TheBloke commited on
Commit
a5791f7
1 Parent(s): 2f56e47

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +53 -41
README.md CHANGED
@@ -36,19 +36,24 @@ tags:
36
  - Model creator: [ddobokki](https://huggingface.co/ddobokki)
37
  - Original model: [Llama 2 70B Orca 200k](https://huggingface.co/ddobokki/Llama-2-70b-orca-200k)
38
 
 
39
  ## Description
40
 
41
  This repo contains GPTQ model files for [ddobokki's Llama 2 70B Orca 200k](https://huggingface.co/ddobokki/Llama-2-70b-orca-200k).
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/Llama-2-70B-Orca-200k-GPTQ)
48
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GGUF)
49
  * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GGML)
50
  * [ddobokki's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/ddobokki/Llama-2-70b-orca-200k)
 
51
 
 
52
  ## Prompt template: Guanaco
53
 
54
  ```
@@ -57,20 +62,23 @@ Multiple GPTQ parameter permutations are provided; see Provided Files below for
57
 
58
  ```
59
 
 
 
 
60
  ## Provided files and GPTQ parameters
61
 
62
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
63
 
64
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
65
 
66
- All GPTQ files are made with AutoGPTQ.
67
 
68
  <details>
69
  <summary>Explanation of GPTQ parameters</summary>
70
 
71
  - Bits: The bit size of the quantised model.
72
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
73
- - 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.
74
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
75
  - 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).
76
  - 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.
@@ -85,6 +93,9 @@ All GPTQ files are made with AutoGPTQ.
85
  | [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 26.77 GB | No | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. |
86
  | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-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. |
87
 
 
 
 
88
  ## How to download from branches
89
 
90
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Llama-2-70B-Orca-200k-GPTQ:gptq-4bit-32g-actorder_True`
@@ -93,73 +104,72 @@ All GPTQ files are made with AutoGPTQ.
93
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GPTQ
94
  ```
95
  - In Python Transformers code, the branch is the `revision` parameter; see below.
96
-
 
97
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
98
 
99
  Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
100
 
101
- It is strongly recommended to use the text-generation-webui one-click-installers unless you know how to make a manual install.
102
 
103
  1. Click the **Model tab**.
104
  2. Under **Download custom model or LoRA**, enter `TheBloke/Llama-2-70B-Orca-200k-GPTQ`.
105
  - To download from a specific branch, enter for example `TheBloke/Llama-2-70B-Orca-200k-GPTQ:gptq-4bit-32g-actorder_True`
106
  - see Provided Files above for the list of branches for each option.
107
  3. Click **Download**.
108
- 4. The model will start downloading. Once it's finished it will say "Done"
109
  5. In the top left, click the refresh icon next to **Model**.
110
  6. In the **Model** dropdown, choose the model you just downloaded: `Llama-2-70B-Orca-200k-GPTQ`
111
  7. The model will automatically load, and is now ready for use!
112
  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.
113
- * Note that you do not need to set GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
114
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
 
115
 
 
116
  ## How to use this GPTQ model from Python code
117
 
118
- First make sure you have [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) 0.3.1 or later installed:
119
 
120
- ```
121
- pip3 install auto-gptq
122
- ```
123
 
124
- If you have problems installing AutoGPTQ, please build from source instead:
 
 
125
  ```
 
 
 
 
126
  pip3 uninstall -y auto-gptq
127
  git clone https://github.com/PanQiWei/AutoGPTQ
128
  cd AutoGPTQ
129
  pip3 install .
130
  ```
131
 
132
- Then try the following example code:
 
 
 
 
 
 
 
 
133
 
134
  ```python
135
- from transformers import AutoTokenizer, pipeline, logging
136
- from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
137
 
138
  model_name_or_path = "TheBloke/Llama-2-70B-Orca-200k-GPTQ"
139
-
140
- use_triton = False
 
 
 
 
141
 
142
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
143
 
144
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
145
- use_safetensors=True,
146
- trust_remote_code=False,
147
- device="cuda:0",
148
- use_triton=use_triton,
149
- quantize_config=None)
150
-
151
- """
152
- # To download from a specific branch, use the revision parameter, as in this example:
153
- # Note that `revision` requires AutoGPTQ 0.3.1 or later!
154
-
155
- model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
156
- revision="gptq-4bit-32g-actorder_True",
157
- use_safetensors=True,
158
- trust_remote_code=False,
159
- device="cuda:0",
160
- quantize_config=None)
161
- """
162
-
163
  prompt = "Tell me about AI"
164
  prompt_template=f'''### Human: {prompt}
165
  ### Assistant:
@@ -174,9 +184,6 @@ print(tokenizer.decode(output[0]))
174
 
175
  # Inference can also be done using transformers' pipeline
176
 
177
- # Prevent printing spurious transformers error when using pipeline with AutoGPTQ
178
- logging.set_verbosity(logging.CRITICAL)
179
-
180
  print("*** Pipeline:")
181
  pipe = pipeline(
182
  "text-generation",
@@ -190,12 +197,17 @@ pipe = pipeline(
190
 
191
  print(pipe(prompt_template)[0]['generated_text'])
192
  ```
 
193
 
 
194
  ## Compatibility
195
 
196
- 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.
 
 
197
 
198
- ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
 
199
 
200
  <!-- footer start -->
201
  <!-- 200823 -->
@@ -220,7 +232,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
220
 
221
  **Special thanks to**: Aemon Algiz.
222
 
223
- **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
224
 
225
 
226
  Thank you to all my generous patrons and donaters!
 
36
  - Model creator: [ddobokki](https://huggingface.co/ddobokki)
37
  - Original model: [Llama 2 70B Orca 200k](https://huggingface.co/ddobokki/Llama-2-70b-orca-200k)
38
 
39
+ <!-- description start -->
40
  ## Description
41
 
42
  This repo contains GPTQ model files for [ddobokki's Llama 2 70B Orca 200k](https://huggingface.co/ddobokki/Llama-2-70b-orca-200k).
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/Llama-2-70B-Orca-200k-GPTQ)
51
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GGUF)
52
  * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GGML)
53
  * [ddobokki's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/ddobokki/Llama-2-70b-orca-200k)
54
+ <!-- repositories-available end -->
55
 
56
+ <!-- prompt-template start -->
57
  ## Prompt template: Guanaco
58
 
59
  ```
 
62
 
63
  ```
64
 
65
+ <!-- prompt-template end -->
66
+
67
+ <!-- README_GPTQ.md-provided-files start -->
68
  ## Provided files and GPTQ parameters
69
 
70
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
71
 
72
  Each separate quant is in a different branch. See below for instructions on fetching from different branches.
73
 
74
+ 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.
75
 
76
  <details>
77
  <summary>Explanation of GPTQ parameters</summary>
78
 
79
  - Bits: The bit size of the quantised model.
80
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
81
+ - 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.
82
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
83
  - 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).
84
  - 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.
 
93
  | [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 26.77 GB | No | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. |
94
  | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-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. |
95
 
96
+ <!-- README_GPTQ.md-provided-files end -->
97
+
98
+ <!-- README_GPTQ.md-download-from-branches start -->
99
  ## How to download from branches
100
 
101
  - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Llama-2-70B-Orca-200k-GPTQ:gptq-4bit-32g-actorder_True`
 
104
  git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Llama-2-70B-Orca-200k-GPTQ
105
  ```
106
  - In Python Transformers code, the branch is the `revision` parameter; see below.
107
+ <!-- README_GPTQ.md-download-from-branches end -->
108
+ <!-- README_GPTQ.md-text-generation-webui start -->
109
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
110
 
111
  Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
112
 
113
+ 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.
114
 
115
  1. Click the **Model tab**.
116
  2. Under **Download custom model or LoRA**, enter `TheBloke/Llama-2-70B-Orca-200k-GPTQ`.
117
  - To download from a specific branch, enter for example `TheBloke/Llama-2-70B-Orca-200k-GPTQ:gptq-4bit-32g-actorder_True`
118
  - see Provided Files above for the list of branches for each option.
119
  3. Click **Download**.
120
+ 4. The model will start downloading. Once it's finished it will say "Done".
121
  5. In the top left, click the refresh icon next to **Model**.
122
  6. In the **Model** dropdown, choose the model you just downloaded: `Llama-2-70B-Orca-200k-GPTQ`
123
  7. The model will automatically load, and is now ready for use!
124
  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.
125
+ * 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`.
126
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
127
+ <!-- README_GPTQ.md-text-generation-webui end -->
128
 
129
+ <!-- README_GPTQ.md-use-from-python start -->
130
  ## How to use this GPTQ model from Python code
131
 
132
+ ### Install the necessary packages
133
 
134
+ Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
 
 
135
 
136
+ ```shell
137
+ pip3 install transformers>=4.32.0 optimum>=1.12.0
138
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
139
  ```
140
+
141
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
142
+
143
+ ```shell
144
  pip3 uninstall -y auto-gptq
145
  git clone https://github.com/PanQiWei/AutoGPTQ
146
  cd AutoGPTQ
147
  pip3 install .
148
  ```
149
 
150
+ ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
151
+
152
+ If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
153
+ ```shell
154
+ pip3 uninstall -y transformers
155
+ pip3 install git+https://github.com/huggingface/transformers.git
156
+ ```
157
+
158
+ ### You can then use the following code
159
 
160
  ```python
161
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
162
 
163
  model_name_or_path = "TheBloke/Llama-2-70B-Orca-200k-GPTQ"
164
+ # To use a different branch, change revision
165
+ # For example: revision="gptq-4bit-32g-actorder_True"
166
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
167
+ torch_dtype=torch.float16,
168
+ device_map="auto",
169
+ revision="main")
170
 
171
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
172
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
173
  prompt = "Tell me about AI"
174
  prompt_template=f'''### Human: {prompt}
175
  ### Assistant:
 
184
 
185
  # Inference can also be done using transformers' pipeline
186
 
 
 
 
187
  print("*** Pipeline:")
188
  pipe = pipeline(
189
  "text-generation",
 
197
 
198
  print(pipe(prompt_template)[0]['generated_text'])
199
  ```
200
+ <!-- README_GPTQ.md-use-from-python end -->
201
 
202
+ <!-- README_GPTQ.md-compatibility start -->
203
  ## Compatibility
204
 
205
+ 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).
206
+
207
+ [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.
208
 
209
+ [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) is compatible with all GPTQ models.
210
+ <!-- README_GPTQ.md-compatibility end -->
211
 
212
  <!-- footer start -->
213
  <!-- 200823 -->
 
232
 
233
  **Special thanks to**: Aemon Algiz.
234
 
235
+ **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
236
 
237
 
238
  Thank you to all my generous patrons and donaters!