | 7.54GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Very high quality, near perfect, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q6_K.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q6_K.gguf) | Q6_K | 6.25GB | Very high quality, near perfect, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q5_K_L.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q5_K_L.gguf) | Q5_K_L | 6.80GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. High quality, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q5_K_M.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q5_K_M.gguf) | Q5_K_M | 5.44GB | High quality, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q5_K_S.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q5_K_S.gguf) | Q5_K_S | 5.31GB | High quality, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q4_K_L.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q4_K_L.gguf) | Q4_K_L | 6.10GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Good quality, uses about 4.83 bits per weight, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q4_K_M.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q4_K_M.gguf) | Q4_K_M | 4.68GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q4_K_S.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q4_K_S.gguf) | Q4_K_S | 4.45GB | Slightly lower quality with more space savings, *recommended*. |
| [Einstein-v7-Qwen2-7B-IQ4_XS.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ4_XS.gguf) | IQ4_XS | 4.21GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
| [Einstein-v7-Qwen2-7B-Q3_K_XL.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF//main/Einstein-v7-Qwen2-7B-Q3_K_XL.gguf) | Q3_K_XL | | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Lower quality but usable, good for low RAM availability. |
| [Einstein-v7-Qwen2-7B-Q3_K_L.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q3_K_L.gguf) | Q3_K_L | 4.08GB | Lower quality but usable, good for low RAM availability. |
| [Einstein-v7-Qwen2-7B-Q3_K_M.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q3_K_M.gguf) | Q3_K_M | 3.80GB | Even lower quality. |
| [Einstein-v7-Qwen2-7B-IQ3_M.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ3_M.gguf) | IQ3_M | 3.57GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
| [Einstein-v7-Qwen2-7B-Q3_K_S.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q3_K_S.gguf) | Q3_K_S | 3.49GB | Low quality, not recommended. |
| [Einstein-v7-Qwen2-7B-IQ3_XS.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ3_XS.gguf) | IQ3_XS | 3.34GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
| [Einstein-v7-Qwen2-7B-IQ3_XXS.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ3_XXS.gguf) | IQ3_XXS | 3.11GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
| [Einstein-v7-Qwen2-7B-Q2_K.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-Q2_K.gguf) | Q2_K | 3.01GB | Very low quality but surprisingly usable. |
| [Einstein-v7-Qwen2-7B-IQ2_M.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ2_M.gguf) | IQ2_M | 2.78GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
| [Einstein-v7-Qwen2-7B-IQ2_S.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ2_S.gguf) | IQ2_S | 2.59GB | Very low quality, uses SOTA techniques to be usable. |
| [Einstein-v7-Qwen2-7B-IQ2_XS.gguf](https://huggingface.co/bartowski/Einstein-v7-Qwen2-7B-GGUF/blob/main/Einstein-v7-Qwen2-7B-IQ2_XS.gguf) | IQ2_XS | 2.46GB | Very low quality, uses SOTA techniques to be usable. |
## Downloading using huggingface-cli
First, make sure you have hugginface-cli installed:
```
pip install -U "huggingface_hub[cli]"
```
Then, you can target the specific file you want:
```
huggingface-cli download bartowski/Einstein-v7-Qwen2-7B-GGUF --include "Einstein-v7-Qwen2-7B-Q4_K_M.gguf" --local-dir ./
```
If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
```
huggingface-cli download bartowski/Einstein-v7-Qwen2-7B-GGUF --include "Einstein-v7-Qwen2-7B-Q8_0.gguf/*" --local-dir Einstein-v7-Qwen2-7B-Q8_0
```
You can either specify a new local-dir (Einstein-v7-Qwen2-7B-Q8_0) or download them all in place (./)
## Which file should I choose?
A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
If you want to get more into the weeds, you can check out this extremely useful feature chart:
[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
| [
"SCIQ"
] |
knowledgator/gliner-bi-large-v1.0 | knowledgator | token-classification | [
"gliner",
"pytorch",
"NER",
"GLiNER",
"information extraction",
"encoder",
"entity recognition",
"token-classification",
"multilingual",
"dataset:urchade/pile-mistral-v0.1",
"dataset:numind/NuNER",
"dataset:knowledgator/GLINER-multi-task-synthetic-data",
"license:apache-2.0",
"region:us"
] | "2024-08-19T07:12:56Z" | 2024-08-25T11:37:40+00:00 | 432 | 23 | ---
datasets:
- urchade/pile-mistral-v0.1
- numind/NuNER
- knowledgator/GLINER-multi-task-synthetic-data
language:
- multilingual
library_name: gliner
license: apache-2.0
pipeline_tag: token-classification
tags:
- NER
- GLiNER
- information extraction
- encoder
- entity recognition
---
# About
GLiNER is a Named Entity Recognition (NER) model capable of identifying any entity type using a bidirectional transformer encoders (BERT-like). It provides a practical alternative to traditional NER models, which are limited to predefined entities, and Large Language Models (LLMs) that, despite their flexibility, are costly and large for resource-constrained scenarios.
This particular version utilize bi-encoder architecture, where textual encoder is [DeBERTa v3 large](microsoft/deberta-v3-large) and entity label encoder is sentence transformer - [BGE-base-en](https://huggingface.co/BAAI/bge-small-en-v1.5).
Such architecture brings several advantages over uni-encoder GLiNER:
* An unlimited amount of entities can be recognized at a single time;
* Faster inference if entity embeddings are preprocessed;
* Better generalization to unseen entities;
However, it has some drawbacks such as a lack of inter-label interactions that make it hard for the model to disambiguate semantically similar but contextually different entities.
### Installation & Usage
Install or update the gliner package:
```bash
pip install gliner -U
```
Once you've downloaded the GLiNER library, you can import the GLiNER class. You can then load this model using `GLiNER.from_pretrained` and predict entities with `predict_entities`.
```python
from gliner import GLiNER
model = GLiNER.from_pretrained("knowledgator/gliner-bi-large-v1.0")
text = """
Cristiano Ronaldo dos Santos Aveiro (Portuguese pronunciation: [kɾiʃˈtjɐnu ʁɔˈnaldu]; born 5 February 1985) is a Portuguese professional footballer who plays as a forward for and captains both Saudi Pro League club Al Nassr and the Portugal national team. Widely regarded as one of the greatest players of all time, Ronaldo has won five Ballon d'Or awards,[note 3] a record three UEFA Men's Player of the Year Awards, and four European Golden Shoes, the most by a European player. He has won 33 trophies in his career, including seven league titles, five UEFA Champions Leagues, the UEFA European Championship and the UEFA Nations League. Ronaldo holds the records for most appearances (183), goals (140) and assists (42) in the Champions League, goals in the European Championship (14), international goals (128) and international appearances (205). He is one of the few players to have made over 1,200 professional career appearances, the most by an outfield player, and has scored over 850 official senior career goals for club and country, making him the top goalscorer of all time.
"""
labels = ["person", "award", "date", "competitions", "teams"]
entities = model.predict_entities(text, labels, threshold=0.3)
for entity in entities:
print(entity["text"], "=>", entity["label"])
```
```
Cristiano Ronaldo dos Santos Aveiro => person
5 February 1985 => date
Al Nassr => teams
Portugal national team => teams
Ballon d'Or => award
UEFA Men's Player of the Year Awards => award
European Golden Shoes => award
UEFA Champions Leagues => competitions
UEFA European Championship => competitions
UEFA Nations League => competitions
Champions League => competitions
European Championship => competitions
```
If you have a large amount of entities and want to pre-embed them, please, refer to the following code snippet:
```python
labels = ["your entities"]
texts = ["your texts"]
entity_embeddings = model.encode_labels(labels, batch_size = 8)
outputs = model.batch_predict_with_embeds(texts, entity_embeddings, labels)
```
### Benchmarks
Below you can see the table with benchmarking results on various named entity recognition datasets:
| Dataset | Score |
|---------|-------|
| ACE 2004 | 29.1% |
| ACE 2005 | 32.7% |
| AnatEM | 35.1% |
| Broad Tweet Corpus | 64.9% |
| CoNLL 2003 | 62.8% |
| FabNER | 21.8% |
| FindVehicle | 37.1% |
| GENIA_NER | 56.2% |
| HarveyNER | 11.7% |
| MultiNERD | 58.8% |
| Ontonotes | 24.0% |
| PolyglotNER | 43.2% |
| TweetNER7 | 35.1% |
| WikiANN en | 54.8% |
| WikiNeural | 70.4% |
| bc2gm | 59.9% |
| bc4chemd | 48.2% |
| bc5cdr | 69.2% |
| ncbi | 67.0% |
| **Average** | **46.4%** |
|||
| CrossNER_AI | 49.2% |
| CrossNER_literature | 62.1% |
| CrossNER_music | 70.3% |
| CrossNER_politics | 70.0% |
| CrossNER_science | 65.7% |
| mit-movie | 36.9% |
| mit-restaurant | 42.5% |
| **Average (zero-shot benchmark)** | **56.7%** |
### Join Our Discord
Connect with our community on Discord for news, support, and discussion about our models. Join [Discord](https://discord.gg/dkyeAgs9DG). | [
"ANATEM",
"BC5CDR"
] |
lunahr/Phi-4-mini-instruct-abliterated | lunahr | text-generation | [
"transformers",
"safetensors",
"phi3",
"text-generation",
"nlp",
"code",
"abliterated",
"conversational",
"custom_code",
"multilingual",
"ar",
"zh",
"cs",
"da",
"nl",
"en",
"fi",
"fr",
"de",
"he",
"hu",
"it",
"ja",
"ko",
"no",
"pl",
"pt",
"ru",
"es",
"sv",
"th",
"tr",
"uk",
"base_model:microsoft/Phi-4-mini-instruct",
"base_model:finetune:microsoft/Phi-4-mini-instruct",
"license:mit",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | "2025-03-01T14:50:48Z" | 2025-03-11T16:54:48+00:00 | 429 | 4 | ---
base_model:
- microsoft/Phi-4-mini-instruct
language:
- multilingual
- ar
- zh
- cs
- da
- nl
- en
- fi
- fr
- de
- he
- hu
- it
- ja
- ko
- 'no'
- pl
- pt
- ru
- es
- sv
- th
- tr
- uk
library_name: transformers
license: mit
license_link: https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/main/LICENSE
pipeline_tag: text-generation
tags:
- nlp
- code
- abliterated
widget:
- messages:
- role: user
content: Can you provide ways to eat combinations of bananas and dragonfruits?
---
# Model Summary
An abliterated version of the below model:
Phi-4-mini-instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data. The model belongs to the Phi-4 model family and supports 128K token context length. The model underwent an enhancement process, incorporating both supervised fine-tuning and direct preference optimization to support precise instruction adherence and robust safety measures.
📰 [Phi-4-mini Microsoft Blog](https://aka.ms/phi4-feb2025) <br>
📖 [Phi-4-mini Technical Report](https://aka.ms/phi-4-multimodal/techreport) <br>
👩🍳 [Phi Cookbook](https://github.com/microsoft/PhiCookBook) <br>
🏡 [Phi Portal](https://azure.microsoft.com/en-us/products/phi) <br>
🖥️ Try It [Azure](https://aka.ms/phi-4-mini/azure), [Huggingface](https://huggingface.co/spaces/microsoft/phi-4-mini) <br>
**Phi-4**:
[[mini-instruct](https://huggingface.co/microsoft/Phi-4-mini-instruct) | [onnx](https://huggingface.co/microsoft/Phi-4-mini-instruct-onnx)];
[multimodal-instruct](https://huggingface.co/microsoft/Phi-4-multimodal-instruct);
## Intended Uses
### Primary Use Cases
The model is intended for broad multilingual commercial and research use. The model provides uses for general purpose AI systems and applications which require:
1) Memory/compute constrained environments
2) Latency bound scenarios
3) Strong reasoning (especially math and logic).
The model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features.
### Use Case Considerations
The model is not specifically designed or evaluated for all downstream purposes. Developers should consider common limitations of language models, as well as performance difference across languages, as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness before using within a specific downstream use case, particularly for high-risk scenarios.
Developers should be aware of and adhere to applicable laws or regulations (including but not limited to privacy, trade compliance laws, etc.) that are relevant to their use case.
***Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.***
## Release Notes
This release of Phi-4-mini-instruct is based on valuable user feedback from the Phi-3 series. The Phi-4-mini model employed new architecture for efficiency, larger vocabulary for multilingual support, and better post-training techniques were used for instruction following, function calling, as well as additional data leading to substantial gains on key capabilities. It is anticipated that most use cases will benefit from this release, but users are encouraged to test in their particular AI applications. The enthusiastic support for the Phi-4 series is greatly appreciated. Feedback on Phi-4-mini-instruct is welcomed and crucial to the model’s evolution and improvement.
### Model Quality
To understand the capabilities, the 3.8B parameters Phi-4-mini-instruct model was compared with a set of models over a variety of benchmarks using an internal benchmark platform (See Appendix A for benchmark methodology). A high-level overview of the model quality is as follows:
| Benchmark | Similar size | | | | |2x size | | | | | |
|----------------------------------|-------------|-------------------|-------------------|-------------------|-----------------|-----------------|-----------------|-----------------|-----------------|-----------------|-----------------|
| | Phi-4 mini-Ins | Phi-3.5-mini-Ins | Llama-3.2-3B-Ins | Mistral-3B | Qwen2.5-3B-Ins | Qwen2.5-7B-Ins | Mistral-8B-2410 | Llama-3.1-8B-Ins | Llama-3.1-Tulu-3-8B | Gemma2-9B-Ins | GPT-4o-mini-2024-07-18 |
| **Popular aggregated benchmark** | | | | | | | | | | | |
| Arena Hard | 32.8 | 34.4 | 17.0 | 26.9 | 32.0 | 55.5 | 37.3 | 25.7 | 42.7 | 43.7 | 53.7 |
| BigBench Hard (0-shot, CoT) | 70.4 | 63.1 | 55.4 | 51.2 | 56.2 | 72.4 | 53.3 | 63.4 | 55.5 | 65.7 | 80.4 |
| MMLU (5-shot) | 67.3 | 65.5 | 61.8 | 60.8 | 65.0 | 72.6 | 63.0 | 68.1 | 65.0 | 71.3 | 77.2 |
| MMLU-Pro (0-shot, CoT) | 52.8 | 47.4 | 39.2 | 35.3 | 44.7 | 56.2 | 36.6 | 44.0 | 40.9 | 50.1 | 62.8 |
| **Reasoning** | | | | | | | | | | | |
| ARC Challenge (10-shot) | 83.7 | 84.6 | 76.1 | 80.3 | 82.6 | 90.1 | 82.7 | 83.1 | 79.4 | 89.8 | 93.5 |
| BoolQ (2-shot) | 81.2 | 77.7 | 71.4 | 79.4 | 65.4 | 80.0 | 80.5 | 82.8 | 79.3 | 85.7 | 88.7 |
| GPQA (0-shot, CoT) | 25.2 | 26.6 | 24.3 | 24.4 | 23.4 | 30.6 | 26.3 | 26.3 | 29.9 | 39.1 | 41.1 |
| HellaSwag (5-shot) | 69.1 | 72.2 | 77.2 | 74.6 | 74.6 | 80.0 | 73.5 | 72.8 | 80.9 | 87.1 | 88.7 |
| OpenBookQA (10-shot) | 79.2 | 81.2 | 72.6 | 79.8 | 79.3 | 82.6 | 80.2 | 84.8 | 79.8 | 90.0 | 90.0 |
| PIQA (5-shot) | 77.6 | 78.2 | 68.2 | 73.2 | 72.6 | 76.2 | 81.2 | 83.2 | 78.3 | 83.7 | 88.7 |
| Social IQA (5-shot) | 72.5 | 75.1 | 68.3 | 73.9 | 75.3 | 75.3 | 77.6 | 71.8 | 73.4 | 74.7 | 82.9 |
| TruthfulQA (MC2) (10-shot) | 66.4 | 65.2 | 59.2 | 62.9 | 64.3 | 69.4 | 63.0 | 69.2 | 64.1 | 76.6 | 78.2 |
| Winogrande (5-shot) | 67.0 | 72.2 | 53.2 | 59.8 | 63.3 | 71.1 | 63.1 | 64.7 | 65.4 | 74.0 | 76.9 |
| **Multilingual** | | | | | | | | | | | |
| Multilingual MMLU (5-shot) | 49.3 | 51.8 | 48.1 | 46.4 | 55.9 | 64.4 | 53.7 | 56.2 | 54.5 | 63.8 | 72.9 |
| MGSM (0-shot, CoT) | 63.9 | 49.6 | 44.6 | 44.6 | 53.5 | 64.5 | 56.7 | 56.7 | 58.6 | 75.1 | 81.7 |
| **Math** | | | | | | | | | | | |
| GSM8K (8-shot, CoT) | 88.6 | 76.9 | 75.6 | 80.1 | 80.6 | 88.7 | 81.9 | 82.4 | 84.3 | 84.9 | 91.3 |
| MATH (0-shot, CoT) | 64.0 | 49.8 | 46.7 | 41.8 | 61.7 | 60.4 | 41.6 | 47.6 | 46.1 | 51.3 | 70.2 |
| **Overall** | **63.5** | **60.5** | **56.2** | **56.9** | **60.1** | **67.9** | **60.2** | **62.3** | **60.9** | **65.0** | **75.5** |
Overall, the model with only 3.8B-param achieves a similar level of multilingual language understanding and reasoning ability as much larger models. However, it is still fundamentally limited by its size for certain tasks. The model simply does not have the capacity to store too much factual knowledge, therefore, users may experience factual incorrectness. However, it may be possible to resolve such weakness by augmenting Phi-4 with a search engine, particularly when using the model under RAG settings.
## Usage
### Tokenizer
Phi-4-mini-instruct supports a vocabulary size of up to `200064` tokens. The [tokenizer files](https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/main/added_tokens.json) already provide placeholder tokens that can be used for downstream fine-tuning, but they can also be extended up to the model's vocabulary size.
### Input Formats
Given the nature of the training data, the Phi-4-mini-instruct
model is best suited for prompts using specific formats.
Below are the two primary formats:
#### Chat format
This format is used for general conversation and instructions:
```yaml
<|system|>Insert System Message<|end|><|user|>Insert User Message<|end|><|assistant|>
```
#### Tool-enabled function-calling format
This format is used when the user wants the model to provide function calls based on the given tools. The user should provide the available tools in the system prompt, wrapped by <|tool|> and <|/tool|> tokens. The tools should be specified in JSON format, using a JSON dump structure. Example:
`
<|system|>You are a helpful assistant with some tools.<|tool|>[{"name": "get_weather_updates", "description": "Fetches weather updates for a given city using the RapidAPI Weather API.", "parameters": {"city": {"description": "The name of the city for which to retrieve weather information.", "type": "str", "default": "London"}}}]<|/tool|><|end|><|user|>What is the weather like in Paris today?<|end|><|assistant|>
`
### Inference with vLLM
#### Requirements
List of required packages:
```
flash_attn==2.7.4.post1
torch==2.5.1
vllm>=0.7.3
```
#### Example
To perform inference using vLLM, you can use the following code snippet:
```python
from vllm import LLM, SamplingParams
llm = LLM(model="microsoft/Phi-4-mini-instruct", trust_remote_code=True)
messages = [
{"role": "system", "content": "You are a helpful AI assistant."},
{"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
{"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."},
{"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
]
sampling_params = SamplingParams(
max_tokens=500,
temperature=0.0,
)
output = llm.chat(messages=messages, sampling_params=sampling_params)
print(output[0].outputs[0].text)
```
### Inference with Transformers
#### Requirements
Phi-4 family has been integrated in the `4.49.0` version of `transformers`. The current `transformers` version can be verified with: `pip list | grep transformers`.
List of required packages:
```
flash_attn==2.7.4.post1
torch==2.5.1
transformers==4.49.0
accelerate==1.3.0
```
Phi-4-mini-instruct is also available in [Azure AI Studio]()
#### Example
After obtaining the Phi-4-mini-instruct chat models, users can use this sample code for inference.
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
torch.random.manual_seed(0)
model_path = "microsoft/Phi-4-mini-instruct"
model = AutoModelForCausalLM.from_pretrained(
model_path,
device_map="auto",
torch_dtype="auto",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(model_path)
messages = [
{"role": "system", "content": "You are a helpful AI assistant."},
{"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
{"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."},
{"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
]
# Prepare the input for the model
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
# Generate the response
generation_args = {
"max_new_tokens": 500,
"temperature": 0.0,
"do_sample": False,
}
output_ids = model.generate(input_ids, **generation_args)
# Decode the generated tokens to text
response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print(response)
```
## Responsible AI Considerations
Like other language models, the Phi family of models can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:
+ Quality of Service: The Phi models are trained primarily on English text and some additional multilingual text. Languages other than English will experience worse performance as well as performance disparities across non-English. English language varieties with less representation in the training data might experience worse performance than standard American English.
+ Multilingual performance and safety gaps: We believe it is important to make language models more widely available across different languages, but the Phi 4 models still exhibit challenges common across multilingual releases. As with any deployment of LLMs, developers will be better positioned to test for performance or safety gaps for their linguistic and cultural context and customize the model with additional fine-tuning and appropriate safeguards.
+ Representation of Harms & Perpetuation of Stereotypes: These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups, cultural contexts, or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
+ Inappropriate or Offensive Content: These models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the case.
+ Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
+ Limited Scope for Code: The majority of Phi 4 training data is based in Python and uses common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, it is strongly recommended that users manually verify all API uses.
+ Long Conversation: Phi 4 models, like other models, can in some cases generate responses that are repetitive, unhelpful, or inconsistent in very long chat sessions in both English and non-English languages. Developers are encouraged to place appropriate mitigations, like limiting conversation turns to account for the possible conversational drift.
Developers should apply responsible AI best practices, including mapping, measuring, and mitigating risks associated with their specific use case and cultural, linguistic context. Phi 4 family of models are general purpose models. As developers plan to deploy these models for specific use cases, they are encouraged to fine-tune the models for their use case and leverage the models as part of broader AI systems with language-specific safeguards in place. Important areas for consideration include:
+ Allocation: Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
+ High-Risk Scenarios: Developers should assess the suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
+ Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
+ Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
+ Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.
## Training
### Model
+ **Architecture:** Phi-4-mini-instruct has 3.8B parameters and is a dense decoder-only Transformer model. When compared with Phi-3.5-mini, the major changes with Phi-4-mini-instruct are 200K vocabulary, grouped-query attention, and shared input and output embedding.<br>
+ **Inputs:** Text. It is best suited for prompts using the chat format.<br>
+ **Context length:** 128K tokens<br>
+ **GPUs:** 512 A100-80G<br>
+ **Training time:** 21 days<br>
+ **Training data:** 5T tokens<br>
+ **Outputs:** Generated text in response to the input<br>
+ **Dates:** Trained between November and December 2024<br>
+ **Status:** This is a static model trained on offline datasets with the cutoff date of June 2024 for publicly available data.<br>
+ **Supported languages:** Arabic, Chinese, Czech, Danish, Dutch, English, Finnish, French, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Thai, Turkish, Ukrainian<br>
+ **Release date:** February 2025<br>
### Training Datasets
Phi-4-mini’s training data includes a wide variety of sources, totaling 5 trillion tokens, and is a combination of
1) publicly available documents filtered for quality, selected high-quality educational data, and code
2) newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (e.g., science, daily activities, theory of mind, etc.)
3) high quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness. Focus was placed on the quality of data that could potentially improve the reasoning ability for the model, and the publicly available documents were filtered to contain a preferred level of knowledge. As an example, the result of a game in premier league on a particular day might be good training data for frontier models, but such information was removed to leave more model capacity for reasoning for the model’s small size. More details about data can be found in the Phi-4-mini-instruct technical report.
The decontamination process involved normalizing and tokenizing the dataset, then generating and comparing n-grams between the target dataset and benchmark datasets. Samples with matching n-grams above a threshold were flagged as contaminated and removed from the dataset. A detailed contamination report was generated, summarizing the matched text, matching ratio, and filtered results for further analysis.
### Fine-tuning
A basic example of multi-GPUs supervised fine-tuning (SFT) with TRL and Accelerate modules is provided [here](https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/main/sample_finetune.py).
## Safety Evaluation and Red-Teaming
Various evaluation techniques including red teaming, adversarial conversation simulations, and multilingual safety evaluation benchmark datasets were leveraged to evaluate Phi-4 models’ propensity to produce undesirable outputs across multiple languages and risk categories. Several approaches were used to compensate for the limitations of one approach alone. Findings across the various evaluation methods indicate that safety post-training that was done as detailed in the Phi 3 Safety Post-Training paper had a positive impact across multiple languages and risk categories as observed by refusal rates (refusal to output undesirable outputs) and robustness to jailbreak techniques. Details on prior red team evaluations across Phi models can be found in the Phi 3 Safety Post-Training paper. For this release, the red team tested the model in English, Chinese, Japanese, Spanish, Portuguese, Arabic, Thai, and Russian for the following potential harms: Hate Speech and Bias, Violent Crimes, Specialized Advice, and Election Information. Their findings indicate that the model is resistant to jailbreak techniques across languages, but that language-specific attack prompts leveraging cultural context can cause the model to output harmful content. Another insight was that with function calling scenarios, the model could sometimes hallucinate function names or URL’s. The model may also be more susceptible to longer multi-turn jailbreak techniques across both English and non-English languages. These findings highlight the need for industry-wide investment in the development of high-quality safety evaluation datasets across multiple languages, including low resource languages, and risk areas that account for cultural nuances where those languages are spoken.
## Software
* [PyTorch](https://github.com/pytorch/pytorch)
* [Transformers](https://github.com/huggingface/transformers)
* [Flash-Attention](https://github.com/HazyResearch/flash-attention)
## Hardware
Note that by default, the Phi-4-mini-instruct model uses flash attention, which requires certain types of GPU hardware to run. We have tested on the following GPU types:
* NVIDIA A100
* NVIDIA A6000
* NVIDIA H100
If you want to run the model on:
* NVIDIA V100 or earlier generation GPUs: call AutoModelForCausalLM.from_pretrained() with attn_implementation="eager"
## License
The model is licensed under the [MIT license](./LICENSE).
## Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow [Microsoft’s Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks). Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.
## Appendix A: Benchmark Methodology
We include a brief word on methodology here - and in particular, how we think about optimizing prompts.
In an ideal world, we would never change any prompts in our benchmarks to ensure it is always an apples-to-apples comparison when comparing different models. Indeed, this is our default approach, and is the case in the vast majority of models we have run to date.
There are, however, some exceptions to this. In some cases, we see a model that performs worse than expected on a given eval due to a failure to respect the output format. For example:
+ A model may refuse to answer questions (for no apparent reason), or in coding tasks models may prefix their response with “Sure, I can help with that. …” which may break the parser. In such cases, we have opted to try different system messages (e.g. “You must always respond to a question” or “Get to the point!”).
+ With some models, we observed that few shots actually hurt model performance. In this case we did allow running the benchmarks with 0-shots for all cases.
+ We have tools to convert between chat and completions APIs. When converting a chat prompt to a completion prompt, some models have different keywords e.g. Human vs User. In these cases, we do allow for model-specific mappings for chat to completion prompts.
However, we do not:
+ Pick different few-shot examples. Few shots will always be the same when comparing different models.
+ Change prompt format: e.g. if it is an A/B/C/D multiple choice, we do not tweak this to 1/2/3/4 multiple choice.
### Benchmark datasets
The model was evaluated across a breadth of public and internal benchmarks to understand the model’s capabilities under multiple tasks and conditions. While most evaluations use English, the leading multilingual benchmark was incorporated that covers performance in select languages. More specifically,
+ Reasoning:
+ Winogrande: commonsense reasoning around pronoun resolution
+ PIQA: physical commonsense reasoning around everyday situations
+ ARC-challenge: grade-school multiple choice science questions
+ GPQA: very hard questions written and validated by experts in biology, physics, and chemistry
+ MedQA: medical questions answering
+ Social IQA: social commonsense intelligence
+ BoolQ: natural questions from context
+ TruthfulQA: grounded reasoning
+ Language understanding:
+ HellaSwag: commonsense natural language inference around everyday events
+ ANLI: adversarial natural language inference
+ Function calling:
+ Berkeley function calling function and tool call
+ Internal function calling benchmarks
+ World knowledge:
+ TriviaQA: trivia question on general topics
+ Math:
+ GSM8K: grade-school math word problems
+ GSM8K Hard: grade-school math word problems with large values and some absurdity.
+ MATH: challenging competition math problems
+ Code:
+ HumanEval HumanEval+, MBPP, MBPP+: python coding tasks
+ LiveCodeBenh, LiveBench: contamination-free code tasks
+ BigCode Bench: challenging programming tasks
+ Spider: SQL query tasks
+ Internal coding benchmarks
+ Instructions following:
+ IFEval: verifiable instructions
+ Internal instructions following benchmarks
+ Multilingual:
+ MGSM: multilingual grade-school math
+ Multilingual MMLU and MMLU-pro
+ MEGA: multilingual NLP tasks
+ Popular aggregated datasets: MMLU, MMLU-pro, BigBench-Hard, AGI Eval
+ Multi-turn conversations:
+ Data generated by in-house adversarial conversation simulation tool
+ Single-turn trustworthiness evaluation:
+ DecodingTrust: a collection of trustworthiness benchmarks in eight different perspectives
+ XSTest: exaggerated safety evaluation
+ Toxigen: adversarial and hate speech detection
+ Red Team:
+ Responses to prompts provided by AI Red Team at Microsoft | [
"MEDQA"
] |
mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF | mradermacher | null | [
"transformers",
"gguf",
"medical",
"climate",
"biology",
"chemistry",
"en",
"dataset:openlifescienceai/medmcqa",
"dataset:bigbio/med_qa",
"dataset:bigbio/pubmed_qa",
"dataset:empirischtech/med-qa-orpo-dpo",
"base_model:empirischtech/Llama-3.1-8B-Instruct-MedQA",
"base_model:quantized:empirischtech/Llama-3.1-8B-Instruct-MedQA",
"license:llama3.1",
"endpoints_compatible",
"region:us",
"conversational"
] | "2025-02-07T15:35:03Z" | 2025-02-11T23:39:31+00:00 | 428 | 1 | ---
base_model: empirischtech/Llama-3.1-8B-Instruct-MedQA
datasets:
- openlifescienceai/medmcqa
- bigbio/med_qa
- bigbio/pubmed_qa
- empirischtech/med-qa-orpo-dpo
language:
- en
library_name: transformers
license: llama3.1
tags:
- medical
- climate
- biology
- chemistry
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
static quants of https://huggingface.co/empirischtech/Llama-3.1-8B-Instruct-MedQA
<!-- provided-files -->
weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion.
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q2_K.gguf) | Q2_K | 3.3 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q3_K_S.gguf) | Q3_K_S | 3.8 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q3_K_M.gguf) | Q3_K_M | 4.1 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q3_K_L.gguf) | Q3_K_L | 4.4 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.IQ4_XS.gguf) | IQ4_XS | 4.6 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q4_K_S.gguf) | Q4_K_S | 4.8 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q4_K_M.gguf) | Q4_K_M | 5.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q5_K_S.gguf) | Q5_K_S | 5.7 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q5_K_M.gguf) | Q5_K_M | 5.8 | |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q6_K.gguf) | Q6_K | 6.7 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.Q8_0.gguf) | Q8_0 | 8.6 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Llama-3.1-8B-Instruct-MedQA-GGUF/resolve/main/Llama-3.1-8B-Instruct-MedQA.f16.gguf) | f16 | 16.2 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
| [
"MEDQA"
] |
TheBloke/juanako-7B-UNA-GGUF | TheBloke | null | [
"transformers",
"gguf",
"mistral",
"alignment-handbook",
"generated_from_trainer",
"juanako",
"UNA",
"dataset:HuggingFaceH4/ultrafeedback_binarized",
"arxiv:2109.07958",
"arxiv:2310.16944",
"arxiv:2305.18290",
"base_model:fblgit/juanako-7b-UNA",
"base_model:quantized:fblgit/juanako-7b-UNA",
"license:apache-2.0",
"model-index",
"region:us",
"conversational"
] | "2023-12-01T19:56:04Z" | 2023-12-01T20:00:24+00:00 | 425 | 12 | ---
base_model: fblgit/juanako-7b-UNA
datasets:
- HuggingFaceH4/ultrafeedback_binarized
license: apache-2.0
tags:
- alignment-handbook
- generated_from_trainer
- juanako
- mistral
- UNA
inference: false
model_creator: FBL
model_type: mistral
prompt_template: '<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
'
quantized_by: TheBloke
model-index:
- name: juanako-7b-UNA
results:
- task:
type: text-generation
name: TruthfulQA (MC2)
dataset:
name: truthful_qa
type: text-generation
config: multiple_choice
split: validation
metrics:
- type: accuracy
value: 65.13
verified: true
- task:
type: text-generation
name: ARC-Challenge
dataset:
name: ai2_arc
type: text-generation
config: ARC-Challenge
split: test
metrics:
- type: accuracy
value: 68.17
verified: true
- task:
type: text-generation
name: HellaSwag
dataset:
name: Rowan/hellaswag
type: text-generation
split: test
metrics:
- type: accuracy
value: 85.34
verified: true
- type: accuracy
value: 83.57
- task:
type: text-generation
name: Winogrande
dataset:
name: winogrande
type: text-generation
config: winogrande_debiased
split: test
metrics:
- type: accuracy
value: 78.85
verified: true
- task:
type: text-generation
name: MMLU
dataset:
name: cais/mmlu
type: text-generation
config: all
split: test
metrics:
- type: accuracy
value: 62.47
verified: true
- task:
type: text-generation
name: DROP
dataset:
name: drop
type: text-generation
split: validation
metrics:
- type: accuracy
value: 38.74
verified: true
- task:
type: text-generation
name: PubMedQA
dataset:
name: bigbio/pubmed_qa
type: text-generation
config: pubmed_qa_artificial_bigbio_qa
split: validation
metrics:
- type: accuracy
value: 76.0
---
<!-- markdownlint-disable MD041 -->
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<div style="display: flex; justify-content: space-between; width: 100%;">
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
</div>
<div style="display: flex; flex-direction: column; align-items: flex-end;">
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
</div>
</div>
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# Juanako 7B UNA - GGUF
- Model creator: [FBL](https://huggingface.co/fblgit)
- Original model: [Juanako 7B UNA](https://huggingface.co/fblgit/juanako-7b-UNA)
<!-- description start -->
## Description
This repo contains GGUF format model files for [FBL's Juanako 7B UNA](https://huggingface.co/fblgit/juanako-7b-UNA).
These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
<!-- description end -->
<!-- README_GGUF.md-about-gguf start -->
### About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplete list of clients and libraries that are known to support GGUF:
* [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
* [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
* [GPT4All](https://gpt4all.io/index.html), a free and open source local running GUI, supporting Windows, Linux and macOS with full GPU accel.
* [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration. Linux available, in beta as of 27/11/2023.
* [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
* [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
* [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
* [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server. Note, as of time of writing (November 27th 2023), ctransformers has not been updated in a long time and does not support many recent models.
<!-- README_GGUF.md-about-gguf end -->
<!-- repositories-available start -->
## Repositories available
* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/juanako-7B-UNA-AWQ)
* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/juanako-7B-UNA-GPTQ)
* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF)
* [FBL's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/fblgit/juanako-7b-UNA)
<!-- repositories-available end -->
<!-- prompt-template start -->
## Prompt template: ChatML
```
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
<!-- prompt-template end -->
<!-- compatibility_gguf start -->
## Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
## Explanation of quantisation methods
<details>
<summary>Click to see details</summary>
The new methods available are:
* GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
* GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
* GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
* GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
</details>
<!-- compatibility_gguf end -->
<!-- README_GGUF.md-provided-files start -->
## Provided files
| Name | Quant method | Bits | Size | Max RAM required | Use case |
| ---- | ---- | ---- | ---- | ---- | ----- |
| [juanako-7b-una.Q2_K.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q2_K.gguf) | Q2_K | 2 | 3.08 GB| 5.58 GB | smallest, significant quality loss - not recommended for most purposes |
| [juanako-7b-una.Q3_K_S.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q3_K_S.gguf) | Q3_K_S | 3 | 3.16 GB| 5.66 GB | very small, high quality loss |
| [juanako-7b-una.Q3_K_M.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q3_K_M.gguf) | Q3_K_M | 3 | 3.52 GB| 6.02 GB | very small, high quality loss |
| [juanako-7b-una.Q3_K_L.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q3_K_L.gguf) | Q3_K_L | 3 | 3.82 GB| 6.32 GB | small, substantial quality loss |
| [juanako-7b-una.Q4_0.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q4_0.gguf) | Q4_0 | 4 | 4.11 GB| 6.61 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| [juanako-7b-una.Q4_K_S.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q4_K_S.gguf) | Q4_K_S | 4 | 4.14 GB| 6.64 GB | small, greater quality loss |
| [juanako-7b-una.Q4_K_M.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q4_K_M.gguf) | Q4_K_M | 4 | 4.37 GB| 6.87 GB | medium, balanced quality - recommended |
| [juanako-7b-una.Q5_0.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q5_0.gguf) | Q5_0 | 5 | 5.00 GB| 7.50 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| [juanako-7b-una.Q5_K_S.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q5_K_S.gguf) | Q5_K_S | 5 | 5.00 GB| 7.50 GB | large, low quality loss - recommended |
| [juanako-7b-una.Q5_K_M.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q5_K_M.gguf) | Q5_K_M | 5 | 5.13 GB| 7.63 GB | large, very low quality loss - recommended |
| [juanako-7b-una.Q6_K.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q6_K.gguf) | Q6_K | 6 | 5.94 GB| 8.44 GB | very large, extremely low quality loss |
| [juanako-7b-una.Q8_0.gguf](https://huggingface.co/TheBloke/juanako-7B-UNA-GGUF/blob/main/juanako-7b-una.Q8_0.gguf) | Q8_0 | 8 | 7.70 GB| 10.20 GB | very large, extremely low quality loss - not recommended |
**Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
<!-- README_GGUF.md-provided-files end -->
<!-- README_GGUF.md-how-to-download start -->
## How to download GGUF files
**Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
* LM Studio
* LoLLMS Web UI
* Faraday.dev
### In `text-generation-webui`
Under Download Model, you can enter the model repo: TheBloke/juanako-7B-UNA-GGUF and below it, a specific filename to download, such as: juanako-7b-una.Q4_K_M.gguf.
Then click Download.
### On the command line, including multiple files at once
I recommend using the `huggingface-hub` Python library:
```shell
pip3 install huggingface-hub
```
Then you can download any individual model file to the current directory, at high speed, with a command like this:
```shell
huggingface-cli download TheBloke/juanako-7B-UNA-GGUF juanako-7b-una.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
<details>
<summary>More advanced huggingface-cli download usage (click to read)</summary>
You can also download multiple files at once with a pattern:
```shell
huggingface-cli download TheBloke/juanako-7B-UNA-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
```
For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
```shell
pip3 install hf_transfer
```
And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
```shell
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/juanako-7B-UNA-GGUF juanako-7b-una.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
```
Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
</details>
<!-- README_GGUF.md-how-to-download end -->
<!-- README_GGUF.md-how-to-run start -->
## Example `llama.cpp` command
Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
```shell
./main -ngl 35 -m juanako-7b-una.Q4_K_M.gguf --color -c 32768 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
```
Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change `-c 32768` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically. Note that longer sequence lengths require much more resources, so you may need to reduce this value.
If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
## How to run in `text-generation-webui`
Further instructions can be found in the text-generation-webui documentation, here: [text-generation-webui/docs/04 ‐ Model Tab.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/04%20%E2%80%90%20Model%20Tab.md#llamacpp).
## How to run from Python code
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.
### How to load this model in Python code, using llama-cpp-python
For full documentation, please see: [llama-cpp-python docs](https://abetlen.github.io/llama-cpp-python/).
#### First install the package
Run one of the following commands, according to your system:
```shell
# Base ctransformers with no GPU acceleration
pip install llama-cpp-python
# With NVidia CUDA acceleration
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
# Or with OpenBLAS acceleration
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
# Or with CLBLast acceleration
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
# Or with AMD ROCm GPU acceleration (Linux only)
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
# Or with Metal GPU acceleration for macOS systems only
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
pip install llama-cpp-python
```
#### Simple llama-cpp-python example code
```python
from llama_cpp import Llama
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = Llama(
model_path="./juanako-7b-una.Q4_K_M.gguf", # Download the model file first
n_ctx=32768, # The max sequence length to use - note that longer sequence lengths require much more resources
n_threads=8, # The number of CPU threads to use, tailor to your system and the resulting performance
n_gpu_layers=35 # The number of layers to offload to GPU, if you have GPU acceleration available
)
# Simple inference example
output = llm(
"<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant", # Prompt
max_tokens=512, # Generate up to 512 tokens
stop=["</s>"], # Example stop token - not necessarily correct for this specific model! Please check before using.
echo=True # Whether to echo the prompt
)
# Chat Completion API
llm = Llama(model_path="./juanako-7b-una.Q4_K_M.gguf", chat_format="llama-2") # Set chat_format according to the model you are using
llm.create_chat_completion(
messages = [
{"role": "system", "content": "You are a story writing assistant."},
{
"role": "user",
"content": "Write a story about llamas."
}
]
)
```
## How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
* [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
* [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
<!-- README_GGUF.md-how-to-run end -->
<!-- footer start -->
<!-- 200823 -->
## Discord
For further support, and discussions on these models and AI in general, join us at:
[TheBloke AI's Discord server](https://discord.gg/theblokeai)
## Thanks, and how to contribute
Thanks to the [chirper.ai](https://chirper.ai) team!
Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
* Patreon: https://patreon.com/TheBlokeAI
* Ko-Fi: https://ko-fi.com/TheBlokeAI
**Special thanks to**: Aemon Algiz.
**Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
<!-- footer end -->
<!-- original-model-card start -->
# Original model card: FBL's Juanako 7B UNA
# juanako-7b-UNA (Uniform Neural Alignment)
This model is a fine-tuned version of [fblgit/juanako-7b-UNA-v2-phase-1](https://huggingface.co/fblgit/juanako-7b-UNA-v2-phase-1) on the HuggingFaceH4/ultrafeedback_binarized dataset.
It outperforms in many aspects most of the current Mistral based models and is the **latest and most powerful juanako version as of now**.
## Scores
The official HuggingFace results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/results/blob/main/fblgit/juanako-7b-UNA/results_2023-11-28T08-33-33.965228.json)
| Model | Average ⬆️| ARC (25-s) ⬆️ | HellaSwag (10-s) ⬆️ | MMLU (5-s) ⬆️| TruthfulQA (MC) (0-s) ⬆️ | Winogrande (5-s) | GSM8K (5-s) | DROP (3-s) |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
|[mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) | 50.32 | 59.58 | 83.31 | 64.16 | 42.15 | 78.37 | 18.12 | 6.14 |
| [Intel/neural-chat-7b-v3-1](https://huggingface.co/Intel/neural-chat-7b-v3-1) | 59.0 | 66.21 | 83.64 | 62.37 | 59.65 | 78.14 | 19.56 | 43.84 |
| [fblgit/juanako-7b-UNA](https://huggingface.co/fblgit/juanako-7b-UNA) | **59.91** | **68.17** | **85.34** | 62.47 | **65.13** | **78.85** | **20.7** | 38.74 |
It scores: **59.91** according HuggingFace LLM Leaderboard.
It scores: **65.1** with `big-refactor` branch of lm-eval-harness
Author [Xavier M.](mailto:xavi@juanako.ai) @fblgit
## Model description
juanako uses UNA, Uniform Neural Alignment. A training technique that ease alignment between transformer layers yet to be published.
### Prompts
The following prompts showed positive results, it may depend the task and needs further experimentation but this should work for starters:
```
<|im_start|>system
- You are a helpful assistant chatbot trained by MosaicML.
- You answer questions.
- You are excited to be able to help the user, but will refuse to do anything that could be considered harmful to the user.
- You are more than just an information source, you are also able to write poetry, short stories, and make jokes.<|im_end|>
<|im_start|>user
Explain QKV<|im_end|>
<|im_start|>assistant
```
```
### Assistant: I am StableVicuna, a large language model created by CarperAI. I am here to chat!
### Human: Explain QKV
### Assistant:
```
```
[Round <|round|>]
问：Explain QKV
答：
```
```
[Round <|round|>]
Question：Explain QKV
Answer：
```
```
Question：Explain QKV
Answer：
```
## Evaluations (lm-eval big-refactor branch)
### TruthfulQA 0-Shot
```
| Tasks |Version|Filter|Metric|Value | |Stderr|
|--------------|-------|------|------|-----:|---|-----:|
|truthfulqa_mc2|Yaml |none |acc |0.6549|± |0.0153|
```
### ARC 25-Shot
```
| Tasks |Version|Filter| Metric |Value | |Stderr|
|-------------|-------|------|--------|-----:|---|-----:|
|arc_challenge|Yaml |none |acc |0.6476|± |0.0140|
| | |none |acc_norm|0.6809|± |0.0136|
```
### HellaSwag 10-Shot
```
| Tasks |Version|Filter| Metric |Value | |Stderr|
|---------|-------|------|--------|-----:|---|-----:|
|hellaswag|Yaml |none |acc |0.6703|± |0.0047|
| | |none |acc_norm|0.8520|± |0.0035|
```
### GSM8k 5-Shot
```
|Tasks|Version| Filter | Metric |Value | |Stderr|
|-----|-------|----------|-----------|-----:|---|-----:|
|gsm8k|Yaml |get-answer|exact_match|0.4898|± |0.0138|
```
### GPT Evaluations 0-Shot
```
| Tasks |Version|Filter| Metric |Value | |Stderr|
|--------------|-------|------|----------|-----:|---|-----:|
|boolq |Yaml |none |acc |0.8703|± |0.0059|
|lambada_openai|Yaml |none |perplexity|3.2598|± |0.0705|
| | |none |acc |0.7336|± |0.0062|
|piqa |Yaml |none |acc |0.8254|± |0.0089|
| | |none |acc_norm |0.8292|± |0.0088|
|sciq |Yaml |none |acc |0.9580|± |0.0063|
| | |none |acc_norm |0.9130|± |0.0089|
```
### MathQA 0-Shot
```
|Tasks |Version|Filter| Metric |Value | |Stderr|
|------|-------|------|--------|-----:|---|-----:|
|mathqa|Yaml |none |acc |0.3752|± |0.0089|
| | |none |acc_norm|0.3772|± |0.0089|
```
### PiQa 1-Shot
```
|Tasks|Version|Filter| Metric |Value | |Stderr|
|-----|-------|------|--------|-----:|---|-----:|
|piqa |Yaml |none |acc |0.8308|± |0.0087|
| | |none |acc_norm|0.8357|± |0.0086|
```
### Winogrande 5-Shot
```
| Tasks |Version|Filter|Metric|Value| |Stderr|
|----------|-------|------|------|----:|---|-----:|
|winogrande|Yaml |none |acc |0.768|± |0.0119|
```
### PubMedQA 0-Shot
```
| Tasks |Version|Filter|Metric|Value| |Stderr|
|--------|-------|------|------|----:|---|-----:|
|pubmedqa|Yaml |none |acc | 0.76|± |0.0191|
```
### RACE 1-Shot
```
|Tasks|Version|Filter|Metric|Value | |Stderr|
|-----|-------|------|------|-----:|---|-----:|
|race |Yaml |none |acc |0.5282|± |0.0154|
```
### MMLU 5-Shot (8-Bit)
```
| Groups |Version|Filter|Metric|Value | |Stderr|
|------------------|-------|------|------|-----:|---|-----:|
|mmlu |N/A |none |acc |0.6137|± |0.1243|
| - humanities |N/A |none |acc |0.5671|± |0.1101|
| - other |N/A |none |acc |0.6859|± |0.1164|
| - social_sciences|N/A |none |acc |0.7195|± |0.0713|
| - stem |N/A |none |acc |0.5087|± |0.1297|
```
### DROP 3-Shot (8-Bit) (Instruct-Eval)
```
{'score': 0.49801113762927607}
{'drop': 49.8}
drop: 49.8
```
### CRASS 0-Shot (Instruct-Eval)
```
{'score': 0.8357664233576643}
{'crass': 83.58}
crass: 83.58
```
## Training Details
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 14
- gradient_accumulation_steps: 16
- total_train_batch_size: 224
- total_eval_batch_size: 14
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.01
- num_epochs: 1
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.4795 | 0.2 | 56 | 0.4958 | -1.3684 | -2.6385 | 0.7552 | 1.2701 | -265.3887 | -241.2612 | -2.2572 | -2.4922 |
| 0.4642 | 0.4 | 112 | 0.4859 | -1.0380 | -1.9769 | 0.7273 | 0.9389 | -258.7718 | -237.9569 | -2.2414 | -2.4751 |
| 0.4758 | 0.61 | 168 | 0.4808 | -1.2594 | -2.3704 | 0.7343 | 1.1110 | -262.7074 | -240.1708 | -2.2305 | -2.4633 |
| 0.4549 | 0.81 | 224 | 0.4768 | -1.1906 | -2.3201 | 0.7552 | 1.1295 | -262.2044 | -239.4827 | -2.2284 | -2.4610 |
### Framework versions
- Transformers 4.35.0-UNA
- Pytorch 2.1.0
- Datasets 2.14.6
- Tokenizers 0.14.1
## Citations
If you find juanako useful please:
```
@misc{juanako7buna,
title={Juanako: Uniform Neural Alignment},
author={Xavier Murias},
year={2023},
publisher = {HuggingFace},
journal = {HuggingFace repository},
howpublished = {\url{https://huggingface.co/fblgit/juanako-7b-UNA}},
}
```
Thanks to all the brilliant humans behind the creation of AI, here some of the ones that we find relevant to our research. If you feel a citation is missing, please contact.
```
@misc{lin2021truthfulqa,
title={TruthfulQA: Measuring How Models Mimic Human Falsehoods},
author={Stephanie Lin and Jacob Hilton and Owain Evans},
year={2021},
eprint={2109.07958},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
@misc{tunstall2023zephyr,
title={Zephyr: Direct Distillation of LM Alignment},
author={Lewis Tunstall and Edward Beeching and Nathan Lambert and Nazneen Rajani and Kashif Rasul and Younes Belkada and Shengyi Huang and Leandro von Werra and Clémentine Fourrier and Nathan Habib and Nathan Sarrazin and Omar Sanseviero and Alexander M. Rush and Thomas Wolf},
year={2023},
eprint={2310.16944},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
@inproceedings{Bisk2020,
author = {Yonatan Bisk and Rowan Zellers and
Ronan Le Bras and Jianfeng Gao
and Yejin Choi},
title = {PIQA: Reasoning about Physical Commonsense in
Natural Language},
booktitle = {Thirty-Fourth AAAI Conference on
Artificial Intelligence},
year = {2020},
}
@software{eval-harness,
author = {Gao, Leo and
Tow, Jonathan and
Biderman, Stella and
Black, Sid and
DiPofi, Anthony and
Foster, Charles and
Golding, Laurence and
Hsu, Jeffrey and
McDonell, Kyle and
Muennighoff, Niklas and
Phang, Jason and
Reynolds, Laria and
Tang, Eric and
Thite, Anish and
Wang, Ben and
Wang, Kevin and
Zou, Andy},
title = {A framework for few-shot language model evaluation},
month = sep,
year = 2021,
publisher = {Zenodo},
version = {v0.0.1},
doi = {10.5281/zenodo.5371628},
url = {https://doi.org/10.5281/zenodo.5371628}
}
@misc{rafailov2023direct,
title={Direct Preference Optimization: Your Language Model is Secretly a Reward Model},
author={Rafael Rafailov and Archit Sharma and Eric Mitchell and Stefano Ermon and Christopher D. Manning and Chelsea Finn},
year={2023},
eprint={2305.18290},
archivePrefix={arXiv},
}
```
<!-- original-model-card end -->
| [
"PUBMEDQA",
"SCIQ"
] |
onekq-ai/OneSQL-v0.1-Qwen-32B-GGUF | onekq-ai | null | [
"transformers",
"gguf",
"qwen2",
"text-generation-inference",
"en",
"base_model:onekq-ai/OneSQL-v0.1-Qwen-32B",
"base_model:quantized:onekq-ai/OneSQL-v0.1-Qwen-32B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | "2025-03-13T04:26:27Z" | 2025-03-18T05:40:52+00:00 | 424 | 0 | ---
base_model: onekq-ai/OneSQL-v0.1-Qwen-32B
language:
- en
license: apache-2.0
tags:
- text-generation-inference
- transformers
- qwen2
- gguf
---
# Introduction
This model is the GGUF version of [OneSQL-v0.1-Qwen-32B](https://huggingface.co/onekq-ai/OneSQL-v0.1-Qwen-32B). You can also find it on [Ollama](https://ollama.com/onekq/OneSQL-v0.1-Qwen).
# Performances
The original model has an EX score of **63.33** on the [BIRD leaderboard](https://bird-bench.github.io/). Below is our self-evaluation for each quantization.
| Quantization |EX score|
|------------|------|
| Q2_K | 47.78 |
| Q3_K_S | 50.26 |
| Q3_K_M | 51.50 |
| Q3_K_L | 51.24 |
| Q4_1 | 46.54 |
| Q4_K_S | 52.47 |
| **Q4_K_M** | **53.79** |
| Q5_0 | 50.23 |
| Q5_1 | 48.36 |
| Q5_K_S | 51.93 |
| Q5_K_M | 50.66 |
| Q6_K | 52.89 |
| Q8_0 | 50.33 |
# Quick start
To use this model, craft your prompt to start with your database schema in the form of **CREATE TABLE**, followed by your natural language query preceded by **--**.
Make sure your prompt ends with **SELECT** in order for the model to finish the query for you. There is no need to set other parameters like temperature or max token limit.
```sh
PROMPT="CREATE TABLE students (
id INTEGER PRIMARY KEY,
name TEXT,
age INTEGER,
grade TEXT
);
-- Find the three youngest students
SELECT "
ollama run onekq-ai/OneSQL-v0.1-Qwen:32B-Q4_K_M "$PROMPT"
```
The model response is the finished SQL query without **SELECT**
```sql
* FROM students ORDER BY age ASC LIMIT 3
```
# Caveats
* The performance drop from the original model is due to quantization itself, and the lack of beam search support in llama.cpp framework. Use at your own discretion.
* The Q4_0 quantization suffers from repetitive output token, hence is not recommended for usage. | [
"CRAFT"
] |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.