TheBloke commited on
Commit
45e05f3
1 Parent(s): 11c8182

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +250 -44
README.md CHANGED
@@ -1,6 +1,16 @@
1
  ---
 
2
  inference: false
3
  license: other
 
 
 
 
 
 
 
 
 
4
  ---
5
 
6
  <!-- header start -->
@@ -20,59 +30,187 @@ license: other
20
  <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
21
  <!-- header end -->
22
 
23
- # WizardLM 13B 1.0 GPTQ
 
 
24
 
25
- These files are GPTQ 4bit model files for [WizardLM 13B 1.0](https://huggingface.co/victor123/WizardLM-13B-1.0).
 
26
 
27
- It is the result of merging the LoRA then quantising to 4bit using [GPTQ-for-LLaMa](https://github.com/qwopqwop200/GPTQ-for-LLaMa).
28
 
29
- ## Other repositories available
30
 
31
- * [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/WizardLM-13B-1.0-GPTQ)
32
- * [4-bit, 5-bit and 8-bit GGML models for CPU(+GPU) inference](https://huggingface.co/TheBloke/WizardLM-13B-1.0-GGML)
33
- * [Merged, unquantised fp16 model in HF format](https://huggingface.co/TheBloke/wizardLM-13B-1.0-fp16)
34
 
35
- ## Prompt Template
 
 
 
 
 
 
 
36
 
37
  ```
38
- A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
39
- USER: prompt goes here
40
- ASSISTANT:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
  ```
 
 
 
 
42
 
43
- ## How to easily download and use this model in text-generation-webui
44
 
45
- Open the text-generation-webui UI as normal.
46
 
47
  1. Click the **Model tab**.
48
  2. Under **Download custom model or LoRA**, enter `TheBloke/WizardLM-13B-1.0-GPTQ`.
 
 
49
  3. Click **Download**.
50
- 4. Wait until it says it's finished downloading.
51
- 5. Click the **Refresh** icon next to **Model** in the top left.
52
- 6. In the **Model drop-down**: choose the model you just downloaded, `WizardLM-13B-1.0-GPTQ`.
53
- 7. If you see an error in the bottom right, ignore it - it's temporary.
54
- 8. Fill out the `GPTQ parameters` on the right: `Bits = 4`, `Groupsize = 128`, `model_type = Llama`
55
- 9. Click **Save settings for this model** in the top right.
56
- 10. Click **Reload the Model** in the top right.
57
- 11. Once it says it's loaded, click the **Text Generation tab** and enter a prompt!
58
 
59
- ## Provided files
 
60
 
61
- **WizardLM-13B-1.0-GPTQ-4bit-128g.no-act-order.safetensors**
62
 
63
- This will work with all versions of GPTQ-for-LLaMa. It has maximum compatibility.
64
 
65
- It was created with groupsize 128 to ensure higher quality inference, without `--act-order` parameter to maximise compatibility.
 
 
 
66
 
67
- * `WizardLM-13B-1.0-GPTQ-4bit-128g.no-act-order.safetensors`
68
- * Works with all versions of GPTQ-for-LLaMa code, both Triton and CUDA branches
69
- * Works with AutoGPTQ
70
- * Works with text-generation-webui one-click-installers
71
- * Parameters: Groupsize = 128. No act-order.
72
- * Command used to create the GPTQ:
73
- ```
74
- python llama.py /workspace/process/wizardLM-13B-1.0/HF wikitext2 --wbits 4 --true-sequential --groupsize 128 --save_safetensors /workspace/process/wizardLM-13B-1.0/gptq/WizardLM-13B-1.0-GPTQ-4bit-128g.no-act-order.safetensors
75
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
76
 
77
  <!-- footer start -->
78
  <!-- 200823 -->
@@ -82,10 +220,12 @@ For further support, and discussions on these models and AI in general, join us
82
 
83
  [TheBloke AI's Discord server](https://discord.gg/theblokeai)
84
 
85
- ## Thanks, and how to contribute.
86
 
87
  Thanks to the [chirper.ai](https://chirper.ai) team!
88
 
 
 
89
  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.
90
 
91
  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.
@@ -97,7 +237,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
97
 
98
  **Special thanks to**: Aemon Algiz.
99
 
100
- **Patreon special mentions**: Sam, theTransient, Jonathan Leane, Steven Wood, webtim, Johann-Peter Hartmann, Geoffrey Montalvo, Gabriel Tamborski, Willem Michiel, John Villwock, Derek Yates, Mesiah Bishop, Eugene Pentland, Pieter, Chadd, Stephen Murray, Daniel P. Andersen, terasurfer, Brandon Frisco, Thomas Belote, Sid, Nathan LeClaire, Magnesian, Alps Aficionado, Stanislav Ovsiannikov, Alex, Joseph William Delisle, Nikolai Manek, Michael Davis, Junyu Yang, K, J, Spencer Kim, Stefan Sabev, Olusegun Samson, transmissions 11, Michael Levine, Cory Kujawski, Rainer Wilmers, zynix, Kalila, Luke @flexchar, Ajan Kanaga, Mandus, vamX, Ai Maven, Mano Prime, Matthew Berman, subjectnull, Vitor Caleffi, Clay Pascal, biorpg, alfie_i, 阿明, Jeffrey Morgan, ya boyyy, Raymond Fosdick, knownsqashed, Olakabola, Leonard Tan, ReadyPlayerEmma, Enrico Ros, Dave, Talal Aujan, Illia Dulskyi, Sean Connelly, senxiiz, Artur Olbinski, Elle, Raven Klaugh, Fen Risland, Deep Realms, Imad Khwaja, Fred von Graf, Will Dee, usrbinkat, SuperWojo, Alexandros Triantafyllidis, Swaroop Kallakuri, Dan Guido, John Detwiler, Pedro Madruga, Iucharbius, Viktor Bowallius, Asp the Wyvern, Edmond Seymore, Trenton Dambrowitz, Space Cruiser, Spiking Neurons AB, Pyrater, LangChain4j, Tony Hughes, Kacper Wikieł, Rishabh Srivastava, David Ziegler, Luke Pendergrass, Andrey, Gabriel Puliatti, Lone Striker, Sebastain Graf, Pierre Kircher, Randy H, NimbleBox.ai, Vadim, danny, Deo Leter
101
 
102
 
103
  Thank you to all my generous patrons and donaters!
@@ -106,13 +246,79 @@ And thank you again to a16z for their generous grant.
106
 
107
  <!-- footer end -->
108
 
109
- # Original model card: WizardLM 13B 1.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
110
 
111
  ## WizardLM: An Instruction-following LLM Using Evol-Instruct
112
  Empowering Large Pre-Trained Language Models to Follow Complex Instructions
113
 
114
  <p align="center" width="100%">
115
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/WizardLM.png" alt="WizardLM" style="width: 20%; min-width: 300px; display: block; margin: auto;"></a>
116
  </p>
117
 
118
  [![Code License](https://img.shields.io/badge/Code%20License-Apache_2.0-green.svg)](https://github.com/tatsu-lab/stanford_alpaca/blob/main/LICENSE)
@@ -137,7 +343,7 @@ At present, our core contributors are preparing the **33B** version and we expec
137
 
138
  We adopt the automatic evaluation framework based on GPT-4 proposed by FastChat to assess the performance of chatbot models. As shown in the following figure, WizardLM-13B achieved better results than Vicuna-13b.
139
  <p align="center" width="100%">
140
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/WizarLM13b-GPT4.png" alt="WizardLM" style="width: 100%; min-width: 300px; display: block; margin: auto;"></a>
141
  </p>
142
 
143
  ### WizardLM-13B performance on different skills.
@@ -145,7 +351,7 @@ We adopt the automatic evaluation framework based on GPT-4 proposed by FastChat
145
  The following figure compares WizardLM-13B and ChatGPT’s skill on Evol-Instruct testset. The result indicates that WizardLM-13B achieves 89.1% of ChatGPT’s performance on average, with almost 100% (or more than) capacity on 10 skills, and more than 90% capacity on 22 skills.
146
 
147
  <p align="center" width="100%">
148
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/evol-testset_skills-13b.png" alt="WizardLM" style="width: 100%; min-width: 300px; display: block; margin: auto;"></a>
149
  </p>
150
 
151
  ## Call for Feedbacks
@@ -164,11 +370,11 @@ We just sample some cases to demonstrate the performance of WizardLM and ChatGPT
164
  [Evol-Instruct](https://github.com/nlpxucan/evol-instruct) is a novel method using LLMs instead of humans to automatically mass-produce open-domain instructions of various difficulty levels and skills range, to improve the performance of LLMs.
165
 
166
  <p align="center" width="100%">
167
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/git_overall.png" alt="WizardLM" style="width: 86%; min-width: 300px; display: block; margin: auto;"></a>
168
  </p>
169
 
170
  <p align="center" width="100%">
171
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/git_running.png" alt="WizardLM" style="width: 86%; min-width: 300px; display: block; margin: auto;"></a>
172
  </p>
173
 
174
  ## Contents
@@ -283,12 +489,12 @@ To evaluate Wizard, we conduct human evaluation on the inputs from our human ins
283
 
284
  WizardLM achieved significantly better results than Alpaca and Vicuna-7b.
285
  <p align="center" width="60%">
286
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/win.png" alt="WizardLM" style="width: 60%; min-width: 300px; display: block; margin: auto;"></a>
287
  </p>
288
 
289
  In the high-difficulty section of our test set (difficulty level >= 8), WizardLM even outperforms ChatGPT, with a win rate 7.9% larger than Chatgpt (42.9% vs. 35.0%). This indicates that our method can significantly improve the ability of large language models to handle complex instructions.
290
  <p align="center" width="60%">
291
- <a ><img src="https://raw.githubusercontent.com/nlpxucan/WizardLM/main/imgs/windiff.png" alt="WizardLM" style="width: 60%; min-width: 300px; display: block; margin: auto;"></a>
292
  </p>
293
 
294
  ### Citation
 
1
  ---
2
+ base_model: https://huggingface.co/WizardLM/WizardLM-13B-V1.0
3
  inference: false
4
  license: other
5
+ model_creator: WizardLM
6
+ model_name: WizardLM 13B 1.0
7
+ model_type: llama
8
+ prompt_template: 'A chat between a curious user and an artificial intelligence assistant.
9
+ The assistant gives helpful, detailed, and polite answers to the user''s questions.
10
+ USER: {prompt} ASSISTANT:
11
+
12
+ '
13
+ quantized_by: TheBloke
14
  ---
15
 
16
  <!-- header start -->
 
30
  <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
31
  <!-- header end -->
32
 
33
+ # WizardLM 13B 1.0 - GPTQ
34
+ - Model creator: [WizardLM](https://huggingface.co/WizardLM)
35
+ - Original model: [WizardLM 13B 1.0](https://huggingface.co/WizardLM/WizardLM-13B-V1.0)
36
 
37
+ <!-- description start -->
38
+ ## Description
39
 
40
+ This repo contains GPTQ model files for [WizardLM's WizardLM 13B 1.0](https://huggingface.co/WizardLM/WizardLM-13B-V1.0).
41
 
42
+ Multiple GPTQ parameter permutations are provided; see Provided Files below for details of the options provided, their parameters, and the software used to create them.
43
 
44
+ <!-- description end -->
45
+ <!-- repositories-available start -->
46
+ ## Repositories available
47
 
48
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/WizardLM-13B-1.0-AWQ)
49
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/WizardLM-13B-1.0-GPTQ)
50
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/WizardLM-13B-1.0-GGUF)
51
+ * [WizardLM's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/TheBloke/wizardLM-13B-1.0-fp16)
52
+ <!-- repositories-available end -->
53
+
54
+ <!-- prompt-template start -->
55
+ ## Prompt template: Vicuna
56
 
57
  ```
58
+ A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: {prompt} ASSISTANT:
59
+
60
+ ```
61
+
62
+ <!-- prompt-template end -->
63
+ <!-- licensing start -->
64
+ ## Licensing
65
+
66
+ The creator of the source model has listed its license as `other`, and this quantization has therefore used that same license.
67
+
68
+ 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.
69
+
70
+ In the meantime, any questions regarding licensing, and in particular how these two licenses might interact, should be directed to the original model repository: [WizardLM's WizardLM 13B 1.0](https://huggingface.co/WizardLM/WizardLM-13B-V1.0).
71
+ <!-- licensing end -->
72
+ <!-- README_GPTQ.md-provided-files start -->
73
+ ## Provided files and GPTQ parameters
74
+
75
+ Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
76
+
77
+ Each separate quant is in a different branch. See below for instructions on fetching from different branches.
78
+
79
+ All recent GPTQ files are made with AutoGPTQ, and all files in non-main branches are made with AutoGPTQ. Files in the `main` branch which were uploaded before August 2023 were made with GPTQ-for-LLaMa.
80
+
81
+ <details>
82
+ <summary>Explanation of GPTQ parameters</summary>
83
+
84
+ - Bits: The bit size of the quantised model.
85
+ - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
86
+ - Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now.
87
+ - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
88
+ - GPTQ dataset: The dataset used for quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s).
89
+ - Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences.
90
+ - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama models in 4-bit.
91
+
92
+ </details>
93
+
94
+ | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
95
+ | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
96
+ | main | 4 | 128 | No | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 8.11 GB | Yes | 4-bit, without Act Order and group size 128g. |
97
+
98
+ <!-- README_GPTQ.md-provided-files end -->
99
+
100
+ <!-- README_GPTQ.md-download-from-branches start -->
101
+ ## How to download from branches
102
+
103
+ - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/WizardLM-13B-1.0-GPTQ:main`
104
+ - With Git, you can clone a branch with:
105
+ ```
106
+ git clone --single-branch --branch main https://huggingface.co/TheBloke/WizardLM-13B-1.0-GPTQ
107
  ```
108
+ - In Python Transformers code, the branch is the `revision` parameter; see below.
109
+ <!-- README_GPTQ.md-download-from-branches end -->
110
+ <!-- README_GPTQ.md-text-generation-webui start -->
111
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
112
 
113
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
114
 
115
+ 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.
116
 
117
  1. Click the **Model tab**.
118
  2. Under **Download custom model or LoRA**, enter `TheBloke/WizardLM-13B-1.0-GPTQ`.
119
+ - To download from a specific branch, enter for example `TheBloke/WizardLM-13B-1.0-GPTQ:main`
120
+ - see Provided Files above for the list of branches for each option.
121
  3. Click **Download**.
122
+ 4. The model will start downloading. Once it's finished it will say "Done".
123
+ 5. In the top left, click the refresh icon next to **Model**.
124
+ 6. In the **Model** dropdown, choose the model you just downloaded: `WizardLM-13B-1.0-GPTQ`
125
+ 7. The model will automatically load, and is now ready for use!
126
+ 8. 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.
127
+ * Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
128
+ 9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
129
+ <!-- README_GPTQ.md-text-generation-webui end -->
130
 
131
+ <!-- README_GPTQ.md-use-from-python start -->
132
+ ## How to use this GPTQ model from Python code
133
 
134
+ ### Install the necessary packages
135
 
136
+ Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
137
 
138
+ ```shell
139
+ pip3 install transformers>=4.32.0 optimum>=1.12.0
140
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
141
+ ```
142
 
143
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
144
+
145
+ ```shell
146
+ pip3 uninstall -y auto-gptq
147
+ git clone https://github.com/PanQiWei/AutoGPTQ
148
+ cd AutoGPTQ
149
+ pip3 install .
150
+ ```
151
+
152
+ ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
153
+
154
+ If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
155
+ ```shell
156
+ pip3 uninstall -y transformers
157
+ pip3 install git+https://github.com/huggingface/transformers.git
158
+ ```
159
+
160
+ ### You can then use the following code
161
+
162
+ ```python
163
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
164
+
165
+ model_name_or_path = "TheBloke/WizardLM-13B-1.0-GPTQ"
166
+ # To use a different branch, change revision
167
+ # For example: revision="main"
168
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
169
+ device_map="auto",
170
+ trust_remote_code=False,
171
+ revision="main")
172
+
173
+ tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
174
+
175
+ prompt = "Tell me about AI"
176
+ prompt_template=f'''A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: {prompt} ASSISTANT:
177
+
178
+ '''
179
+
180
+ print("\n\n*** Generate:")
181
+
182
+ input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
183
+ output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512)
184
+ print(tokenizer.decode(output[0]))
185
+
186
+ # Inference can also be done using transformers' pipeline
187
+
188
+ print("*** Pipeline:")
189
+ pipe = pipeline(
190
+ "text-generation",
191
+ model=model,
192
+ tokenizer=tokenizer,
193
+ max_new_tokens=512,
194
+ do_sample=True,
195
+ temperature=0.7,
196
+ top_p=0.95,
197
+ top_k=40,
198
+ repetition_penalty=1.1
199
+ )
200
+
201
+ print(pipe(prompt_template)[0]['generated_text'])
202
+ ```
203
+ <!-- README_GPTQ.md-use-from-python end -->
204
+
205
+ <!-- README_GPTQ.md-compatibility start -->
206
+ ## Compatibility
207
+
208
+ The files provided are tested to work with AutoGPTQ, both via Transformers and using AutoGPTQ directly. They should also work with [Occ4m's GPTQ-for-LLaMa fork](https://github.com/0cc4m/KoboldAI).
209
+
210
+ [ExLlama](https://github.com/turboderp/exllama) is compatible with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
211
+
212
+ [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) is compatible with all GPTQ models.
213
+ <!-- README_GPTQ.md-compatibility end -->
214
 
215
  <!-- footer start -->
216
  <!-- 200823 -->
 
220
 
221
  [TheBloke AI's Discord server](https://discord.gg/theblokeai)
222
 
223
+ ## Thanks, and how to contribute
224
 
225
  Thanks to the [chirper.ai](https://chirper.ai) team!
226
 
227
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
228
+
229
  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.
230
 
231
  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.
 
237
 
238
  **Special thanks to**: Aemon Algiz.
239
 
240
+ **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
241
 
242
 
243
  Thank you to all my generous patrons and donaters!
 
246
 
247
  <!-- footer end -->
248
 
249
+ # Original model card: WizardLM's WizardLM 13B 1.0
250
+
251
+
252
+ <!-- header start -->
253
+ <div style="width: 100%;">
254
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
255
+ </div>
256
+ <div style="display: flex; justify-content: space-between; width: 100%;">
257
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
258
+ <p><a href="https://discord.gg/Jq4vkcDakD">Chat & support: my new Discord server</a></p>
259
+ </div>
260
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
261
+ <p><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
262
+ </div>
263
+ </div>
264
+ <!-- header end -->
265
+
266
+ # WizardLM 13B 1.0 fp16
267
+
268
+ These files are fp16 unquantised format model files for [WizardLM 13B 1.0](https://huggingface.co/victor123/WizardLM-13B-1.0).
269
+
270
+ It is the result of merging the deltas provided in the above repo.
271
+
272
+ ## Need support? Want to discuss? I now have a Discord!
273
+
274
+ Join me at: https://discord.gg/UBgz4VXf
275
+
276
+ ## Other repositories available
277
+
278
+ * [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/WizardLM-13B-1.0-GPTQ)
279
+ * [4-bit, 5-bit and 8-bit GGML models for CPU(+GPU) inference](https://huggingface.co/TheBloke/WizardLM-13B-1.0-GGML)
280
+ * [Merged, unquantised fp16 model in HF format](https://huggingface.co/TheBloke/WizardLM-13B-1.0-HF)
281
+
282
+ ## Prompt Template
283
+
284
+ ```
285
+ A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
286
+ USER: prompt goes here
287
+ ASSISTANT:
288
+ ```
289
+
290
+ <!-- footer start -->
291
+ ## Discord
292
+
293
+ For further support, and discussions on these models and AI in general, join us at:
294
+
295
+ [TheBloke AI's Discord server](https://discord.gg/Jq4vkcDakD)
296
+
297
+ ## Thanks, and how to contribute.
298
+
299
+ Thanks to the [chirper.ai](https://chirper.ai) team!
300
+
301
+ 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.
302
+
303
+ 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.
304
+
305
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
306
+
307
+ * Patreon: https://patreon.com/TheBlokeAI
308
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
309
+
310
+ **Patreon special mentions**: Aemon Algiz, Dmitriy Samsonov, Nathan LeClaire, Trenton Dambrowitz, Mano Prime, David Flickinger, vamX, Nikolai Manek, senxiiz, Khalefa Al-Ahmad, Illia Dulskyi, Jonathan Leane, Talal Aujan, V. Lukas, Joseph William Delisle, Pyrater, Oscar Rangel, Lone Striker, Luke Pendergrass, Eugene Pentland, Sebastain Graf, Johann-Peter Hartman.
311
+
312
+ Thank you to all my generous patrons and donaters!
313
+ <!-- footer end -->
314
+
315
+ # Original model card
316
 
317
  ## WizardLM: An Instruction-following LLM Using Evol-Instruct
318
  Empowering Large Pre-Trained Language Models to Follow Complex Instructions
319
 
320
  <p align="center" width="100%">
321
+ <a ><img src="imgs/WizardLM.png" alt="WizardLM" style="width: 20%; min-width: 300px; display: block; margin: auto;"></a>
322
  </p>
323
 
324
  [![Code License](https://img.shields.io/badge/Code%20License-Apache_2.0-green.svg)](https://github.com/tatsu-lab/stanford_alpaca/blob/main/LICENSE)
 
343
 
344
  We adopt the automatic evaluation framework based on GPT-4 proposed by FastChat to assess the performance of chatbot models. As shown in the following figure, WizardLM-13B achieved better results than Vicuna-13b.
345
  <p align="center" width="100%">
346
+ <a ><img src="imgs/WizarLM13b-GPT4.png" alt="WizardLM" style="width: 100%; min-width: 300px; display: block; margin: auto;"></a>
347
  </p>
348
 
349
  ### WizardLM-13B performance on different skills.
 
351
  The following figure compares WizardLM-13B and ChatGPT’s skill on Evol-Instruct testset. The result indicates that WizardLM-13B achieves 89.1% of ChatGPT’s performance on average, with almost 100% (or more than) capacity on 10 skills, and more than 90% capacity on 22 skills.
352
 
353
  <p align="center" width="100%">
354
+ <a ><img src="imgs/evol-testset_skills-13b.png" alt="WizardLM" style="width: 100%; min-width: 300px; display: block; margin: auto;"></a>
355
  </p>
356
 
357
  ## Call for Feedbacks
 
370
  [Evol-Instruct](https://github.com/nlpxucan/evol-instruct) is a novel method using LLMs instead of humans to automatically mass-produce open-domain instructions of various difficulty levels and skills range, to improve the performance of LLMs.
371
 
372
  <p align="center" width="100%">
373
+ <a ><img src="imgs/git_overall.png" alt="WizardLM" style="width: 86%; min-width: 300px; display: block; margin: auto;"></a>
374
  </p>
375
 
376
  <p align="center" width="100%">
377
+ <a ><img src="imgs/git_running.png" alt="WizardLM" style="width: 86%; min-width: 300px; display: block; margin: auto;"></a>
378
  </p>
379
 
380
  ## Contents
 
489
 
490
  WizardLM achieved significantly better results than Alpaca and Vicuna-7b.
491
  <p align="center" width="60%">
492
+ <a ><img src="imgs/win.png" alt="WizardLM" style="width: 60%; min-width: 300px; display: block; margin: auto;"></a>
493
  </p>
494
 
495
  In the high-difficulty section of our test set (difficulty level >= 8), WizardLM even outperforms ChatGPT, with a win rate 7.9% larger than Chatgpt (42.9% vs. 35.0%). This indicates that our method can significantly improve the ability of large language models to handle complex instructions.
496
  <p align="center" width="60%">
497
+ <a ><img src="imgs/windiff.png" alt="WizardLM" style="width: 60%; min-width: 300px; display: block; margin: auto;"></a>
498
  </p>
499
 
500
  ### Citation