transformers
Fix step shifting when accumulate gradient
#33673
Merged

Fix step shifting when accumulate gradient #33673

kibitzing
kibitzing replace total_batched_samples with step while counting grad accum step
6720364a
kibitzing remove unused variable
59273e69
SunMarc SunMarc requested a review from SunMarc SunMarc 1 year ago
HuggingFaceDocBuilderDev
SunMarc
SunMarc commented on 2024-09-26
kibitzing simplify condition for update step
81215c0c
kibitzing
kibitzing fix format by ruff
e4cc3609
kibitzing
SunMarc
kibitzing
SunMarc
kibitzing
SunMarc
kibitzing kibitzing force pushed to e4cc3609 1 year ago
kibitzing
kibitzing
kibitzing simplify update step condition using accelerator.sync_gradients
8348ff4e
kibitzing
kibitzing
muellerzr
muellerzr approved these changes on 2024-10-24
muellerzr muellerzr requested a review from SunMarc SunMarc 1 year ago
muellerzr
SunMarc
SunMarc commented on 2024-10-24
kibitzing
kibitzing commented on 2024-10-25
kibitzing Merge branch 'main' into fix-step-shifting-when-accum-grad
ffd4c83e
kibitzing simplify update condition using do_sync_step
ca715e78
kibitzing Merge branch 'main' into fix-step-shifting-when-accum-grad
53c2ed84
kibitzing remove print for test
8f210263
kibitzing
SunMarc
SunMarc approved these changes on 2024-10-25
muellerzr Merge branch 'main' into fix-step-shifting-when-accum-grad
a69efb02
muellerzr
muellerzr Merge branch 'main' into fix-step-shifting-when-accum-grad
0333ba08
muellerzr
kibitzing Merge branch 'main' into fix-step-shifting-when-accum-grad
041971e7
kibitzing
LysandreJik
LysandreJik approved these changes on 2024-10-31
muellerzr muellerzr merged dca93ca0 into main 1 year ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone