DeepSpeed
[Engine] Only scale gradients if scale_wrt_gas is True
#7724
Merged

[Engine] Only scale gradients if scale_wrt_gas is True #7724

tohtana merged 5 commits into deepspeedai:master from kashif:fix-hook
kashif
kashif kashif requested a review from tjruwase tjruwase 38 days ago
kashif kashif requested a review from tohtana tohtana 38 days ago
kashif Only scale gradients if scale_wrt_gas is True
a1f022bd
kashif kashif force pushed from 271dfca7 to a1f022bd 38 days ago
stas00
tohtana
tohtana approved these changes on 2025-12-11
tohtana
tohtana add test to verify scale_wrt_gas=False
149b22f9
tohtana
kashif Merge pull request #1 from tohtana/tohtana/add_test_disable_scaling
01e6d730
kashif kashif requested a review from loadams loadams 37 days ago
kashif Merge branch 'master' into fix-hook
5df5d990
kashif
kashif dry up the method
1428da69
kashif kashif force pushed from 52e2120f to 1428da69 37 days ago
tohtana tohtana enabled auto-merge (squash) 37 days ago
tohtana tohtana merged d568375e into master 37 days ago
kashif kashif deleted the fix-hook branch 37 days ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone