TheBloke commited on
Commit
0bdd11f
1 Parent(s): 3af3644

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +55 -41
README.md CHANGED
@@ -36,19 +36,24 @@ tags:
36
  - Model creator: [NousResearch](https://huggingface.co/NousResearch)
37
  - Original model: [Nous Hermes Llama2 70B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b)
38
 
 
39
  ## Description
40
 
41
  This repo contains GPTQ model files for [NousResearch's Nous Hermes Llama2 70B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b).
42
 
43
  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.
44
 
 
 
45
  ## Repositories available
46
 
47
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ)
48
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GGUF)
49
  * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GGML)
50
  * [NousResearch's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b)
 
51
 
 
52
  ## Prompt template: Alpaca-InstructOnly
53
 
54
  ```
@@ -57,22 +62,26 @@ Multiple GPTQ parameter permutations are provided; see Provided Files below for
57
  {prompt}
58
 
59
  ### Response:
 
60
  ```
61
 
 
 
 
62
  ## Provided files and GPTQ parameters
63
 
64
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
65
 
66
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
67
 
68
- All GPTQ files are made with AutoGPTQ.
69
 
70
  <details>
71
  <summary>Explanation of GPTQ parameters</summary>
72
 
73
  - Bits: The bit size of the quantised model.
74
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
75
- - 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.
76
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
77
  - 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).
78
  - 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.
@@ -89,6 +98,9 @@ All GPTQ files are made with AutoGPTQ.
89
  | [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 26.78 GB | No | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. |
90
  | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ/tree/gptq-3bit-128g-actorder_True) | 3 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 28.03 GB | No | 3-bit, with group size 128g and act-order. Higher quality than 128g-False but poor AutoGPTQ CUDA speed. |
91
 
 
 
 
92
  ## How to download from branches
93
 
94
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Nous-Hermes-Llama2-70B-GPTQ:gptq-4bit-32g-actorder_True`
@@ -97,79 +109,79 @@ All GPTQ files are made with AutoGPTQ.
97
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ
98
  ```
99
  - In Python Transformers code, the branch is the `revision` parameter; see below.
100
-
 
101
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
102
 
103
  Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
104
 
105
- It is strongly recommended to use the text-generation-webui one-click-installers unless you know how to make a manual install.
106
 
107
  1. Click the **Model tab**.
108
  2. Under **Download custom model or LoRA**, enter `TheBloke/Nous-Hermes-Llama2-70B-GPTQ`.
109
  - To download from a specific branch, enter for example `TheBloke/Nous-Hermes-Llama2-70B-GPTQ:gptq-4bit-32g-actorder_True`
110
  - see Provided Files above for the list of branches for each option.
111
  3. Click **Download**.
112
- 4. The model will start downloading. Once it's finished it will say "Done"
113
  5. In the top left, click the refresh icon next to **Model**.
114
  6. In the **Model** dropdown, choose the model you just downloaded: `Nous-Hermes-Llama2-70B-GPTQ`
115
  7. The model will automatically load, and is now ready for use!
116
  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.
117
- * Note that you do not need to set GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
118
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
 
119
 
 
120
  ## How to use this GPTQ model from Python code
121
 
122
- First make sure you have [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) 0.3.1 or later installed:
123
 
124
- ```
125
- pip3 install auto-gptq
126
- ```
127
 
128
- If you have problems installing AutoGPTQ, please build from source instead:
 
 
129
  ```
 
 
 
 
130
  pip3 uninstall -y auto-gptq
131
  git clone https://github.com/PanQiWei/AutoGPTQ
132
  cd AutoGPTQ
133
  pip3 install .
134
  ```
135
 
136
- Then try the following example code:
 
 
 
 
 
 
 
 
137
 
138
  ```python
139
- from transformers import AutoTokenizer, pipeline, logging
140
- from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
141
 
142
  model_name_or_path = "TheBloke/Nous-Hermes-Llama2-70B-GPTQ"
143
-
144
- use_triton = False
 
 
 
 
145
 
146
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
147
 
148
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
149
- use_safetensors=True,
150
- trust_remote_code=False,
151
- device="cuda:0",
152
- use_triton=use_triton,
153
- quantize_config=None)
154
-
155
- """
156
- # To download from a specific branch, use the revision parameter, as in this example:
157
- # Note that `revision` requires AutoGPTQ 0.3.1 or later!
158
-
159
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
160
- revision="gptq-4bit-32g-actorder_True",
161
- use_safetensors=True,
162
- trust_remote_code=False,
163
- device="cuda:0",
164
- quantize_config=None)
165
- """
166
-
167
  prompt = "Tell me about AI"
168
  prompt_template=f'''### Instruction:
169
 
170
  {prompt}
171
 
172
  ### Response:
 
173
  '''
174
 
175
  print("\n\n*** Generate:")
@@ -180,9 +192,6 @@ print(tokenizer.decode(output[0]))
180
 
181
  # Inference can also be done using transformers' pipeline
182
 
183
- # Prevent printing spurious transformers error when using pipeline with AutoGPTQ
184
- logging.set_verbosity(logging.CRITICAL)
185
-
186
  print("*** Pipeline:")
187
  pipe = pipeline(
188
  "text-generation",
@@ -196,12 +205,17 @@ pipe = pipeline(
196
 
197
  print(pipe(prompt_template)[0]['generated_text'])
198
  ```
 
199
 
 
200
  ## Compatibility
201
 
202
- 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.
 
 
203
 
204
- ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
 
205
 
206
  <!-- footer start -->
207
  <!-- 200823 -->
@@ -226,7 +240,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
226
 
227
  **Special thanks to**: Aemon Algiz.
228
 
229
- **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
230
 
231
 
232
  Thank you to all my generous patrons and donaters!
 
36
  - Model creator: [NousResearch](https://huggingface.co/NousResearch)
37
  - Original model: [Nous Hermes Llama2 70B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b)
38
 
39
+ <!-- description start -->
40
  ## Description
41
 
42
  This repo contains GPTQ model files for [NousResearch's Nous Hermes Llama2 70B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b).
43
 
44
  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.
45
 
46
+ <!-- description end -->
47
+ <!-- repositories-available start -->
48
  ## Repositories available
49
 
50
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ)
51
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GGUF)
52
  * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GGML)
53
  * [NousResearch's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b)
54
+ <!-- repositories-available end -->
55
 
56
+ <!-- prompt-template start -->
57
  ## Prompt template: Alpaca-InstructOnly
58
 
59
  ```
 
62
  {prompt}
63
 
64
  ### Response:
65
+
66
  ```
67
 
68
+ <!-- prompt-template end -->
69
+
70
+ <!-- README_GPTQ.md-provided-files start -->
71
  ## Provided files and GPTQ parameters
72
 
73
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
74
 
75
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
76
 
77
+ 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.
78
 
79
  <details>
80
  <summary>Explanation of GPTQ parameters</summary>
81
 
82
  - Bits: The bit size of the quantised model.
83
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
84
+ - 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.
85
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
86
  - 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).
87
  - 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.
 
98
  | [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 26.78 GB | No | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. |
99
  | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ/tree/gptq-3bit-128g-actorder_True) | 3 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 28.03 GB | No | 3-bit, with group size 128g and act-order. Higher quality than 128g-False but poor AutoGPTQ CUDA speed. |
100
 
101
+ <!-- README_GPTQ.md-provided-files end -->
102
+
103
+ <!-- README_GPTQ.md-download-from-branches start -->
104
  ## How to download from branches
105
 
106
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Nous-Hermes-Llama2-70B-GPTQ:gptq-4bit-32g-actorder_True`
 
109
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Nous-Hermes-Llama2-70B-GPTQ
110
  ```
111
  - In Python Transformers code, the branch is the `revision` parameter; see below.
112
+ <!-- README_GPTQ.md-download-from-branches end -->
113
+ <!-- README_GPTQ.md-text-generation-webui start -->
114
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
115
 
116
  Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
117
 
118
+ 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.
119
 
120
  1. Click the **Model tab**.
121
  2. Under **Download custom model or LoRA**, enter `TheBloke/Nous-Hermes-Llama2-70B-GPTQ`.
122
  - To download from a specific branch, enter for example `TheBloke/Nous-Hermes-Llama2-70B-GPTQ:gptq-4bit-32g-actorder_True`
123
  - see Provided Files above for the list of branches for each option.
124
  3. Click **Download**.
125
+ 4. The model will start downloading. Once it's finished it will say "Done".
126
  5. In the top left, click the refresh icon next to **Model**.
127
  6. In the **Model** dropdown, choose the model you just downloaded: `Nous-Hermes-Llama2-70B-GPTQ`
128
  7. The model will automatically load, and is now ready for use!
129
  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.
130
+ * 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`.
131
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
132
+ <!-- README_GPTQ.md-text-generation-webui end -->
133
 
134
+ <!-- README_GPTQ.md-use-from-python start -->
135
  ## How to use this GPTQ model from Python code
136
 
137
+ ### Install the necessary packages
138
 
139
+ Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
 
 
140
 
141
+ ```shell
142
+ pip3 install transformers>=4.32.0 optimum>=1.12.0
143
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
144
  ```
145
+
146
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
147
+
148
+ ```shell
149
  pip3 uninstall -y auto-gptq
150
  git clone https://github.com/PanQiWei/AutoGPTQ
151
  cd AutoGPTQ
152
  pip3 install .
153
  ```
154
 
155
+ ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
156
+
157
+ If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
158
+ ```shell
159
+ pip3 uninstall -y transformers
160
+ pip3 install git+https://github.com/huggingface/transformers.git
161
+ ```
162
+
163
+ ### You can then use the following code
164
 
165
  ```python
166
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
167
 
168
  model_name_or_path = "TheBloke/Nous-Hermes-Llama2-70B-GPTQ"
169
+ # To use a different branch, change revision
170
+ # For example: revision="gptq-4bit-32g-actorder_True"
171
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
172
+ torch_dtype=torch.float16,
173
+ device_map="auto",
174
+ revision="main")
175
 
176
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
177
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
178
  prompt = "Tell me about AI"
179
  prompt_template=f'''### Instruction:
180
 
181
  {prompt}
182
 
183
  ### Response:
184
+
185
  '''
186
 
187
  print("\n\n*** Generate:")
 
192
 
193
  # Inference can also be done using transformers' pipeline
194
 
 
 
 
195
  print("*** Pipeline:")
196
  pipe = pipeline(
197
  "text-generation",
 
205
 
206
  print(pipe(prompt_template)[0]['generated_text'])
207
  ```
208
+ <!-- README_GPTQ.md-use-from-python end -->
209
 
210
+ <!-- README_GPTQ.md-compatibility start -->
211
  ## Compatibility
212
 
213
+ 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).
214
+
215
+ [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.
216
 
217
+ [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) is compatible with all GPTQ models.
218
+ <!-- README_GPTQ.md-compatibility end -->
219
 
220
  <!-- footer start -->
221
  <!-- 200823 -->
 
240
 
241
  **Special thanks to**: Aemon Algiz.
242
 
243
+ **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
244
 
245
 
246
  Thank you to all my generous patrons and donaters!