onnxruntime
Extend CUDA QMoE packed INT2 prefill block sizes and weight combinations
#33092

Merged

Extend CUDA QMoE packed INT2 prefill block sizes and weight combinations #33092

jiafatom merged 4 commits into main from qmoe-int2-prefill-block128

Extend packed QMoE INT2 prefill block sizes and weight combinations

f4e80c9c

jiafatom requested a review from

copilot-pull-request-reviewer 35 days ago

github-actions commented on 2026-10-05

copilot-pull-request-reviewer commented on 2026-10-05

Use unset defaults for packed QMoE prefill configuration

71428910

Preserve INT2 decode fallback test with packed prefill enabled

69bc25fb

jiafatom marked this pull request as ready for review 35 days ago

apsonawane requested a review from

copilot-pull-request-reviewer 35 days ago

apsonawane commented on 2026-10-05

copilot-pull-request-reviewer commented on 2026-10-05

Gate QMoE block-size-32 prepacking on packed prefill

db8a7625

tianleiwu commented on 2026-10-05

tianleiwu approved these changes on 2026-10-05

jiafatom enabled auto-merge (squash) 35 days ago

jiafatom merged 0c67e77b into main 34 days ago

jiafatom deleted the qmoe-int2-prefill-block128 branch 34 days ago

Reviewers

tianleiwu

apsonawane

github-actions

Assignees

No one assigned

Labels

None yet

Milestone

No milestone