transformers
fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel
#45074

Merged

fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel #45074

Rocketknight1 merged 4 commits into huggingface:main from harshaljanjani:fix/switch-transformers-timm-wrapper-bf16-dtype

fix: Cast inputs to match weight dtype

cd1a4c94

new: Add test

e05bd6ad

harshaljanjani marked this pull request as ready for review 5 days ago

github-actions requested a review from

ArthurZucker 5 days ago

github-actions requested a review from

ydshieh 5 days ago

Merge branch 'main' into fix/switch-transformers-timm-wrapper-bf16-dtype

32817246

Rocketknight1 commented on 2026-03-30

change: Upcast to float32 instead of downcasting

faa66b3b

harshaljanjani requested a review from

Rocketknight1 3 days ago

Rocketknight1 approved these changes on 2026-04-02

Rocketknight1 merged 423f2a31 into main 1 hour ago

harshaljanjani deleted the fix/switch-transformers-timm-wrapper-bf16-dtype branch 1 hour ago

Reviewers

Rocketknight1

ArthurZucker

ydshieh

Assignees

No one assigned

Labels

None yet

Milestone

No milestone