TheBloke commited on
Commit
b144529
1 Parent(s): 7756381

Update for Transformers GPTQ support

Browse files
README.md CHANGED
@@ -235,17 +235,20 @@ widget:
235
  ---
236
 
237
  <!-- header start -->
238
- <div style="width: 100%;">
239
- <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
 
240
  </div>
241
  <div style="display: flex; justify-content: space-between; width: 100%;">
242
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
243
- <p><a href="https://discord.gg/theblokeai">Chat & support: my new Discord server</a></p>
244
  </div>
245
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
246
- <p><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
247
  </div>
248
  </div>
 
 
249
  <!-- header end -->
250
 
251
  # Octocoder - GPTQ
@@ -294,11 +297,11 @@ All GPTQ files are made with AutoGPTQ.
294
 
295
  | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
296
  | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
297
- | [main](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/main) | 4 | 128 | No | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 9.20 GB | No | Most compatible option. Good inference speed in AutoGPTQ and GPTQ-for-LLaMa. Lower inference quality than other options. |
298
- | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 10.09 GB | No | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. Poor AutoGPTQ CUDA speed. |
299
- | [gptq-4bit-64g-actorder_True](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/gptq-4bit-64g-actorder_True) | 4 | 64 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 9.49 GB | No | 4-bit, with Act Order and group size 64g. Uses less VRAM than 32g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
300
- | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 9.20 GB | No | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
301
- | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/Octocoder-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) | 8192 | 16.49 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements and to improve AutoGPTQ speed. |
302
  | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/Octocoder-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) | 8192 | 16.84 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. Poor AutoGPTQ CUDA speed. |
303
 
304
  ## How to download from branches
@@ -413,6 +416,7 @@ The files provided will work with AutoGPTQ (CUDA and Triton modes), GPTQ-for-LLa
413
  ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
414
 
415
  <!-- footer start -->
 
416
  ## Discord
417
 
418
  For further support, and discussions on these models and AI in general, join us at:
@@ -434,11 +438,13 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
434
 
435
  **Special thanks to**: Aemon Algiz.
436
 
437
- **Patreon special mentions**: Ajan Kanaga, David Ziegler, Raymond Fosdick, SuperWojo, Sam, webtim, Steven Wood, knownsqashed, Tony Hughes, Junyu Yang, J, Olakabola, Dan Guido, Stephen Murray, John Villwock, vamX, William Sang, Sean Connelly, LangChain4j, Olusegun Samson, Fen Risland, Derek Yates, Karl Bernard, transmissions 11, Trenton Dambrowitz, Pieter, Preetika Verma, Swaroop Kallakuri, Andrey, Slarti, Jonathan Leane, Michael Levine, Kalila, Joseph William Delisle, Rishabh Srivastava, Deo Leter, Luke Pendergrass, Spencer Kim, Geoffrey Montalvo, Thomas Belote, Jeffrey Morgan, Mandus, ya boyyy, Matthew Berman, Magnesian, Ai Maven, senxiiz, Alps Aficionado, Luke @flexchar, Raven Klaugh, Imad Khwaja, Gabriel Puliatti, Johann-Peter Hartmann, usrbinkat, Spiking Neurons AB, Artur Olbinski, chris gileta, danny, Willem Michiel, WelcomeToTheClub, Deep Realms, alfie_i, Dave, Leonard Tan, NimbleBox.ai, Randy H, Daniel P. Andersen, Pyrater, Will Dee, Elle, Space Cruiser, Gabriel Tamborski, Asp the Wyvern, Illia Dulskyi, Nikolai Manek, Sid, Brandon Frisco, Nathan LeClaire, Edmond Seymore, Enrico Ros, Pedro Madruga, Eugene Pentland, John Detwiler, Mano Prime, Stanislav Ovsiannikov, Alex, Vitor Caleffi, K, biorpg, Michael Davis, Lone Striker, Pierre Kircher, theTransient, Fred von Graf, Sebastain Graf, Vadim, Iucharbius, Clay Pascal, Chadd, Mesiah Bishop, terasurfer, Rainer Wilmers, Alexandros Triantafyllidis, Stefan Sabev, Talal Aujan, Cory Kujawski, Viktor Bowallius, subjectnull, ReadyPlayerEmma, zynix
438
 
439
 
440
  Thank you to all my generous patrons and donaters!
441
 
 
 
442
  <!-- footer end -->
443
 
444
  # Original model card: BigCode's Octocoder
@@ -463,27 +469,27 @@ Thank you to all my generous patrons and donaters!
463
  - **OctoPack🐙🎒:**
464
  <table>
465
  <tr>
466
- <th>Data</t>
467
  <th><a href=https://huggingface.co/datasets/bigcode/commitpack>CommitPack</a></th>
468
  <td>4TB of GitHub commits across 350 programming languages</td>
469
  </tr>
470
  <tr>
471
- <th></t>
472
  <th><a href=https://huggingface.co/datasets/bigcode/commitpackft>CommitPackFT</a></th>
473
  <td>Filtered version of CommitPack for high-quality commit messages that resemble instructions</td>
474
  </tr>
475
  <tr>
476
- <th>Model</t>
477
  <th><a href=https://huggingface.co/bigcode/octocoder>OctoCoder</a></th>
478
  <td>StarCoder (16B parameters) instruction tuned on CommitPackFT + OASST</td>
479
  </tr>
480
  <tr>
481
- <th></t>
482
  <th><a href=https://huggingface.co/bigcode/octogeex>OctoGeeX</a></th>
483
  <td>CodeGeeX2 (6B parameters) instruction tuned on CommitPackFT + OASST</td>
484
  </tr>
485
  <tr>
486
- <th>Evaluation&nbsp;&nbsp;</t>
487
  <th><a href=https://huggingface.co/datasets/bigcode/humanevalpack>HumanEvalPack</a></th>
488
  <td>Extension of OpenAI's HumanEval to cover 3 scenarios across 6 languages</td>
489
  </tr>
@@ -541,7 +547,7 @@ print(tokenizer.decode(outputs[0]))
541
 
542
  ```bibtex
543
  @article{muennighoff2023octopack,
544
- title={OctoPack: Instruction Tuning Code Large Language Models},
545
  author={Niklas Muennighoff and Qian Liu and Armel Zebaze and Qinkai Zheng and Binyuan Hui and Terry Yue Zhuo and Swayam Singh and Xiangru Tang and Leandro von Werra and Shayne Longpre},
546
  journal={arXiv preprint arXiv:2308.07124},
547
  year={2023}
 
235
  ---
236
 
237
  <!-- header start -->
238
+ <!-- 200823 -->
239
+ <div style="width: auto; margin-left: auto; margin-right: auto">
240
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
241
  </div>
242
  <div style="display: flex; justify-content: space-between; width: 100%;">
243
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
244
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
245
  </div>
246
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
247
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
248
  </div>
249
  </div>
250
+ <div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
251
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
252
  <!-- header end -->
253
 
254
  # Octocoder - GPTQ
 
297
 
298
  | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
299
  | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
300
+ | [main](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/main) | 4 | 128 | No | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 9.20 GB | No | Most compatible option. Good inference speed in AutoGPTQ and GPTQ-for-LLaMa. Lower inference quality than other options. |
301
+ | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 10.09 GB | No | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. Poor AutoGPTQ CUDA speed. |
302
+ | [gptq-4bit-64g-actorder_True](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/gptq-4bit-64g-actorder_True) | 4 | 64 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 9.49 GB | No | 4-bit, with Act Order and group size 64g. Uses less VRAM than 32g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
303
+ | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/Octocoder-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [Evol Instruct Code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1) | 8192 | 9.20 GB | No | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
304
+ | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/Octocoder-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) | 8192 | 16.49 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements and to improve AutoGPTQ speed. |
305
  | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/Octocoder-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) | 8192 | 16.84 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. Poor AutoGPTQ CUDA speed. |
306
 
307
  ## How to download from branches
 
416
  ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
417
 
418
  <!-- footer start -->
419
+ <!-- 200823 -->
420
  ## Discord
421
 
422
  For further support, and discussions on these models and AI in general, join us at:
 
438
 
439
  **Special thanks to**: Aemon Algiz.
440
 
441
+ **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
442
 
443
 
444
  Thank you to all my generous patrons and donaters!
445
 
446
+ And thank you again to a16z for their generous grant.
447
+
448
  <!-- footer end -->
449
 
450
  # Original model card: BigCode's Octocoder
 
469
  - **OctoPack🐙🎒:**
470
  <table>
471
  <tr>
472
+ <th>Data</t>
473
  <th><a href=https://huggingface.co/datasets/bigcode/commitpack>CommitPack</a></th>
474
  <td>4TB of GitHub commits across 350 programming languages</td>
475
  </tr>
476
  <tr>
477
+ <th></t>
478
  <th><a href=https://huggingface.co/datasets/bigcode/commitpackft>CommitPackFT</a></th>
479
  <td>Filtered version of CommitPack for high-quality commit messages that resemble instructions</td>
480
  </tr>
481
  <tr>
482
+ <th>Model</t>
483
  <th><a href=https://huggingface.co/bigcode/octocoder>OctoCoder</a></th>
484
  <td>StarCoder (16B parameters) instruction tuned on CommitPackFT + OASST</td>
485
  </tr>
486
  <tr>
487
+ <th></t>
488
  <th><a href=https://huggingface.co/bigcode/octogeex>OctoGeeX</a></th>
489
  <td>CodeGeeX2 (6B parameters) instruction tuned on CommitPackFT + OASST</td>
490
  </tr>
491
  <tr>
492
+ <th>Evaluation&nbsp;&nbsp;</t>
493
  <th><a href=https://huggingface.co/datasets/bigcode/humanevalpack>HumanEvalPack</a></th>
494
  <td>Extension of OpenAI's HumanEval to cover 3 scenarios across 6 languages</td>
495
  </tr>
 
547
 
548
  ```bibtex
549
  @article{muennighoff2023octopack,
550
+ title={OctoPack: Instruction Tuning Code Large Language Models},
551
  author={Niklas Muennighoff and Qian Liu and Armel Zebaze and Qinkai Zheng and Binyuan Hui and Terry Yue Zhuo and Swayam Singh and Xiangru Tang and Leandro von Werra and Shayne Longpre},
552
  journal={arXiv preprint arXiv:2308.07124},
553
  year={2023}
config.json CHANGED
@@ -1,40 +1,51 @@
1
  {
2
- "_name_or_path": "starcoder-guanaco-commits-35",
3
- "activation_function": "gelu",
4
- "architectures": [
5
- "GPTBigCodeForCausalLM"
6
- ],
7
- "attention_softmax_in_fp32": true,
8
- "attn_pdrop": 0.1,
9
- "bos_token_id": 0,
10
- "embd_pdrop": 0.1,
11
- "eos_token_id": 0,
12
- "inference_runner": 0,
13
- "initializer_range": 0.02,
14
- "layer_norm_epsilon": 1e-05,
15
- "max_batch_size": null,
16
- "max_sequence_length": null,
17
- "model_type": "gpt_bigcode",
18
- "multi_query": true,
19
- "n_embd": 6144,
20
- "n_head": 48,
21
- "n_inner": 24576,
22
- "n_layer": 40,
23
- "n_positions": 8192,
24
- "pad_key_length": true,
25
- "pre_allocate_kv_cache": false,
26
- "resid_pdrop": 0.1,
27
- "scale_attention_softmax_in_fp32": true,
28
- "scale_attn_weights": true,
29
- "summary_activation": null,
30
- "summary_first_dropout": 0.1,
31
- "summary_proj_to_labels": true,
32
- "summary_type": "cls_index",
33
- "summary_use_proj": true,
34
- "torch_dtype": "float32",
35
- "transformers_version": "4.31.0.dev0",
36
- "use_cache": true,
37
- "validate_runner_input": true,
38
- "vocab_size": 49152,
39
- "pretraining_tp": 1
 
 
 
 
 
 
 
 
 
 
 
40
  }
 
1
  {
2
+ "_name_or_path": "starcoder-guanaco-commits-35",
3
+ "activation_function": "gelu",
4
+ "architectures": [
5
+ "GPTBigCodeForCausalLM"
6
+ ],
7
+ "attention_softmax_in_fp32": true,
8
+ "attn_pdrop": 0.1,
9
+ "bos_token_id": 0,
10
+ "embd_pdrop": 0.1,
11
+ "eos_token_id": 0,
12
+ "inference_runner": 0,
13
+ "initializer_range": 0.02,
14
+ "layer_norm_epsilon": 1e-05,
15
+ "max_batch_size": null,
16
+ "max_sequence_length": null,
17
+ "model_type": "gpt_bigcode",
18
+ "multi_query": true,
19
+ "n_embd": 6144,
20
+ "n_head": 48,
21
+ "n_inner": 24576,
22
+ "n_layer": 40,
23
+ "n_positions": 8192,
24
+ "pad_key_length": true,
25
+ "pre_allocate_kv_cache": false,
26
+ "resid_pdrop": 0.1,
27
+ "scale_attention_softmax_in_fp32": true,
28
+ "scale_attn_weights": true,
29
+ "summary_activation": null,
30
+ "summary_first_dropout": 0.1,
31
+ "summary_proj_to_labels": true,
32
+ "summary_type": "cls_index",
33
+ "summary_use_proj": true,
34
+ "torch_dtype": "float32",
35
+ "transformers_version": "4.31.0.dev0",
36
+ "use_cache": true,
37
+ "validate_runner_input": true,
38
+ "vocab_size": 49152,
39
+ "pretraining_tp": 1,
40
+ "quantization_config": {
41
+ "bits": 4,
42
+ "group_size": 128,
43
+ "damp_percent": 0.1,
44
+ "desc_act": false,
45
+ "sym": true,
46
+ "true_sequential": true,
47
+ "model_name_or_path": null,
48
+ "model_file_base_name": "model",
49
+ "quant_method": "gptq"
50
+ }
51
  }
gptq_model-4bit-128g.safetensors → model.safetensors RENAMED
File without changes
quantize_config.json CHANGED
@@ -6,5 +6,5 @@
6
  "sym": true,
7
  "true_sequential": true,
8
  "model_name_or_path": null,
9
- "model_file_base_name": null
10
  }
 
6
  "sym": true,
7
  "true_sequential": true,
8
  "model_name_or_path": null,
9
+ "model_file_base_name": "model"
10
  }