TheBloke commited on
Commit
b0d423c
1 Parent(s): a849819

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +35 -17
README.md CHANGED
@@ -1,4 +1,5 @@
1
  ---
 
2
  datasets:
3
  - conceptofmind/cot_submix_original
4
  - conceptofmind/flan2021_submix_original
@@ -9,10 +10,22 @@ language:
9
  - en
10
  license: llama2
11
  model_creator: Stability AI
12
- model_link: https://huggingface.co/stabilityai/StableBeluga-7b
13
  model_name: StableBeluga 7B
14
  model_type: llama
15
  pipeline_tag: text-generation
 
 
 
 
 
 
 
 
 
 
 
 
 
16
  quantized_by: TheBloke
17
  ---
18
 
@@ -48,9 +61,9 @@ Multiple GPTQ parameter permutations are provided; see Provided Files below for
48
  <!-- repositories-available start -->
49
  ## Repositories available
50
 
 
51
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ)
52
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/StableBeluga-7B-GGUF)
53
- * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/StableBeluga-7B-GGML)
54
  * [Stability AI's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/stabilityai/StableBeluga-7b)
55
  <!-- repositories-available end -->
56
 
@@ -70,6 +83,7 @@ Multiple GPTQ parameter permutations are provided; see Provided Files below for
70
 
71
  <!-- prompt-template end -->
72
 
 
73
  <!-- README_GPTQ.md-provided-files start -->
74
  ## Provided files and GPTQ parameters
75
 
@@ -94,22 +108,22 @@ All recent GPTQ files are made with AutoGPTQ, and all files in non-main branches
94
 
95
  | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
96
  | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
97
- | [main](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/main) | 4 | 128 | No | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 3.90 GB | Yes | Most compatible option. Good inference speed in AutoGPTQ and GPTQ-for-LLaMa. Lower inference quality than other options. |
98
- | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 4.28 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. Poor AutoGPTQ CUDA speed. |
99
- | [gptq-4bit-64g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-4bit-64g-actorder_True) | 4 | 64 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 4.02 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. |
100
- | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 3.90 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. |
101
- | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-8bit--1g-actorder_True) | 8 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 7.01 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements and to improve AutoGPTQ speed. |
102
- | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-8bit-128g-actorder_True) | 8 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 7.16 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher 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/StableBeluga-7B-GPTQ:gptq-4bit-32g-actorder_True`
110
  - With Git, you can clone a branch with:
111
  ```
112
- git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ
113
  ```
114
  - In Python Transformers code, the branch is the `revision` parameter; see below.
115
  <!-- README_GPTQ.md-download-from-branches end -->
@@ -122,7 +136,7 @@ It is strongly recommended to use the text-generation-webui one-click-installers
122
 
123
  1. Click the **Model tab**.
124
  2. Under **Download custom model or LoRA**, enter `TheBloke/StableBeluga-7B-GPTQ`.
125
- - To download from a specific branch, enter for example `TheBloke/StableBeluga-7B-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".
@@ -170,10 +184,10 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
170
 
171
  model_name_or_path = "TheBloke/StableBeluga-7B-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)
@@ -192,7 +206,7 @@ prompt_template=f'''### System:
192
  print("\n\n*** Generate:")
193
 
194
  input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
195
- output = model.generate(inputs=input_ids, temperature=0.7, max_new_tokens=512)
196
  print(tokenizer.decode(output[0]))
197
 
198
  # Inference can also be done using transformers' pipeline
@@ -203,9 +217,11 @@ pipe = pipeline(
203
  model=model,
204
  tokenizer=tokenizer,
205
  max_new_tokens=512,
 
206
  temperature=0.7,
207
  top_p=0.95,
208
- repetition_penalty=1.15
 
209
  )
210
 
211
  print(pipe(prompt_template)[0]['generated_text'])
@@ -230,10 +246,12 @@ For further support, and discussions on these models and AI in general, join us
230
 
231
  [TheBloke AI's Discord server](https://discord.gg/theblokeai)
232
 
233
- ## Thanks, and how to contribute.
234
 
235
  Thanks to the [chirper.ai](https://chirper.ai) team!
236
 
 
 
237
  I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
238
 
239
  If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
@@ -245,7 +263,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
245
 
246
  **Special thanks to**: Aemon Algiz.
247
 
248
- **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
249
 
250
 
251
  Thank you to all my generous patrons and donaters!
 
1
  ---
2
+ base_model: https://huggingface.co/stabilityai/StableBeluga-7b
3
  datasets:
4
  - conceptofmind/cot_submix_original
5
  - conceptofmind/flan2021_submix_original
 
10
  - en
11
  license: llama2
12
  model_creator: Stability AI
 
13
  model_name: StableBeluga 7B
14
  model_type: llama
15
  pipeline_tag: text-generation
16
+ prompt_template: '### System:
17
+
18
+ {system_message}
19
+
20
+
21
+ ### User:
22
+
23
+ {prompt}
24
+
25
+
26
+ ### Assistant:
27
+
28
+ '
29
  quantized_by: TheBloke
30
  ---
31
 
 
61
  <!-- repositories-available start -->
62
  ## Repositories available
63
 
64
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/StableBeluga-7B-AWQ)
65
  * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ)
66
  * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/StableBeluga-7B-GGUF)
 
67
  * [Stability AI's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/stabilityai/StableBeluga-7b)
68
  <!-- repositories-available end -->
69
 
 
83
 
84
  <!-- prompt-template end -->
85
 
86
+
87
  <!-- README_GPTQ.md-provided-files start -->
88
  ## Provided files and GPTQ parameters
89
 
 
108
 
109
  | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
110
  | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
111
+ | [main](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/main) | 4 | 128 | No | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 3.90 GB | Yes | 4-bit, without Act Order and group size 128g. |
112
+ | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 4.28 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. |
113
+ | [gptq-4bit-64g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-4bit-64g-actorder_True) | 4 | 64 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 4.02 GB | Yes | 4-bit, with Act Order and group size 64g. Uses less VRAM than 32g, but with slightly lower accuracy. |
114
+ | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 3.90 GB | Yes | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. |
115
+ | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-8bit--1g-actorder_True) | 8 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 7.01 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements. |
116
+ | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ/tree/gptq-8bit-128g-actorder_True) | 8 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 7.16 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. |
117
 
118
  <!-- README_GPTQ.md-provided-files end -->
119
 
120
  <!-- README_GPTQ.md-download-from-branches start -->
121
  ## How to download from branches
122
 
123
+ - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/StableBeluga-7B-GPTQ:main`
124
  - With Git, you can clone a branch with:
125
  ```
126
+ git clone --single-branch --branch main https://huggingface.co/TheBloke/StableBeluga-7B-GPTQ
127
  ```
128
  - In Python Transformers code, the branch is the `revision` parameter; see below.
129
  <!-- README_GPTQ.md-download-from-branches end -->
 
136
 
137
  1. Click the **Model tab**.
138
  2. Under **Download custom model or LoRA**, enter `TheBloke/StableBeluga-7B-GPTQ`.
139
+ - To download from a specific branch, enter for example `TheBloke/StableBeluga-7B-GPTQ:main`
140
  - see Provided Files above for the list of branches for each option.
141
  3. Click **Download**.
142
  4. The model will start downloading. Once it's finished it will say "Done".
 
184
 
185
  model_name_or_path = "TheBloke/StableBeluga-7B-GPTQ"
186
  # To use a different branch, change revision
187
+ # For example: revision="main"
188
  model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
 
189
  device_map="auto",
190
+ trust_remote_code=False,
191
  revision="main")
192
 
193
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
 
206
  print("\n\n*** Generate:")
207
 
208
  input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
209
+ output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512)
210
  print(tokenizer.decode(output[0]))
211
 
212
  # Inference can also be done using transformers' pipeline
 
217
  model=model,
218
  tokenizer=tokenizer,
219
  max_new_tokens=512,
220
+ do_sample=True,
221
  temperature=0.7,
222
  top_p=0.95,
223
+ top_k=40,
224
+ repetition_penalty=1.1
225
  )
226
 
227
  print(pipe(prompt_template)[0]['generated_text'])
 
246
 
247
  [TheBloke AI's Discord server](https://discord.gg/theblokeai)
248
 
249
+ ## Thanks, and how to contribute
250
 
251
  Thanks to the [chirper.ai](https://chirper.ai) team!
252
 
253
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
254
+
255
  I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
256
 
257
  If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
 
263
 
264
  **Special thanks to**: Aemon Algiz.
265
 
266
+ **Patreon special mentions**: Alicia Loh, Stephen Murray, K, Ajan Kanaga, RoA, Magnesian, Deo Leter, Olakabola, Eugene Pentland, zynix, Deep Realms, Raymond Fosdick, Elijah Stavena, Iucharbius, Erik Bjäreholt, Luis Javier Navarrete Lozano, Nicholas, theTransient, John Detwiler, alfie_i, knownsqashed, Mano Prime, Willem Michiel, Enrico Ros, LangChain4j, OG, Michael Dempsey, Pierre Kircher, Pedro Madruga, James Bentley, Thomas Belote, Luke @flexchar, Leonard Tan, Johann-Peter Hartmann, Illia Dulskyi, Fen Risland, Chadd, S_X, Jeff Scroggin, Ken Nordquist, Sean Connelly, Artur Olbinski, Swaroop Kallakuri, Jack West, Ai Maven, David Ziegler, Russ Johnson, transmissions 11, John Villwock, Alps Aficionado, Clay Pascal, Viktor Bowallius, Subspace Studios, Rainer Wilmers, Trenton Dambrowitz, vamX, Michael Levine, 준교 김, Brandon Frisco, Kalila, Trailburnt, Randy H, Talal Aujan, Nathan Dryer, Vadim, 阿明, ReadyPlayerEmma, Tiffany J. Kim, George Stoitzev, Spencer Kim, Jerry Meng, Gabriel Tamborski, Cory Kujawski, Jeffrey Morgan, Spiking Neurons AB, Edmond Seymore, Alexandros Triantafyllidis, Lone Striker, Cap'n Zoog, Nikolai Manek, danny, ya boyyy, Derek Yates, usrbinkat, Mandus, TL, Nathan LeClaire, subjectnull, Imad Khwaja, webtim, Raven Klaugh, Asp the Wyvern, Gabriel Puliatti, Caitlyn Gatomon, Joseph William Delisle, Jonathan Leane, Luke Pendergrass, SuperWojo, Sebastain Graf, Will Dee, Fred von Graf, Andrey, Dan Guido, Daniel P. Andersen, Nitin Borwankar, Elle, Vitor Caleffi, biorpg, jjj, NimbleBox.ai, Pieter, Matthew Berman, terasurfer, Michael Davis, Alex, Stanislav Ovsiannikov
267
 
268
 
269
  Thank you to all my generous patrons and donaters!