ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
2,348 commits 391 branches 58 tags 304 MiB
Commit graph

2,348 commits

This branch
This branch
All branches
Author SHA1 Message Date
Daniel Han
b087bfaf55 Update rl.py 2025-05-28 12:27:55 -07:00
Daniel Han
b290557814 Update rl.py 2025-05-28 12:27:43 -07:00
Daniel Han
acb972cee0 versioning 2025-05-28 12:19:43 -07:00
Daniel Han
85b959b9bb Merge branch 'main' into nightly 2025-05-28 12:11:41 -07:00
DoubleMathew
95452eed81 Fix SFTtraining for new trl (#2647) 
* fix sft training with trl>0.15.2 with trl DataCollator

* Update fix to accomodate both trl and transformers DataCollatorForLanguageModeling
 2025-05-28 11:55:48 -07:00
Daniel Han
2b1462e813 Merge branch 'main' into nightly 2025-05-28 06:15:34 -07:00
Daniel Han
623060ba29 Latest TRL, GRPO + Bug fixes (#2645) 
* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* model_type_arch

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* check

* Update _utils.py

* Update loader.py

* Update loader.py

* Remove prints

* Update README.md

typo

* Update _utils.py

* Update _utils.py

* versioning

* Update _utils.py

* Update _utils.py

* Update _utils.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update vision.py

* HF Transfer

* fix(utils): add missing importlib import to fix NameError (#2134)

This commit fixes a NameError that occurs when `importlib` is referenced in _utils.py
without being imported, especially when UNSLOTH_USE_MODELSCOPE=1 is enabled.
By adding the missing import statement, the code will no longer throw a NameError.

* Add QLoRA Train and Merge16bit Test (#2130)

* add reference and unsloth lora merging tests

* add test / dataset printing to test scripts

* allow running tests from repo root

* add qlora test readme

* more readme edits

* ruff formatting

* additional readme comments

* forgot to add actual tests

* add apache license

* Update pyproject.toml

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* Update loader.py

* Revert

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Bug fix

* Update mapper.py

* check SDPA for Mistral 3, Pixtral

* Update vision.py

* Versioning

* Update rl_replacements.py

* Update README.md

* add model registry

* move hf hub utils to unsloth/utils

* refactor global model info dicts to dataclasses

* fix dataclass init

* fix llama registration

* remove deprecated key function

* start registry reog

* add llama vision

* quant types -> Enum

* remap literal quant types to QuantType Enum

* add llama model registration

* fix quant tag mapping

* add qwen2.5 models to registry

* add option to include original model in registry

* handle quant types per model size

* separate registration of base and instruct llama3.2

* add QwenQVQ to registry

* add gemma3 to registry

* add phi

* add deepseek v3

* add deepseek r1 base

* add deepseek r1 zero

* add deepseek distill llama

* add deepseek distill models

* remove redundant code when constructing model names

* add mistral small to registry

* rename model registration methods

* rename deepseek registration methods

* refactor naming for mistral and phi

* add global register models

* refactor model registration tests for new registry apis

* add model search method

* remove deprecated registration api

* add quant type test

* add registry readme

* make llama registration more specific

* clear registry when executing individual model registration file

* more registry readme updates

* Update _auto_install.py

* Llama4

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Synthetic data

* Update mapper.py

* Xet and Synthetic

* Update synthetic.py

* Update loader.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update pyproject.toml

* Delete .gitignore

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update _utils.py

* Update pyproject.toml

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update chat_templates.py

* Seasame force float16 / float32

* Fix Seasame

* Update loader.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* is_multimodal

* Update loader.py

* Update loader.py

* Update loader.py

* Update loader.py

* Update vision.py

* Update vision.py

* Update vision.py

* UNSLOTH_DISABLE_STATIC_GENERATION

* Update vision.py

* Auto vision detection

* Sesame

* Whisper

* Update loader.py

* Update loader.py

* Update loader.py

* Update mapper.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* Update loader.py

* Update loader.py

* Update loader.py

* Update _utils.py

* Update rl.py

* versioning

* Update rl.py

* Update rl.py

* Update rl.py

* Update rl.py

* Update rl.py

* logging

* Update pyproject.toml

* Update rl.py

---------

Co-authored-by: Jack Shi Wei Lun <87535974+jackswl@users.noreply.github.com>
Co-authored-by: naliazheli <nalia0316@gmail.com>
Co-authored-by: jeromeku <jerome.ku@gmail.com>
Co-authored-by: Michael Han <107991372+shimmyshimmer@users.noreply.github.com>
 2025-05-28 06:15:12 -07:00
Daniel Han
251066289d Update rl.py 2025-05-28 06:04:23 -07:00
Daniel Han
12912022f4 Update pyproject.toml 2025-05-28 05:58:02 -07:00
Daniel Han
c94e31888c logging 2025-05-28 05:56:19 -07:00
Daniel Han
725d84997b Update rl.py 2025-05-28 05:26:01 -07:00
Daniel Han
1dabfcd2d3 Update rl.py 2025-05-28 05:23:07 -07:00
Daniel Han
86cd1d2786 Update rl.py 2025-05-28 05:09:03 -07:00
Daniel Han
e7f76d53bb Update rl.py 2025-05-28 05:06:24 -07:00
Daniel Han
4fb4d3a36c Update rl.py 2025-05-28 05:04:12 -07:00
Daniel Han
dc48358b2f versioning 2025-05-28 05:02:05 -07:00
Daniel Han
89c30967df Update rl.py 2025-05-28 04:57:30 -07:00
Daniel Han
026ba8e678 Create LICENSE 2025-05-28 03:27:48 -07:00
jeromeku
0b5ac8f2ab Llama4 MoE Grouped GEMM (#2639) 
* add llama4 reference layer

* add llama4 reference impl

* formatting
 2025-05-28 03:26:35 -07:00
Daniel Han
86ecf655a9 Merge branch 'main' into nightly 2025-05-28 03:24:15 -07:00
Premik
d8bf17959a Check the skip_prepare_dataset before accessing dataset fields. #2496 (#2633) 2025-05-28 03:23:59 -07:00
Daniel Han
0ba75a4359 Merge branch 'main' into nightly 2025-05-28 02:06:42 -07:00
Michael Han
ce9e54755f Update README.md 
Better Qwen3 notebook
 2025-05-26 23:44:41 -07:00
Daniel Han
5327d1d36d Flash Attention whls 2025-05-26 22:48:46 -07:00
Datta Nimmaturi
811422bbb4 Upgrade trl fix (#2544) 
* Update llama.py making set and reset functions in order to properly use autoSequenceClassification

* Update fast_lora.py, added mixed precising pytorch autocasting

* Update llama.py did not included rotary embeddings in the reset functions correctly

* Update rl.py: correct get reward model added as well as the eval step stuff

* Update rl.py removed function that did not need to be patched

* Update llama.py: kept reset functions and made their names generic

* Update fast_lora.py

* Update rl.py, try except

* Update fast_lora.py, removing downcasting stuff

* Update llama.py removed depircate LLamaLinearScalingRotaryEmbedding

* Update rl.py for VLLM RLOO and PPO

* Update rl.py reverted

* Update rl.py with peft cahnges

* Update rl.py, disabling adapters screws inference up

* Update rl.py getting PPO support

* Update rl.py cleanup

* Update rl.py cleaned up not useful commented code

* Update llama.py, enabled new flag, keep padding

* Upgrade trl fix

Signed-off-by: Dattu Sharma <venkatadattasainimmaturi@gmail.com>

* Update rl.py made changes relative to the review

* Revert accidental patch block for non grpo

Signed-off-by: Dattu Sharma <venkatadattasainimmaturi@gmail.com>

* Fixup sampling params issue

* Fix rl.py regex

Signed-off-by: Dattu Sharma <venkatadattasainimmaturi@gmail.com>

* loss type: grpo, drgrpo and bnpo

Signed-off-by: Dattu Sharma <venkatadattasainimmaturi@gmail.com>

* Add trl version check for vllm colocate mode for RL trainers

* Update rl.py

For TRL 0.18.0 (Main branch of TRL at the time because its on 0.17.0) , the SFT trainer for some reason deletes the labels column and unsloth internal loss funcitons need that column for hte claculations so I add it back in like this.

* Update llama.py, merge it to be dattas llama version

* Update rl.py, sft changes to get 0.18.0 to be working

* Update rl_replacements.py, added hidden state stuff

* Update rl_replacements.py

* Update rl_replacements.py

* Update rl_replacements.py, rechanged the accumlated loss

* Fixup num_iterations>1 for grpo

Signed-off-by: datta0 <venkatadattasainimmaturi@gmail.com>

* Update rl_replacements.py

* no unnecessary logits upcast. fix naming

Signed-off-by: datta0 <venkatadattasainimmaturi@gmail.com>

* Update rl_replacements.py returned hidden states from logprobs

* Update rl_replacements.py removed debug logic

* Update rl_replacements.py, should be fine now

* Update rl_replacements.py, should take new args for GRPO trainer

* Update rl_replacements.py, made it compatible with trl 0.15.2

* Update rl_replacements.py, fixed typo in per tokne-Logps

---------

Signed-off-by: Dattu Sharma <venkatadattasainimmaturi@gmail.com>
Signed-off-by: datta0 <venkatadattasainimmaturi@gmail.com>
Co-authored-by: pluesclues <136766175+pluesclues@users.noreply.github.com>
 2025-05-26 17:20:57 -07:00
Daniel Han
a66f3f4cda Colocate vLLM 2025-05-26 00:37:04 -07:00
Michael Han
1f4e74cb96 Update README.md 2025-05-25 03:35:43 -07:00
Quentin Gallouédec
ce5c2d2145 Remove dataset_text_field from SFTConfig (#2609) 2025-05-25 03:20:16 -07:00
Richi
f6c4be39b7 add: path checking for failed llama cpp builds (#2603) 2025-05-25 03:18:07 -07:00
Daniel Han
9b8e6b1e22 Merge branch 'main' into nightly 2025-05-21 23:21:12 -07:00
Daniel Han
dd43200718 Devstral, MedGemma 2025-05-21 07:35:36 -07:00
Michael Han
e771760e53 Update README.md 
Updating model support
 2025-05-20 09:51:55 -07:00
Michael Han
61b68725e9 Update README.md 2025-05-19 21:26:19 -07:00
Daniel Han
b5269556c2 Update issue templates 2025-05-17 18:30:17 -07:00
Daniel Han
25c11cf2d8 Update issue templates 2025-05-17 18:29:27 -07:00
Daniel Han
7a0c8c7da1 Update issue templates 2025-05-17 05:42:10 -07:00
Daniel Han
4d5f2172f4 Fix Whisper, ModernBERT (#2565) 
* Update vision.py

* Update vision.py

* Update mapper.py

* Update vision.py

* fix: config.torch_dtype in LlamaModel_fast_forward_inference (#2091)

* fix: config.torch_dtype in LlamaModel_fast_forward_inference

* Update llama.py

* update for consistency

---------

Co-authored-by: Daniel Han <danielhanchen@gmail.com>

* versioning

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* model_type_arch

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* check

* Update _utils.py

* Update loader.py

* Update loader.py

* Remove prints

* Update README.md

typo

* Update _utils.py

* Update _utils.py

* versioning

* Update _utils.py

* Update _utils.py

* Update _utils.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update llama.py

* Update vision.py

* HF Transfer

* fix(utils): add missing importlib import to fix NameError (#2134)

This commit fixes a NameError that occurs when `importlib` is referenced in _utils.py
without being imported, especially when UNSLOTH_USE_MODELSCOPE=1 is enabled.
By adding the missing import statement, the code will no longer throw a NameError.

* Add QLoRA Train and Merge16bit Test (#2130)

* add reference and unsloth lora merging tests

* add test / dataset printing to test scripts

* allow running tests from repo root

* add qlora test readme

* more readme edits

* ruff formatting

* additional readme comments

* forgot to add actual tests

* add apache license

* Update pyproject.toml

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* Update loader.py

* Revert

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Bug fix

* Update mapper.py

* check SDPA for Mistral 3, Pixtral

* Update vision.py

* Versioning

* Update rl_replacements.py

* Update README.md

* add model registry

* move hf hub utils to unsloth/utils

* refactor global model info dicts to dataclasses

* fix dataclass init

* fix llama registration

* remove deprecated key function

* start registry reog

* add llama vision

* quant types -> Enum

* remap literal quant types to QuantType Enum

* add llama model registration

* fix quant tag mapping

* add qwen2.5 models to registry

* add option to include original model in registry

* handle quant types per model size

* separate registration of base and instruct llama3.2

* add QwenQVQ to registry

* add gemma3 to registry

* add phi

* add deepseek v3

* add deepseek r1 base

* add deepseek r1 zero

* add deepseek distill llama

* add deepseek distill models

* remove redundant code when constructing model names

* add mistral small to registry

* rename model registration methods

* rename deepseek registration methods

* refactor naming for mistral and phi

* add global register models

* refactor model registration tests for new registry apis

* add model search method

* remove deprecated registration api

* add quant type test

* add registry readme

* make llama registration more specific

* clear registry when executing individual model registration file

* more registry readme updates

* Update _auto_install.py

* Llama4

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Synthetic data

* Update mapper.py

* Xet and Synthetic

* Update synthetic.py

* Update loader.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update pyproject.toml

* Delete .gitignore

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update _utils.py

* Update pyproject.toml

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update synthetic.py

* Update chat_templates.py

* Seasame force float16 / float32

* Fix Seasame

* Update loader.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* is_multimodal

* Update loader.py

* Update loader.py

* Update loader.py

* Update loader.py

* Update vision.py

* Update vision.py

* Update vision.py

* UNSLOTH_DISABLE_STATIC_GENERATION

* Update vision.py

* Auto vision detection

* Sesame

* Whisper

* Update loader.py

* Update loader.py

* Update loader.py

* Update mapper.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update vision.py

* Update loader.py

* Update loader.py

* Update loader.py

* Update loader.py

* Update _utils.py

---------

Co-authored-by: lurf21 <93976703+lurf21@users.noreply.github.com>
Co-authored-by: Jack Shi Wei Lun <87535974+jackswl@users.noreply.github.com>
Co-authored-by: naliazheli <nalia0316@gmail.com>
Co-authored-by: jeromeku <jerome.ku@gmail.com>
Co-authored-by: Michael Han <107991372+shimmyshimmer@users.noreply.github.com>
 2025-05-17 05:11:50 -07:00
Daniel Han
416b89e901 Update _utils.py 2025-05-17 05:11:36 -07:00
Daniel Han
f0bf761832 Merge branch 'main' into nightly 2025-05-17 05:10:00 -07:00
Emmanuel Ferdman
2a1caa746b Display the model name in RoPE scaling unsupported error (#2564) 
Signed-off-by: Emmanuel Ferdman <emmanuelferdman@gmail.com>
 2025-05-17 05:09:06 -07:00
Daniel Han
382406a9ef Update loader.py 2025-05-17 00:29:55 -07:00
Daniel Han
87b98cf970 Update loader.py 2025-05-17 00:29:04 -07:00
Daniel Han
811a86007d Update loader.py 2025-05-17 00:24:27 -07:00
Daniel Han
611463b428 Update loader.py 2025-05-17 00:03:20 -07:00
Daniel Han
74d44dc611 Update vision.py 2025-05-16 23:31:15 -07:00
Daniel Han
c0ae3602e2 Update vision.py 2025-05-16 23:25:32 -07:00
Daniel Han
fe3f1c02f9 Update vision.py 2025-05-16 23:24:32 -07:00
Daniel Han
18489d1c99 Update vision.py 2025-05-16 23:20:57 -07:00
Daniel Han
b790b82e42 Update vision.py 2025-05-16 23:20:18 -07:00
Daniel Han
3c475c834d Update vision.py 2025-05-16 23:03:04 -07:00