TheBloke commited on
Commit
f8c6abe
1 Parent(s): 054fbf6

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +104 -111
README.md CHANGED
@@ -3,8 +3,12 @@ inference: false
3
  language:
4
  - en
5
  license: other
 
 
 
6
  model_type: llama
7
  pipeline_tag: text-generation
 
8
  tags:
9
  - facebook
10
  - meta
@@ -30,76 +34,76 @@ tags:
30
  <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
31
  <!-- header end -->
32
 
33
- # Meta's Llama 2 70B Chat GPTQ
 
 
34
 
35
- These files are GPTQ model files for [Meta's Llama 2 70B Chat](https://huggingface.co/meta-llama/Llama-2-70b-chat-hf).
 
36
 
37
- 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.
38
-
39
- Many thanks to William Beauchamp from [Chai](https://chai-research.com/) for providing the hardware for these quantisations!
40
-
41
- ## ExLlama support for 70B is here!
42
-
43
- As of [this commit](https://github.com/turboderp/exllama/commit/b3aea521859b83cfd889c4c00c05a323313b7fee), ExLlama has support for Llama 2 70B models.
44
-
45
- Please make sure you update ExLlama to the latest version. If you are a text-generation-webui one-click user, you must first uninstall the ExLlama wheel, then clone ExLlama into `text-generation-webui/repositories`; full instructions are below.
46
 
47
- Now that we have ExLlama, that is the recommended loader to use for these models, as performance should be better than with AutoGPTQ and GPTQ-for-LLaMa, and you will be able to use the higher accuracy models, eg 128g + Act-Order.
48
-
49
- Reminder: ExLlama does not support 3-bit models, so if you wish to try those quants, you will need to use AutoGPTQ or GPTQ-for-LLaMa.
50
-
51
- ## AutoGPTQ and GPTQ-for-LLaMa compatibility
52
-
53
- Please update AutoGPTQ to version 0.3.1 or later. This will also update Transformers to 4.31.0, which is required for Llama 70B compatibility.
54
-
55
- If you're using GPTQ-for-LLaMa, please update Transformers manually with:
56
- ```
57
- pip3 install "transformers>=4.31.0"
58
- ```
59
 
 
 
60
  ## Repositories available
61
 
62
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ)
63
- * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference.](https://huggingface.co/TheBloke/Llama-2-70B-chat-GGML)
64
- * [Original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/TheBloke/Llama-2-70B-chat-fp16)
 
 
65
 
 
66
  ## Prompt template: Llama-2-Chat
67
 
68
  ```
69
  [INST] <<SYS>>
70
  You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
71
  <</SYS>>
72
-
73
- {prompt} [/INST]
74
- ```
75
-
76
- To continue a conversation:
77
 
78
  ```
79
- [INST] <<SYS>>
80
- You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
81
- <</SYS>>
82
 
83
- {prompt} [/INST] {model_reply} [INST] {prompt} [/INST]
84
- ```
85
 
86
- ## Provided files
 
87
 
88
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
89
 
90
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
91
 
92
- | Branch | Bits | Group Size | Act Order (desc_act) | File Size | ExLlama Compatible? | Made With | Description |
93
- | ------ | ---- | ---------- | -------------------- | --------- | ------------------- | --------- | ----------- |
94
- | main | 4 | -1 | True | 35.33 GB | True | AutoGPTQ | Most compatible option. Good inference speed in AutoGPTQ and GPTQ-for-LLaMa. Lower inference quality than other options. |
95
- | gptq-4bit-32g-actorder_True | 4 | 32 | True | 40.66 GB | True | AutoGPTQ | 4-bit, with Act Order and group size. 32g gives highest possible inference quality, with maximum VRAM usage. Poor AutoGPTQ CUDA speed. |
96
- | gptq-4bit-64g-actorder_True | 4 | 64 | True | 37.99 GB | True | AutoGPTQ | 4-bit, with Act Order and group size. 64g uses less VRAM than 32g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
97
- | gptq-4bit-128g-actorder_True | 4 | 128 | True | 36.65 GB | True | AutoGPTQ | 4-bit, with Act Order and group size. 128g uses even less VRAM, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
98
- | gptq-3bit--1g-actorder_True | 3 | None | True | 26.78 GB | False | AutoGPTQ | 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_False | 3 | 128 | False | 28.03 GB | False | AutoGPTQ | 3-bit, with group size 128g but no act-order. Slightly higher VRAM requirements than 3-bit None. |
100
- | gptq-3bit-128g-actorder_True | 3 | 128 | True | 28.03 GB | False | AutoGPTQ | 3-bit, with group size 128g and act-order. Higher quality than 128g-False but poor AutoGPTQ CUDA speed. |
101
- | gptq-3bit-64g-actorder_True | 3 | 64 | True | 29.30 GB | False | AutoGPTQ | 3-bit, with group size 64g and act-order. Highest quality 3-bit option. Poor AutoGPTQ CUDA speed. |
102
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
103
  ## How to download from branches
104
 
105
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Llama-2-70B-chat-GPTQ:gptq-4bit-32g-actorder_True`
@@ -108,91 +112,78 @@ Each separate quant is in a different branch. See below for instructions on fet
108
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ
109
  ```
110
  - In Python Transformers code, the branch is the `revision` parameter; see below.
 
 
 
111
 
112
- ### How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
113
-
114
- Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui), which includes support for Llama 2 models.
115
 
116
- It is strongly recommended to use the text-generation-webui one-click-installers unless you know how to make a manual install.
117
-
118
- ### Use ExLlama (4-bit models only) - recommended option if you have enough VRAM for 4-bit
119
-
120
- ExLlama has now been updated to support Llama 2 70B. Make sure you're using the latest version of ExLlama, and text-generation-webui if you're using that.
121
-
122
- ### Downloading and running the model in text-generation-webui
123
 
124
  1. Click the **Model tab**.
125
  2. Under **Download custom model or LoRA**, enter `TheBloke/Llama-2-70B-chat-GPTQ`.
126
  - To download from a specific branch, enter for example `TheBloke/Llama-2-70B-chat-GPTQ:gptq-4bit-32g-actorder_True`
127
  - see Provided Files above for the list of branches for each option.
128
  3. Click **Download**.
129
- 4. The model will start downloading. Once it's finished it will say "Done"
130
- 5. Set Loader to ExLlama if you plan to use a 4-bit file, or else choose AutoGPTQ or GPTQ-for-LLaMA.
131
- - If you use AutoGPTQ, make sure "No inject fused attention" is ticked
132
- 6. In the top left, click the refresh icon next to **Model**.
133
- 7. In the **Model** dropdown, choose the model you just downloaded: `TheBloke/Llama-2-70B-chat-GPTQ`
134
- 8. The model will automatically load, and is now ready for use!
135
- 9. Then click **Save settings for this model** followed by **Reload the Model** in the top right to make sure your settings are persisted.
136
- 10. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
137
-
 
138
  ## How to use this GPTQ model from Python code
139
 
140
- First make sure you have [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) installed, version 0.3.1 or 0.3.2 or later:
141
 
142
- ```
143
- pip3 install auto-gptq
 
 
 
144
  ```
145
 
146
- You also need the latest Transformers code from Github:
147
 
148
- ```
149
- pip3 install "transformers>=4.31.0"
 
 
 
150
  ```
151
 
152
- You must set `inject_fused_attention=False` as shown below.
153
 
154
- Then try the following example code:
 
 
 
 
 
 
155
 
156
  ```python
157
- from transformers import AutoTokenizer, pipeline, logging
158
- from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
159
 
160
  model_name_or_path = "TheBloke/Llama-2-70B-chat-GPTQ"
161
- model_basename = "model"
162
-
163
- use_triton = False
 
 
 
164
 
165
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
166
 
167
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
168
- model_basename=model_basename,
169
- inject_fused_attention=False, # Required for Llama 2 70B model at this time.
170
- use_safetensors=True,
171
- trust_remote_code=False,
172
- device="cuda:0",
173
- use_triton=use_triton,
174
- quantize_config=None)
175
-
176
- """
177
- To download from a specific branch, use the revision parameter, as in this example:
178
-
179
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
180
- revision="gptq-4bit-32g-actorder_True",
181
- model_basename=model_basename,
182
- inject_fused_attention=False, # Required for Llama 2 70B model at this time.
183
- use_safetensors=True,
184
- trust_remote_code=False,
185
- device="cuda:0",
186
- quantize_config=None)
187
- """
188
-
189
  prompt = "Tell me about AI"
190
- system_message = "You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information."
191
  prompt_template=f'''[INST] <<SYS>>
192
- {system_message}
193
  <</SYS>>
 
194
 
195
- {prompt} [/INST]
196
  '''
197
 
198
  print("\n\n*** Generate:")
@@ -203,9 +194,6 @@ print(tokenizer.decode(output[0]))
203
 
204
  # Inference can also be done using transformers' pipeline
205
 
206
- # Prevent printing spurious transformers error when using pipeline with AutoGPTQ
207
- logging.set_verbosity(logging.CRITICAL)
208
-
209
  print("*** Pipeline:")
210
  pipe = pipeline(
211
  "text-generation",
@@ -219,14 +207,17 @@ pipe = pipeline(
219
 
220
  print(pipe(prompt_template)[0]['generated_text'])
221
  ```
 
222
 
 
223
  ## Compatibility
224
 
225
- 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.
226
 
227
- ExLlama is now compatible with Llama 2 70B models, as of [this commit](https://github.com/turboderp/exllama/commit/b3aea521859b83cfd889c4c00c05a323313b7fee).
228
 
229
- Please see the Provided Files table above for per-file compatibility.
 
230
 
231
  <!-- footer start -->
232
  <!-- 200823 -->
@@ -251,7 +242,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
251
 
252
  **Special thanks to**: Aemon Algiz.
253
 
254
- **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
255
 
256
 
257
  Thank you to all my generous patrons and donaters!
@@ -260,7 +251,7 @@ And thank you again to a16z for their generous grant.
260
 
261
  <!-- footer end -->
262
 
263
- # Original model card: Meta's Llama 2 70B Chat
264
 
265
  # **Llama 2**
266
  Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 70B fine-tuned model, optimized for dialogue use cases and converted for the Hugging Face Transformers format. Links to other models can be found in the index at the bottom.
@@ -295,6 +286,8 @@ Meta developed and publicly released the Llama 2 family of large language models
295
 
296
  **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/)
297
 
 
 
298
  ## Intended Use
299
  **Intended Use Cases** Llama 2 is intended for commercial and research use in English. Tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.
300
 
 
3
  language:
4
  - en
5
  license: other
6
+ model_creator: Meta Llama 2
7
+ model_link: https://huggingface.co/meta-llama/Llama-2-70b-chat-hf
8
+ model_name: Llama 2 70B Chat
9
  model_type: llama
10
  pipeline_tag: text-generation
11
+ quantized_by: TheBloke
12
  tags:
13
  - facebook
14
  - meta
 
34
  <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
35
  <!-- header end -->
36
 
37
+ # Llama 2 70B Chat - GPTQ
38
+ - Model creator: [Meta Llama 2](https://huggingface.co/meta-llama)
39
+ - Original model: [Llama 2 70B Chat](https://huggingface.co/meta-llama/Llama-2-70b-chat-hf)
40
 
41
+ <!-- description start -->
42
+ ## Description
43
 
44
+ This repo contains GPTQ model files for [Meta Llama 2's Llama 2 70B Chat](https://huggingface.co/meta-llama/Llama-2-70b-chat-hf).
 
 
 
 
 
 
 
 
45
 
46
+ 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.
 
 
 
 
 
 
 
 
 
 
 
47
 
48
+ <!-- description end -->
49
+ <!-- repositories-available start -->
50
  ## Repositories available
51
 
52
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ)
53
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Llama-2-70B-chat-GGUF)
54
+ * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/Llama-2-70B-chat-GGML)
55
+ * [Meta Llama 2's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/meta-llama/Llama-2-70b-chat-hf)
56
+ <!-- repositories-available end -->
57
 
58
+ <!-- prompt-template start -->
59
  ## Prompt template: Llama-2-Chat
60
 
61
  ```
62
  [INST] <<SYS>>
63
  You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
64
  <</SYS>>
65
+ {prompt}[/INST]
 
 
 
 
66
 
67
  ```
 
 
 
68
 
69
+ <!-- prompt-template end -->
 
70
 
71
+ <!-- README_GPTQ.md-provided-files start -->
72
+ ## Provided files and GPTQ parameters
73
 
74
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
75
 
76
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
77
 
78
+ 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.
79
+
80
+ <details>
81
+ <summary>Explanation of GPTQ parameters</summary>
 
 
 
 
 
 
82
 
83
+ - Bits: The bit size of the quantised model.
84
+ - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
85
+ - 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.
86
+ - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
87
+ - 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).
88
+ - 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
+ - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama models in 4-bit.
90
+
91
+ </details>
92
+
93
+ | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
94
+ | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
95
+ | [main](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/main) | 4 | None | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 35.33 GB | Yes | Most compatible option. Good inference speed in AutoGPTQ and GPTQ-for-LLaMa. Lower inference quality than other options. |
96
+ | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 40.66 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. Poor AutoGPTQ CUDA speed. |
97
+ | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 36.65 GB | Yes | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
98
+ | [gptq-3bit-64g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-3bit-64g-actorder_True) | 3 | 64 | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 29.30 GB | No | 3-bit, with group size 64g and act-order. Poor AutoGPTQ CUDA speed. |
99
+ | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-3bit-128g-actorder_True) | 3 | 128 | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 28.03 GB | No | 3-bit, with group size 128g and act-order. Higher quality than 128g-False but poor AutoGPTQ CUDA speed. |
100
+ | [gptq-3bit-128g-actorder_False](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-3bit-128g-actorder_False) | 3 | 128 | No | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 28.03 GB | No | 3-bit, with group size 128g but no act-order. Slightly higher VRAM requirements than 3-bit None. |
101
+ | [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 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. |
102
+ | [gptq-4bit-64g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ/tree/gptq-4bit-64g-actorder_True) | 4 | 64 | Yes | 0.01 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 2048 | 37.99 GB | Yes | 4-bit, with Act Order and group size 64g. Uses less VRAM than 32g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
103
+
104
+ <!-- README_GPTQ.md-provided-files end -->
105
+
106
+ <!-- README_GPTQ.md-download-from-branches start -->
107
  ## How to download from branches
108
 
109
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Llama-2-70B-chat-GPTQ:gptq-4bit-32g-actorder_True`
 
112
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Llama-2-70B-chat-GPTQ
113
  ```
114
  - In Python Transformers code, the branch is the `revision` parameter; see below.
115
+ <!-- README_GPTQ.md-download-from-branches end -->
116
+ <!-- README_GPTQ.md-text-generation-webui start -->
117
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
118
 
119
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
 
 
120
 
121
+ 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.
 
 
 
 
 
 
122
 
123
  1. Click the **Model tab**.
124
  2. Under **Download custom model or LoRA**, enter `TheBloke/Llama-2-70B-chat-GPTQ`.
125
  - To download from a specific branch, enter for example `TheBloke/Llama-2-70B-chat-GPTQ:gptq-4bit-32g-actorder_True`
126
  - see Provided Files above for the list of branches for each option.
127
  3. Click **Download**.
128
+ 4. The model will start downloading. Once it's finished it will say "Done".
129
+ 5. In the top left, click the refresh icon next to **Model**.
130
+ 6. In the **Model** dropdown, choose the model you just downloaded: `Llama-2-70B-chat-GPTQ`
131
+ 7. The model will automatically load, and is now ready for use!
132
+ 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.
133
+ * 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`.
134
+ 9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
135
+ <!-- README_GPTQ.md-text-generation-webui end -->
136
+
137
+ <!-- README_GPTQ.md-use-from-python start -->
138
  ## How to use this GPTQ model from Python code
139
 
140
+ ### Install the necessary packages
141
 
142
+ Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
143
+
144
+ ```shell
145
+ pip3 install transformers>=4.32.0 optimum>=1.12.0
146
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
147
  ```
148
 
149
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
150
 
151
+ ```shell
152
+ pip3 uninstall -y auto-gptq
153
+ git clone https://github.com/PanQiWei/AutoGPTQ
154
+ cd AutoGPTQ
155
+ pip3 install .
156
  ```
157
 
158
+ ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
159
 
160
+ If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
161
+ ```shell
162
+ pip3 uninstall -y transformers
163
+ pip3 install git+https://github.com/huggingface/transformers.git
164
+ ```
165
+
166
+ ### You can then use the following code
167
 
168
  ```python
169
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
170
 
171
  model_name_or_path = "TheBloke/Llama-2-70B-chat-GPTQ"
172
+ # To use a different branch, change revision
173
+ # For example: revision="gptq-4bit-32g-actorder_True"
174
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
175
+ torch_dtype=torch.float16,
176
+ device_map="auto",
177
+ revision="main")
178
 
179
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
180
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
181
  prompt = "Tell me about AI"
 
182
  prompt_template=f'''[INST] <<SYS>>
183
+ You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
184
  <</SYS>>
185
+ {prompt}[/INST]
186
 
 
187
  '''
188
 
189
  print("\n\n*** Generate:")
 
194
 
195
  # Inference can also be done using transformers' pipeline
196
 
 
 
 
197
  print("*** Pipeline:")
198
  pipe = pipeline(
199
  "text-generation",
 
207
 
208
  print(pipe(prompt_template)[0]['generated_text'])
209
  ```
210
+ <!-- README_GPTQ.md-use-from-python end -->
211
 
212
+ <!-- README_GPTQ.md-compatibility start -->
213
  ## Compatibility
214
 
215
+ 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).
216
 
217
+ [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.
218
 
219
+ [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) is compatible with all GPTQ models.
220
+ <!-- README_GPTQ.md-compatibility end -->
221
 
222
  <!-- footer start -->
223
  <!-- 200823 -->
 
242
 
243
  **Special thanks to**: Aemon Algiz.
244
 
245
+ **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
246
 
247
 
248
  Thank you to all my generous patrons and donaters!
 
251
 
252
  <!-- footer end -->
253
 
254
+ # Original model card: Meta Llama 2's Llama 2 70B Chat
255
 
256
  # **Llama 2**
257
  Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 70B fine-tuned model, optimized for dialogue use cases and converted for the Hugging Face Transformers format. Links to other models can be found in the index at the bottom.
 
286
 
287
  **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/)
288
 
289
+ **Research Paper** ["Llama-2: Open Foundation and Fine-tuned Chat Models"](arxiv.org/abs/2307.09288)
290
+
291
  ## Intended Use
292
  **Intended Use Cases** Llama 2 is intended for commercial and research use in English. Tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.
293