PR #8644 Add opt-in fused AutoEP DeepEP row weighting

Add opt-in fused AutoEP DeepEP row weighting #8644

pengdurice merged 8 commits into deepspeedai:master from yh0903:yh0903/autoep-fused-row-weighting

yh0903 marked this pull request as ready for review 21 days ago

yh0903 requested a review from

tohtana 21 days ago

yh0903 requested a review from

hwchen2017 21 days ago

yh0903 requested a review from

loadams 21 days ago

yh0903 requested a review from

tjruwase 21 days ago

Add fused AutoEP DeepEP row weighting

b671eea7

Report the gradient behind a failing optimizer delta

07f8913b

Check every gradient before any update, and report both at a failure

7414e9a5

Compare the fused weight gradient against FP64, and the model with it…

f1d3753e

Cover native fused row weighting through the router

8dd4b87f

Clarify BF16 optimizer and pre-clipping parity semantics

1c26fe61

yh0903 force pushed from c95b7683 to 1c26fe61 20 days ago

pengdurice added will-review

tohtana approved these changes on 2026-09-27

hwchen2017 requested a review from

copilot-pull-request-reviewer 15 days ago

copilot-pull-request-reviewer commented on 2026-09-29

Scope higher parity learning rate to explicit row weighting

90597296

Merge branch 'master' into yh0903/autoep-fused-row-weighting

306735ae

pengdurice approved these changes on 2026-10-01

pengdurice merged 2c4ead4b into master 12 days ago

Reviewers

pengdurice

tohtana

copilot-pull-request-reviewer

hwchen2017

loadams

tjruwase

Assignees

No one assigned

Labels

will-review

Milestone

No milestone

DeepSpeed Add opt-in fused AutoEP DeepEP row weighting #8644 Merged

Add opt-in fused AutoEP DeepEP row weighting #8644

DeepSpeed
Add opt-in fused AutoEP DeepEP row weighting
#8644

Merged