Commit graph

3,104 commits

Author SHA1 Message Date
Daniel Han
e3918e7063 Merge branch 'main' into nightly 2025-12-12 05:53:19 -08:00
Daniel Han
8490f6efc4
Nightly (#3720)
* Update _utils.py

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

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

* [FIX] [Transformers] VLM input embeds fix for gradients (#3715)

* Fix get_input_embeds call for VLMs

* patch input_require_grads instead

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

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

* cleanup old patch

* cleanup old patch

* cleanup

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

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

* Apply suggestion from @danielhanchen

* use logger instead of prints

* Move unsloth present set

* [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>

* Update rope_embedding.py

* Fixes

* Update _utils.py

* Update import_fixes.py

* Update rl_replacements.py

* fix_openenv_no_vllm

* Fix

* Update __init__.py

* Update __init__.py

* Update __init__.py

* Update import_fixes.py

* Update import_fixes.py

* Update import_fixes.py

* logger

* Update __init__.py

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

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

* Update __init__.py

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Datta Nimmaturi <venkatadattasainimmaturi@gmail.com>
2025-12-12 05:53:08 -08:00
Daniel Han
12ccc47ed7 Merge branch 'nightly' of https://github.com/unslothai/unsloth into nightly 2025-12-12 05:51:51 -08:00
Daniel Han
efb5801a74 Update __init__.py 2025-12-12 05:51:31 -08:00
pre-commit-ci[bot]
b5b57b3378 [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
2025-12-12 13:48:24 +00:00
Daniel Han
0c9288f311 Update __init__.py 2025-12-12 05:46:50 -08:00
Daniel Han
32b52a0002 logger 2025-12-12 05:44:38 -08:00
Daniel Han
04ad21cdb0 Update import_fixes.py 2025-12-12 05:40:56 -08:00
Daniel Han
f34eb0abb6 Update import_fixes.py 2025-12-12 05:38:24 -08:00
Daniel Han
e17b62f76e Update import_fixes.py 2025-12-12 05:36:40 -08:00
Daniel Han
f4f2a7f907 Update __init__.py 2025-12-12 05:34:29 -08:00
Daniel Han
fb763f34ec Update __init__.py 2025-12-12 05:31:36 -08:00
Daniel Han
680f19f156 Update __init__.py 2025-12-12 05:29:25 -08:00
Daniel Han
9a9813942d Fix 2025-12-12 05:27:42 -08:00
Daniel Han
ac54d6e1c8 fix_openenv_no_vllm 2025-12-12 05:20:09 -08:00
Daniel Han
696a540c1e Update rl_replacements.py 2025-12-12 05:11:12 -08:00
Daniel Han
01319d3681 Update import_fixes.py 2025-12-12 05:10:45 -08:00
Daniel Han
c94f59513f Update _utils.py 2025-12-12 05:01:43 -08:00
Daniel Han
b5f1a77482 Fixes 2025-12-12 04:58:43 -08:00
Daniel Han
449430db47 Merge branch 'main' into nightly 2025-12-12 04:07:32 -08:00
Scott Roy
6a3b5a22b3
Update torchao save (#3679)
* Update torchao save

* up

* up

* up

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

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

* Apply suggestion from @danielhanchen

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
2025-12-12 04:07:02 -08:00
Daniel Han
2c22ce662f Update rope_embedding.py 2025-12-12 03:41:09 -08:00
Daniel Han
a63a33762f Merge branch 'main' into nightly 2025-12-12 03:38:09 -08:00
Datta Nimmaturi
cde2d42caf
[FIX] [Transformers] VLM input embeds fix for gradients (#3715)
* Fix get_input_embeds call for VLMs

* patch input_require_grads instead

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

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

* cleanup old patch

* cleanup old patch

* cleanup

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

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

* Apply suggestion from @danielhanchen

* use logger instead of prints

* Move unsloth present set

* [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>
2025-12-12 03:33:39 -08:00
Dan Saunders
345f5a5eb4
Mistral packing, train on completions only, simplifications (#3709)
* pipe kwargs through mistral model

* simplify / bugfix

* bugfix for train_on_completions_only

* wire up is_unsupported_model

* nits, edge cases
2025-12-10 23:15:59 -08:00
Lei Zhenyuan
e2f66beb48
[intel] skip xpu fbgemm fp8 (#3625)
* skip xpu fbgemm fp8

* Apply suggestion from @gemini-code-assist[bot]

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

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

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

* Apply suggestion from @danielhanchen

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
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 21:13:29 -08:00
Michael Han
947eda3ad6
Padding free packing update 2025-12-10 21:12:13 -08:00
Michael Han
173be4ff77
Adding new padding free packing support 2025-12-10 21:10:19 -08:00
Dan Saunders
aa063de198
update TRL filter (#3707)
* update TRL filter

* both filters

* Apply suggestion from @danielhanchen

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

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

---------

Co-authored-by: Daniel Han <danielhanchen@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-12-10 07:49:52 -08:00
Daniel Han
f357fc58cd Merge branch 'main' into nightly 2025-12-10 06:25:56 -08:00
Daniel Han
1f17f4a984 Gemma issue 2025-12-10 06:25:49 -08:00
Daniel Han
30b78c5ca3 Merge branch 'main' into nightly 2025-12-10 06:11:32 -08:00
Daniel Han
2163100aaa Update _utils.py 2025-12-10 06:11:14 -08:00
Daniel Han
7840005526 Update trainer.py 2025-12-10 06:11:03 -08:00
Daniel Han
94bbcaa504 Merge branch 'main' into nightly 2025-12-10 04:15:57 -08:00
Daniel Han
335f9cc095
Nightly (#3706)
* Update _utils.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>
2025-12-10 04:13:13 -08:00
pre-commit-ci[bot]
30ade52416 [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
2025-12-10 12:12:49 +00:00
Daniel Han
f754bd2b50 Merge branch 'main' into nightly 2025-12-10 04:12:19 -08:00
Daniel Han
ff78e784fe Update import_fixes.py 2025-12-10 04:05:14 -08:00
Daniel Han
d7480d310d Update _utils.py 2025-12-10 03:48:01 -08:00
Datta Nimmaturi
3f92868da4
[FIX] fbgemm version check (#3704)
* fbgemm version check

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

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

* safer version check

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

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

* Add check for torchvision-torch compatibility

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

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

* refactor package check logic

* Remove logs and enforce torch

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2025-12-10 03:46:30 -08:00
Dan Saunders
35606da943
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
Daniel Han
6b908cfae7 Merge branch 'main' into nightly 2025-12-09 17:37:18 -08:00
Dan Saunders
50325e0be7
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
25a6250581 Merge branch 'main' into nightly 2025-12-09 03:31:30 -08:00
Daniel Han
984bed300f
Update _utils.py (#3698) 2025-12-09 03:31:20 -08:00
Daniel Han
150eadfb8d Merge branch 'main' into nightly 2025-12-09 03:30:42 -08:00
Datta Nimmaturi
d85e359c45
[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
Daniel Han
e53e1852a8 Update _utils.py 2025-12-09 01:02:26 -08:00
Datta Nimmaturi
6324476a76
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