transformers
[Trainer] use output.loss when using liger-kernel
#42444
Merged

[Trainer] use output.loss when using liger-kernel #42444

SunMarc merged 6 commits into main from issue-42414
kashif
kashif use output.loss when using liger
e6e8ec06
kashif kashif requested a review from Rocketknight1 Rocketknight1 211 days ago
kashif Clarify Liger-kernel loss computation in comments
9fe9da14
HuggingFaceDocBuilderDev
kashif Merge branch 'main' into issue-42414
160b320d
kashif kashif requested a review from SunMarc SunMarc 211 days ago
SunMarc
SunMarc commented on 2025-11-27
kashif Both standard transformers and Liger models handle shift_labels corre…
10b4ae85
kashif removed unused shift_labels reference in loss computation
4f0c18aa
kashif Remove unused model unwrapping
9a5654f7
kashif
kashif kashif added bug
SunMarc
SunMarc approved these changes on 2025-11-28
SunMarc SunMarc merged 6db43321 into main 210 days ago
SunMarc SunMarc deleted the issue-42414 branch 210 days ago
zhangwj618
SunMarc
zhangwj618
SunMarc

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone