PR #7184 Reland perf fix for nan inf check

Reland perf fix for nan inf check #7184

loadams merged 6 commits into deepspeedai:master from nelyahu:reland_perf_fix_for_nan_inf_check

nelyahu requested a review from

tjruwase 1 year ago

nelyahu requested a review from

tohtana 1 year ago

nelyahu force pushed from 6f123cc7 to 2166ff45 1 year ago

tjruwase commented on 2025-03-30

Revert "Fix issue #5242 grad_norm and loss is nan (#7171)"

07cf3e21

replace boolen compute with torch.where for nan/inf detection

37a0454d

move logic to runtime.utils and reuse code

cddef4c7

nelyahu force pushed from 889c90d9 to cddef4c7 1 year ago

tjruwase approved these changes on 2025-03-31

fix formatting issues

5797162c

Merge branch 'master' into reland_perf_fix_for_nan_inf_check

efff1bf4

Merge branch 'master' into reland_perf_fix_for_nan_inf_check

4a003e01

loadams merged 3c1817f3 into master 1 year ago

Reviewers

tjruwase

tohtana

Assignees

No one assigned

Labels

None yet

Milestone

No milestone