DeepSpeed
AutoEP: one program per token in the fused weighted restore forward
#8678

Merged

AutoEP: one program per token in the fused weighted restore forward #8678

tohtana merged 2 commits into deepspeedai:master from yh0903:yh0903/autoep-restore-whole-row

Give each fused weighted restore program a whole token

3c477efa

Document whole-token restore numerics and guard against FP contraction

0ec88e95

yh0903 marked this pull request as ready for review 20 days ago

yh0903 requested a review from

tohtana 20 days ago

yh0903 requested a review from

tjruwase 20 days ago

yh0903 requested a review from

loadams 20 days ago

hwchen2017 requested a review from

copilot-pull-request-reviewer 18 days ago

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

tohtana approved these changes on 2026-09-29

tohtana enabled auto-merge 18 days ago

pengdurice added will-review

tohtana merged 11b24d21 into master 17 days ago

Reviewers

tohtana

tjruwase

loadams

Assignees

No one assigned

Labels

will-review

Milestone

No milestone