TheBloke commited on
Commit
519401c
1 Parent(s): c22d9ed

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +56 -43
README.md CHANGED
@@ -34,39 +34,47 @@ tags:
34
  - Model creator: [Meta](https://huggingface.co/meta-llama)
35
  - Original model: [CodeLlama 34B](https://huggingface.co/codellama/CodeLlama-34b-hf)
36
 
 
37
  ## Description
38
 
39
  This repo contains GPTQ model files for [Meta's CodeLlama 34B](https://huggingface.co/codellama/CodeLlama-34b-hf).
40
 
41
  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.
42
 
 
 
43
  ## Repositories available
44
 
45
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ)
46
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/CodeLlama-34B-GGUF)
47
- * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/CodeLlama-34B-GGML)
48
  * [Meta's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/codellama/CodeLlama-34b-hf)
 
49
 
 
50
  ## Prompt template: None
51
 
52
  ```
53
  {prompt}
 
54
  ```
55
 
 
 
 
56
  ## Provided files and GPTQ parameters
57
 
58
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
59
 
60
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
61
 
62
- All GPTQ files are made with AutoGPTQ.
63
 
64
  <details>
65
  <summary>Explanation of GPTQ parameters</summary>
66
 
67
  - Bits: The bit size of the quantised model.
68
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
69
- - Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have issues with models that use Act Order plus Group Size.
70
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
71
  - 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).
72
  - 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.
@@ -83,6 +91,9 @@ All GPTQ files are made with AutoGPTQ.
83
  | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ/tree/gptq-8bit--1g-actorder_True) | 8 | None | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 4096 | 34.30 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements and to improve AutoGPTQ speed. |
84
  | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ/tree/gptq-8bit-128g-actorder_True) | 8 | 128 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 4096 | 35.07 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. Poor AutoGPTQ CUDA speed. |
85
 
 
 
 
86
  ## How to download from branches
87
 
88
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/CodeLlama-34B-GPTQ:gptq-4bit-32g-actorder_True`
@@ -91,75 +102,75 @@ All GPTQ files are made with AutoGPTQ.
91
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ
92
  ```
93
  - In Python Transformers code, the branch is the `revision` parameter; see below.
94
-
 
95
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
96
 
97
  Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
98
 
99
- It is strongly recommended to use the text-generation-webui one-click-installers unless you know how to make a manual install.
100
 
101
  1. Click the **Model tab**.
102
  2. Under **Download custom model or LoRA**, enter `TheBloke/CodeLlama-34B-GPTQ`.
103
  - To download from a specific branch, enter for example `TheBloke/CodeLlama-34B-GPTQ:gptq-4bit-32g-actorder_True`
104
  - see Provided Files above for the list of branches for each option.
105
  3. Click **Download**.
106
- 4. The model will start downloading. Once it's finished it will say "Done"
107
  5. In the top left, click the refresh icon next to **Model**.
108
  6. In the **Model** dropdown, choose the model you just downloaded: `CodeLlama-34B-GPTQ`
109
  7. The model will automatically load, and is now ready for use!
110
  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.
111
- * Note that you do not need to set GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
112
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
 
113
 
 
114
  ## How to use this GPTQ model from Python code
115
 
116
- First make sure you have [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) 0.3.1 or later installed:
117
 
118
- ```
119
- pip3 install auto-gptq
120
- ```
121
 
122
- If you have problems installing AutoGPTQ, please build from source instead:
 
 
123
  ```
 
 
 
 
124
  pip3 uninstall -y auto-gptq
125
  git clone https://github.com/PanQiWei/AutoGPTQ
126
  cd AutoGPTQ
127
  pip3 install .
128
  ```
129
 
130
- Then try the following example code:
 
 
 
 
 
 
 
 
131
 
132
  ```python
133
- from transformers import AutoTokenizer, pipeline, logging
134
- from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
135
 
136
  model_name_or_path = "TheBloke/CodeLlama-34B-GPTQ"
137
-
138
- use_triton = False
 
 
 
 
139
 
140
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
141
 
142
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
143
- use_safetensors=True,
144
- trust_remote_code=False,
145
- device="cuda:0",
146
- use_triton=use_triton,
147
- quantize_config=None)
148
-
149
- """
150
- # To download from a specific branch, use the revision parameter, as in this example:
151
- # Note that `revision` requires AutoGPTQ 0.3.1 or later!
152
-
153
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
154
- revision="gptq-4bit-32g-actorder_True",
155
- use_safetensors=True,
156
- trust_remote_code=False,
157
- device="cuda:0",
158
- quantize_config=None)
159
- """
160
-
161
  prompt = "Tell me about AI"
162
  prompt_template=f'''{prompt}
 
163
  '''
164
 
165
  print("\n\n*** Generate:")
@@ -170,9 +181,6 @@ print(tokenizer.decode(output[0]))
170
 
171
  # Inference can also be done using transformers' pipeline
172
 
173
- # Prevent printing spurious transformers error when using pipeline with AutoGPTQ
174
- logging.set_verbosity(logging.CRITICAL)
175
-
176
  print("*** Pipeline:")
177
  pipe = pipeline(
178
  "text-generation",
@@ -186,12 +194,17 @@ pipe = pipeline(
186
 
187
  print(pipe(prompt_template)[0]['generated_text'])
188
  ```
 
189
 
 
190
  ## Compatibility
191
 
192
- The files provided will work with AutoGPTQ (CUDA and Triton modes), GPTQ-for-LLaMa (only CUDA has been tested), and Occ4m's GPTQ-for-LLaMa fork.
 
 
193
 
194
- ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
 
195
 
196
  <!-- footer start -->
197
  <!-- 200823 -->
@@ -216,7 +229,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
216
 
217
  **Special thanks to**: Aemon Algiz.
218
 
219
- **Patreon special mentions**: Kacper Wikieł, knownsqashed, Leonard Tan, Asp the Wyvern, Daniel P. Andersen, Luke Pendergrass, Stanislav Ovsiannikov, RoA, Dave, Ai Maven, Kalila, Will Dee, Imad Khwaja, Nitin Borwankar, Joseph William Delisle, Tony Hughes, Cory Kujawski, Rishabh Srivastava, Russ Johnson, Stephen Murray, Lone Striker, Johann-Peter Hartmann, Elle, J, Deep Realms, SuperWojo, Raven Klaugh, Sebastain Graf, ReadyPlayerEmma, Alps Aficionado, Mano Prime, Derek Yates, Gabriel Puliatti, Mesiah Bishop, Magnesian, Sean Connelly, biorpg, Iucharbius, Olakabola, Fen Risland, Space Cruiser, theTransient, Illia Dulskyi, Thomas Belote, Spencer Kim, Pieter, John Detwiler, Fred von Graf, Michael Davis, Swaroop Kallakuri, subjectnull, Clay Pascal, Subspace Studios, Chris Smitley, Enrico Ros, usrbinkat, Steven Wood, alfie_i, David Ziegler, Willem Michiel, Matthew Berman, Andrey, Pyrater, Jeffrey Morgan, vamX, LangChain4j, Luke @flexchar, Trenton Dambrowitz, Pierre Kircher, Alex, Sam, James Bentley, Edmond Seymore, Eugene Pentland, Pedro Madruga, Rainer Wilmers, Dan Guido, Nathan LeClaire, Spiking Neurons AB, Talal Aujan, zynix, Artur Olbinski, Michael Levine, 阿明, K, John Villwock, Nikolai Manek, Femi Adebogun, senxiiz, Deo Leter, NimbleBox.ai, Viktor Bowallius, Geoffrey Montalvo, Mandus, Ajan Kanaga, ya boyyy, Jonathan Leane, webtim, Brandon Frisco, danny, Alexandros Triantafyllidis, Gabriel Tamborski, Randy H, terasurfer, Vadim, Junyu Yang, Vitor Caleffi, Chadd, transmissions 11
220
 
221
 
222
  Thank you to all my generous patrons and donaters!
@@ -309,7 +322,7 @@ All variants are available in sizes of 7B, 13B and 34B parameters.
309
 
310
  **License** A custom commercial license is available at: [https://ai.meta.com/resources/models-and-libraries/llama-downloads/](https://ai.meta.com/resources/models-and-libraries/llama-downloads/)
311
 
312
- **Research Paper** More information can be found in the paper "[Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/)".
313
 
314
  ## Intended Use
315
  **Intended Use Cases** Code Llama and its variants is intended for commercial and research use in English and relevant programming languages. The base model Code Llama can be adapted for a variety of code synthesis and understanding tasks, Code Llama - Python is designed specifically to handle the Python programming language, and Code Llama - Instruct is intended to be safer to use for code assistant and generation applications.
 
34
  - Model creator: [Meta](https://huggingface.co/meta-llama)
35
  - Original model: [CodeLlama 34B](https://huggingface.co/codellama/CodeLlama-34b-hf)
36
 
37
+ <!-- description start -->
38
  ## Description
39
 
40
  This repo contains GPTQ model files for [Meta's CodeLlama 34B](https://huggingface.co/codellama/CodeLlama-34b-hf).
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
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ)
49
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/CodeLlama-34B-GGUF)
 
50
  * [Meta's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/codellama/CodeLlama-34b-hf)
51
+ <!-- repositories-available end -->
52
 
53
+ <!-- prompt-template start -->
54
  ## Prompt template: None
55
 
56
  ```
57
  {prompt}
58
+
59
  ```
60
 
61
+ <!-- prompt-template end -->
62
+
63
+ <!-- README_GPTQ.md-provided-files start -->
64
  ## Provided files and GPTQ parameters
65
 
66
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
67
 
68
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
69
 
70
+ 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.
71
 
72
  <details>
73
  <summary>Explanation of GPTQ parameters</summary>
74
 
75
  - Bits: The bit size of the quantised model.
76
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
77
+ - 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.
78
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
79
  - 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).
80
  - 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.
 
91
  | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ/tree/gptq-8bit--1g-actorder_True) | 8 | None | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 4096 | 34.30 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements and to improve AutoGPTQ speed. |
92
  | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ/tree/gptq-8bit-128g-actorder_True) | 8 | 128 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 4096 | 35.07 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. Poor AutoGPTQ CUDA speed. |
93
 
94
+ <!-- README_GPTQ.md-provided-files end -->
95
+
96
+ <!-- README_GPTQ.md-download-from-branches start -->
97
  ## How to download from branches
98
 
99
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/CodeLlama-34B-GPTQ:gptq-4bit-32g-actorder_True`
 
102
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/CodeLlama-34B-GPTQ
103
  ```
104
  - In Python Transformers code, the branch is the `revision` parameter; see below.
105
+ <!-- README_GPTQ.md-download-from-branches end -->
106
+ <!-- README_GPTQ.md-text-generation-webui start -->
107
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
108
 
109
  Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
110
 
111
+ 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.
112
 
113
  1. Click the **Model tab**.
114
  2. Under **Download custom model or LoRA**, enter `TheBloke/CodeLlama-34B-GPTQ`.
115
  - To download from a specific branch, enter for example `TheBloke/CodeLlama-34B-GPTQ:gptq-4bit-32g-actorder_True`
116
  - see Provided Files above for the list of branches for each option.
117
  3. Click **Download**.
118
+ 4. The model will start downloading. Once it's finished it will say "Done".
119
  5. In the top left, click the refresh icon next to **Model**.
120
  6. In the **Model** dropdown, choose the model you just downloaded: `CodeLlama-34B-GPTQ`
121
  7. The model will automatically load, and is now ready for use!
122
  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.
123
+ * 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`.
124
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
125
+ <!-- README_GPTQ.md-text-generation-webui end -->
126
 
127
+ <!-- README_GPTQ.md-use-from-python start -->
128
  ## How to use this GPTQ model from Python code
129
 
130
+ ### Install the necessary packages
131
 
132
+ Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
 
 
133
 
134
+ ```shell
135
+ pip3 install transformers>=4.32.0 optimum>=1.12.0
136
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
137
  ```
138
+
139
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
140
+
141
+ ```shell
142
  pip3 uninstall -y auto-gptq
143
  git clone https://github.com/PanQiWei/AutoGPTQ
144
  cd AutoGPTQ
145
  pip3 install .
146
  ```
147
 
148
+ ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
149
+
150
+ If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
151
+ ```shell
152
+ pip3 uninstall -y transformers
153
+ pip3 install git+https://github.com/huggingface/transformers.git
154
+ ```
155
+
156
+ ### You can then use the following code
157
 
158
  ```python
159
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
160
 
161
  model_name_or_path = "TheBloke/CodeLlama-34B-GPTQ"
162
+ # To use a different branch, change revision
163
+ # For example: revision="gptq-4bit-32g-actorder_True"
164
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
165
+ torch_dtype=torch.float16,
166
+ device_map="auto",
167
+ revision="main")
168
 
169
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
170
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
171
  prompt = "Tell me about AI"
172
  prompt_template=f'''{prompt}
173
+
174
  '''
175
 
176
  print("\n\n*** Generate:")
 
181
 
182
  # Inference can also be done using transformers' pipeline
183
 
 
 
 
184
  print("*** Pipeline:")
185
  pipe = pipeline(
186
  "text-generation",
 
194
 
195
  print(pipe(prompt_template)[0]['generated_text'])
196
  ```
197
+ <!-- README_GPTQ.md-use-from-python end -->
198
 
199
+ <!-- README_GPTQ.md-compatibility start -->
200
  ## Compatibility
201
 
202
+ 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).
203
+
204
+ [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.
205
 
206
+ [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) is compatible with all GPTQ models.
207
+ <!-- README_GPTQ.md-compatibility end -->
208
 
209
  <!-- footer start -->
210
  <!-- 200823 -->
 
229
 
230
  **Special thanks to**: Aemon Algiz.
231
 
232
+ **Patreon special mentions**: Russ Johnson, J, alfie_i, Alex, NimbleBox.ai, Chadd, Mandus, Nikolai Manek, Ken Nordquist, ya boyyy, Illia Dulskyi, Viktor Bowallius, vamX, Iucharbius, zynix, Magnesian, Clay Pascal, Pierre Kircher, Enrico Ros, Tony Hughes, Elle, Andrey, knownsqashed, Deep Realms, Jerry Meng, Lone Striker, Derek Yates, Pyrater, Mesiah Bishop, James Bentley, Femi Adebogun, Brandon Frisco, SuperWojo, Alps Aficionado, Michael Dempsey, Vitor Caleffi, Will Dee, Edmond Seymore, usrbinkat, LangChain4j, Kacper Wikieł, Luke Pendergrass, John Detwiler, theTransient, Nathan LeClaire, Tiffany J. Kim, biorpg, Eugene Pentland, Stanislav Ovsiannikov, Fred von Graf, terasurfer, Kalila, Dan Guido, Nitin Borwankar, 阿明, Ai Maven, John Villwock, Gabriel Puliatti, Stephen Murray, Asp the Wyvern, danny, Chris Smitley, ReadyPlayerEmma, S_X, Daniel P. Andersen, Olakabola, Jeffrey Morgan, Imad Khwaja, Caitlyn Gatomon, webtim, Alicia Loh, Trenton Dambrowitz, Swaroop Kallakuri, Erik Bjäreholt, Leonard Tan, Spiking Neurons AB, Luke @flexchar, Ajan Kanaga, Thomas Belote, Deo Leter, RoA, Willem Michiel, transmissions 11, subjectnull, Matthew Berman, Joseph William Delisle, David Ziegler, Michael Davis, Johann-Peter Hartmann, Talal Aujan, senxiiz, Artur Olbinski, Rainer Wilmers, Spencer Kim, Fen Risland, Cap'n Zoog, Rishabh Srivastava, Michael Levine, Geoffrey Montalvo, Sean Connelly, Alexandros Triantafyllidis, Pieter, Gabriel Tamborski, Sam, Subspace Studios, Junyu Yang, Pedro Madruga, Vadim, Cory Kujawski, K, Raven Klaugh, Randy H, Mano Prime, Sebastain Graf, Space Cruiser
233
 
234
 
235
  Thank you to all my generous patrons and donaters!
 
322
 
323
  **License** A custom commercial license is available at: [https://ai.meta.com/resources/models-and-libraries/llama-downloads/](https://ai.meta.com/resources/models-and-libraries/llama-downloads/)
324
 
325
+ **Research Paper** More information can be found in the paper "[Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/)" or its [arXiv page](https://arxiv.org/abs/2308.12950).
326
 
327
  ## Intended Use
328
  **Intended Use Cases** Code Llama and its variants is intended for commercial and research use in English and relevant programming languages. The base model Code Llama can be adapted for a variety of code synthesis and understanding tasks, Code Llama - Python is designed specifically to handle the Python programming language, and Code Llama - Instruct is intended to be safer to use for code assistant and generation applications.