DeepSpeed
Fix gradient checkpointing with use_reentrant=True / PyTorch-style backward / ZeRO-3
#7780
Merged
Merged
Fix gradient checkpointing with use_reentrant=True / PyTorch-style backward / ZeRO-3 #7780
tohtana
merged
15 commits into
deepspeedai:master
from
tohtana:tohtana/backward_with_reentrant
fix backward with checkpointing and reentrant
a09084b0
Update README with newer status badges for CI
8211447d
Add timeout to test workflows (#7774)
f6026d19
Remove cron/PR triggers for outdated V100 tests (#7777)
3cf426cd
fix yapf formatting in test file
9116f4a9
Merge branch 'master' into tohtana/backward_with_reentrant
cc00bd9b
added sync in tests
f61abc90
Merge branch 'master' into tohtana/backward_with_reentrant
0c994459
loadams
approved these changes
on 2026-01-15
extract function to clear params
84fa1db0
Merge branch 'tohtana/backward_with_reentrant' of github.com:tohtana/…
90926ba1
fix issue with backward count
ddb54e76
fix backward hook state management
3f6938ea
fix for zero1
db1ff062
fix micro step id count
e2de9a4a
Merge branch 'master' into tohtana/backward_with_reentrant
c33e14f2
Login to write a write a comment.
Login via GitHub