onnxruntime
Add CUDA correctness fallback for INT2 and mixed-width QMoE
#32743

Merged

Add CUDA correctness fallback for INT2 and mixed-width QMoE #32743

jiafatom merged 8 commits into main from jiafa/int2-qmoe-cuda-correctness

jiafatom requested a review from

copilot-pull-request-reviewer 18 days ago

jiafatom marked this pull request as draft 18 days ago

copilot-pull-request-reviewer commented on 2026-09-22

jiafatom marked this pull request as ready for review 18 days ago

xadupre requested a review from

copilot-pull-request-reviewer 17 days ago

copilot-pull-request-reviewer commented on 2026-09-23

tianleiwu requested changes on 2026-09-23

jiafatom requested a review from

tianleiwu 17 days ago

jiafatom force pushed from 38ec92f8 to a9a18c8d 17 days ago

tianleiwu requested a review from

copilot-pull-request-reviewer 17 days ago

copilot-pull-request-reviewer commented on 2026-09-23

Add CUDA correctness fallback for mixed-width QMoE

ffb81534

Support canonical INT2 CUDA QMoE fallback

55120382

Address CUDA QMoE fallback review feedback

89142215

Gate BF16 QMoE test on kernel build

45d23c91

Validate CUDA QMoE fallback inputs early

24f13d0b

Test mixed INT8 CUDA QMoE fallback

8ef2158a

Handle single-block QMoE scales

5dcede4b

jiafatom force pushed from a9a18c8d to 5dcede4b 17 days ago

jiafatom enabled auto-merge (squash) 16 days ago

tianleiwu requested changes on 2026-09-24

tianleiwu commented on 2026-09-24

Address QMoE fallback review feedback

0284b087

tianleiwu approved these changes on 2026-09-24

jiafatom merged ee5f6e7c into main 16 days ago

jiafatom deleted the jiafa/int2-qmoe-cuda-correctness branch 16 days ago

Reviewers

tianleiwu

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime Add CUDA correctness fallback for INT2 and mixed-width QMoE #32743 Merged

Add CUDA correctness fallback for INT2 and mixed-width QMoE #32743

onnxruntime
Add CUDA correctness fallback for INT2 and mixed-width QMoE
#32743

Merged