README.md

---
license: apache-2.0
tags:
- gemma
- gguf
- quantized
inference: false
---

GGUF-IQ-Imatrix quants for [YeungNLP/firefly-gemma-7b](https://huggingface.co/YeungNLP/firefly-gemma-7b):

```python
    quantization_options = [
        "IQ2_XXS", "IQ2_XS", "IQ2_S", "IQ2_M", "Q3_K_M",
        "Q4_K_M", "Q4_K_S", "IQ4_XS", "Q5_K_M", "Q5_K_S",
        "Q6_K", "Q8_0", "IQ3_M", "IQ3_S", "IQ3_XXS"
    ]
```

[Requested by Cran-May.](https://huggingface.co/Lewdiculous/Model-Requests/discussions/8)

**Model card image:**

![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/65d4cf2693a0a3744a27536c/SrOekTxdpnxHyWWmMiAvc.jpeg)

## Model Card for Firefly-Gemma:

[firefly-gemma-7b](https://huggingface.co/YeungNLP/firefly-gemma-7b) is trained based on [gemma-7b](https://huggingface.co/google/gemma-7b) to act as a helpful and harmless AI assistant. 
We use [Firefly](https://github.com/yangjianxin1/Firefly) to train the model on **a single V100 GPU** with QLoRA.

Our model outperforms the official [gemma-7b-it](https://huggingface.co/google/gemma-7b-it), [zephyr-7b-gemma-v0.1](https://huggingface.co/HuggingFaceH4/zephyr-7b-gemma-v0.1), [Qwen1.5-7B-Chat](https://huggingface.co/Qwen/Qwen1.5-7B-Chat) and [Zephyr-7B-Beta](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta) on [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).

![image/png](https://cdn-uploads.huggingface.co/production/uploads/65d4cf2693a0a3744a27536c/xeZeemMWs8_NLL-BnjGrN.png)

We advise you to install transformers>=4.38.1.

## Performance
We evaluate our models on [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard), they achieve good performance.

| Model                          | Average | ARC    | HellaSwag | MMLU   | TruthfulQA | Winogrande | GSM8K  |
|--------------------------------|--------|--------|-----------|--------|------------|-----------|--------|
| **firefly-gemma-7b**           | 62.93  | 	62.12 | 79.77     | 61.57  | 49.41      | 75.45     | 49.28  |
| zephyr-7b-gemma-v0.1           |62.41|58.45|83.48|60.68|52.07|	74.19| 45.56|
| firefly-qwen1.5-en-7b-dpo-v0.1 | 62.36  | 54.35  | 76.04     | 61.21  | 56.4       | 72.06     | 54.13  |
| zephyr-7b-beta                 | 61.95  | 62.03  | 84.36     | 61.07  | 	57.45     | 77.74     | 	29.04 |
| firefly-qwen1.5-en-7b          | 61.44  | 53.41  | 	75.51          | 61.67       |51.96          |70.72        | 55.34       |
| vicuna-13b-v1.5                | 55.41  | 57.08  | 	81.24    | 56.67  | 51.51      | 	74.66    | 11.3   |
| Xwin-LM-13B-V0.1               | 55.29  | 	62.54 | 82.8      | 56.53  | 45.96      | 74.27     | 9.63   |
| Qwen1.5-7B-Chat                | 55.15  | 	55.89 | 78.56     | 61.65  | 53.54      | 	67.72    | 13.57  |
| gemma-7b-it                    | 53.56  | 51.45  | 71.96     | 53.52  | 47.29      | 	67.96    | 	29.19 |



## Usage
The chat template of our chat models is similar as Official gemma-7b-it:
```text
<bos><start_of_turn>user
hello, who are you?<end_of_turn>
<start_of_turn>model
I am a AI program developed by Firefly<eos>
```

You can use script to inference in [Firefly](https://github.com/yangjianxin1/Firefly/blob/master/script/chat/chat.py).

You can also use the following code:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name_or_path = "YeungNLP/firefly-gemma-7b"
model = AutoModelForCausalLM.from_pretrained(
    model_name_or_path,
    trust_remote_code=True,
    low_cpu_mem_usage=True,
    torch_dtype=torch.float16,
    device_map='auto',
)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)

prompt = "Compose an engaging travel blog post about a recent trip to Hawaii, highlighting cultural experiences and must-see attractions. "
text = f"""
<bos><start_of_turn>user
{prompt}<end_of_turn>
<start_of_turn>model
""".strip()
model_inputs = tokenizer([text], return_tensors="pt").to('cuda')

generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=1500,
    top_p = 0.9,
    temperature = 0.35,
    repetition_penalty = 1.0,
    eos_token_id=tokenizer.encode('<eos>', add_special_tokens=False)
)
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)
```