transformers
Keep relevant weights in fp32 when `model._keep_in_fp32_modules` is set even when `accelerate` is not installed
#26225
Merged
Merged
Keep relevant weights in fp32 when `model._keep_in_fp32_modules` is set even when `accelerate` is not installed #26225
fxmarty
merged
4 commits into
huggingface:main
from
fxmarty:fix-keep-fp32-without-accelerate
fix bug where weight would not be kept in fp32
1eb9eed9
nit
69d6efd2
younesbelkada
approved these changes
on 2023-09-20
address review comments
28a67971
ArthurZucker
approved these changes
on 2023-09-21
younesbelkada
commented
on 2023-09-21
fix test
f326a733
younesbelkada
approved these changes
on 2023-09-21
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub