Commit graph

3,059 commits

Author SHA1 Message Date
Dan Saunders
89b042f23b Auto-enable padding-free SFT (#3672)
* implement (sdpa, xformers, fa2) sample packing

* attention dispatching

* ddp working OOTB with CLI

* packed SWA and softcap support

* enable batch flattening

* LGPL license headers

* mask packed sequence boundaries

* auto-enable sample packing

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add explicit toggle for sample packing

* Add explicit toggle for sample packing

* Update __init__.py

* Update unsloth/kernels/rope_embedding.py

* Update unsloth/kernels/rope_embedding.py

* remove grad output clones; restore deleted FastLanguageModel arg

* fix

* restore rope embedding clones

* xformers mask cache

* implement (sdpa, xformers, fa2) sample packing

* attention dispatching

* ddp working OOTB with CLI

* packed SWA and softcap support

* enable batch flattening

* LGPL license headers

* mask packed sequence boundaries

* auto-enable sample packing

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add explicit toggle for sample packing

* Add explicit toggle for sample packing

* Update __init__.py

* Update unsloth/kernels/rope_embedding.py

* Update unsloth/kernels/rope_embedding.py

* remove grad output clones; restore deleted FastLanguageModel arg

* fix

* restore rope embedding clones

* xformers mask cache

* add back accidental deletion

* Update unsloth/kernels/rope_embedding.py

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

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix merge conflicts

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add **kwargs

* add back clobbered

* Update rope_embedding.py

* Update rope_embedding.py

* simplify trl warnings filter

* docstring

* nit

* bugfix

* add padding-free seqlen metadata

* auto-enable padding free

* gemma2 disable

* Apply suggestion from @danielhanchen

* Update trainer.py

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Update trainer.py

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
2025-12-10 03:07:29 -08:00
Dan Saunders
538d2c0719 SFT sample packing (#3566)
* implement (sdpa, xformers, fa2) sample packing

* attention dispatching

* ddp working OOTB with CLI

* packed SWA and softcap support

* enable batch flattening

* LGPL license headers

* mask packed sequence boundaries

* auto-enable sample packing

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add explicit toggle for sample packing

* Add explicit toggle for sample packing

* Update __init__.py

* Update unsloth/kernels/rope_embedding.py

* Update unsloth/kernels/rope_embedding.py

* remove grad output clones; restore deleted FastLanguageModel arg

* fix

* restore rope embedding clones

* xformers mask cache

* implement (sdpa, xformers, fa2) sample packing

* attention dispatching

* ddp working OOTB with CLI

* packed SWA and softcap support

* enable batch flattening

* LGPL license headers

* mask packed sequence boundaries

* auto-enable sample packing

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add explicit toggle for sample packing

* Add explicit toggle for sample packing

* Update __init__.py

* Update unsloth/kernels/rope_embedding.py

* Update unsloth/kernels/rope_embedding.py

* remove grad output clones; restore deleted FastLanguageModel arg

* fix

* restore rope embedding clones

* xformers mask cache

* add back accidental deletion

* Update unsloth/kernels/rope_embedding.py

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

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix merge conflicts

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add **kwargs

* add back clobbered

* Update rope_embedding.py

* Update rope_embedding.py

* simplify trl warnings filter

* docstring

* nit

* bugfix

* Apply suggestion from @danielhanchen

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Update unsloth/trainer.py

* Update unsloth/trainer.py

* Update unsloth/trainer.py

* Update unsloth/trainer.py

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
2025-12-09 17:36:45 -08:00
Daniel Han
1b70a5f15e Update _utils.py (#3698) 2025-12-09 03:31:20 -08:00
Datta Nimmaturi
c8c8f168bb [Fix] [TRL] load_lora for multi line llm.chat/generate (#3696)
* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Remove reload_weights rpc call from grpo trainer

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Use regex instead of static string

* patch openenv reload_weights call

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Better handle sleep and wakeup

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Reset indentation

* Handle multi line self.llm.chat better

* Use logger

* re-indent

* Stricter regex to replace wildcard

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
2025-12-09 03:30:23 -08:00
Datta Nimmaturi
b414e43b74 Remove reload_weights rpc call from grpo trainer (#3673)
* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Remove reload_weights rpc call from grpo trainer

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Use regex instead of static string

* patch openenv reload_weights call

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Better handle sleep and wakeup

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Reset indentation

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
2025-12-08 23:36:22 -08:00
pre-commit-ci[bot]
cde46b2054 [pre-commit.ci] pre-commit autoupdate (#3694)
updates:
- [github.com/astral-sh/ruff-pre-commit: v0.14.7 → v0.14.8](https://github.com/astral-sh/ruff-pre-commit/compare/v0.14.7...v0.14.8)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-12-08 19:44:56 -08:00
Daniel Han
609357d8aa Versioning 2025-12-08 04:19:10 -08:00
Daniel Han
28c058d1b8 Update pyproject.toml 2025-12-08 04:13:45 -08:00
Daniel Han
8447c6d9bc Versioning 2025-12-08 04:06:01 -08:00
Noah Kirschmann
8618876411 Update transformers version constraint in pyproject.toml (#3689)
* Update transformers version constraint in pyproject.toml

The latest transformers version just fixes the local training.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Update transformers version constraint in pyproject.toml

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-12-08 03:27:18 -08:00
Daniel Han
16ea28daad Add **kwargs 2025-12-08 03:24:51 -08:00
Daniel Han
c8d0134b7d Update rl.py 2025-12-08 02:23:43 -08:00
Daniel Han
e0b36af542 Update vision.py 2025-12-07 23:09:13 -08:00
Daniel Han
f687d7afc2 Update _utils.py 2025-12-07 16:52:59 -08:00
Daniel Han
66e07ede3b Xformers fix 2025-12-07 16:40:51 -08:00
Michael Han
907264d077 Update README.md 2025-12-04 08:21:20 -08:00
Daniel Han
6db8cee887 Update README.md 2025-12-02 04:08:54 -08:00
Daniel Han
4f30c9104f Update README.md 2025-12-02 03:52:50 -08:00
pre-commit-ci[bot]
24adac2296 [pre-commit.ci] pre-commit autoupdate (#3666)
updates:
- [github.com/astral-sh/ruff-pre-commit: v0.14.6 → v0.14.7](https://github.com/astral-sh/ruff-pre-commit/compare/v0.14.6...v0.14.7)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-12-01 17:45:13 -08:00
Daniel Han
53e6bd1568 Update _utils.py 2025-12-01 08:01:05 -08:00
Daniel Han
d33a5b909c Update rl.py 2025-12-01 08:00:08 -08:00
Daniel Han
a00ec814e5 Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"
This reverts commit 3beade641d.
2025-12-01 07:24:58 -08:00
pre-commit-ci[bot]
3beade641d [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
2025-12-01 15:24:34 +00:00
Daniel Han
0613fafefe Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"
This reverts commit 727da805d9.
2025-12-01 07:24:21 -08:00
pre-commit-ci[bot]
727da805d9 [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
2025-12-01 15:23:44 +00:00
Daniel Han
1bf2af5165 Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"
This reverts commit 3722c95e42.
2025-12-01 07:23:31 -08:00
pre-commit-ci[bot]
3722c95e42 [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
2025-12-01 15:20:22 +00:00
Daniel Han
9ab2d23233 Update unsloth/models/rl.py
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-12-01 07:20:00 -08:00
Daniel Han
e38209de83 Update unsloth/models/rl.py
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-12-01 07:19:50 -08:00
Daniel Han
89fc908d85 Update qwen3_moe.py 2025-12-01 07:19:07 -08:00
Datta Nimmaturi
2bc0765a2d Vllm guided decoding (#3663)
* vllm sampling params fix

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* do not patch base_trainer

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* seperate vllm fixes

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Fixup deletion

* Fix indentation

* revert to old style

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-12-01 07:11:28 -08:00
Daniel Han
d15b9b8acb Verisoning 2025-12-01 07:09:17 -08:00
Daniel Han
b80f2761ac Update rl.py 2025-12-01 06:23:23 -08:00
Daniel Han
5f0c9cff5e Revert "[FIX] Vllm guided decoding params (#3662)"
This reverts commit e4d566e159.
2025-12-01 05:43:45 -08:00
Datta Nimmaturi
e4d566e159 [FIX] Vllm guided decoding params (#3662)
* vllm sampling params fix

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* do not patch base_trainer

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* seperate vllm fixes

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Apply suggestion from @danielhanchen

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"

This reverts commit fe334c4dffda0c799295a69b3fc5f4c071d83176.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"

This reverts commit ae94be7d7e0a4780820adfbcde3549aec0ec49cf.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"

This reverts commit cde2dde170d35bce2b50211eeaf4c9788f33734c.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
2025-12-01 05:42:37 -08:00
Daniel Han
64bd3e552f Merge branch 'main' into nightly 2025-12-01 04:21:27 -08:00
Santosh Bhavani
6abd9afe99 Fix: Pass gradient_checkpointing parameter to model.for_training() calls (#3659) 2025-12-01 04:18:41 -08:00
Daniel Han
08a399092f Update vision.py 2025-12-01 01:21:26 -08:00
Daniel Han
a40a7b03fb Typos 2025-12-01 00:01:07 -08:00
Daniel Han
3ef6b66f39 Update qwen3_moe.py 2025-11-30 23:37:32 -08:00
Daniel Han
f691b2616e Update vision.py 2025-11-30 21:32:07 -08:00
VED
5d0523c1ca set defualt [128, 128] insted of none (#3658)
Co-authored-by: Ved <ved.work2024@gmail.com>
2025-11-30 17:00:31 -08:00
Daniel Han
ad3221a7fd Update rl.py 2025-11-30 04:40:03 -08:00
Daniel Han
a40189fa4d Merge branch 'main' into nightly 2025-11-30 04:39:55 -08:00
DoubleMathew
69f7dfdb90 make unsloth_tiled_mlp a from_pretrained arg (#3655)
* make unsloth_tiled_mlp a from_pretrained arg

* adjust patching logic
2025-11-29 22:47:51 -08:00
Bhuvan Prakash
27ae5c335c Fix: prevent load_in_fp8 kwarg from reaching Qwen3MoeForCausalLM constructor (Fix #3649) (#3654)
* Fix: remove load_in_fp8 from kwargs to prevent Qwen3Moe init TypeError (Fix #3649)

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-11-29 20:18:11 -08:00
gitpullpull
d8ea261924 Fix broken link for Advanced pip install instructions (#3652) 2025-11-29 15:33:48 -08:00
Michael Han
f730ad0b3e Update README.md 2025-11-29 08:01:00 -08:00
DoubleMathew
f3803cdee0 fix rope_theta -> rope_parameters['rope_theta'] (#3651) 2025-11-29 06:44:26 -08:00
Michael Han
2769e27566 Update README.md 2025-11-27 20:52:27 -08:00