onnxruntime
[CUDA] Add Tensor-Core BF16 INT2 QMoE GEMV kernel and 3-mask BF16 converter
#33786

Merged

[CUDA] Add Tensor-Core BF16 INT2 QMoE GEMV kernel and 3-mask BF16 converter #33786

tianleiwu merged 2 commits into main from tlwu/20261031/qmoe-cuda-int2-tc

tianleiwu requested a review from

copilot-pull-request-reviewer 11 days ago

copilot-pull-request-reviewer commented on 2026-10-31

[CUDA] Add Tensor-Core BF16 INT2 QMoE GEMV kernel and 3-mask BF16 con…

9ae92834

tianleiwu force pushed from 36a4284b to 9ae92834 11 days ago

tianleiwu added model:qwen-3.8

tianleiwu requested a review from

copilot-pull-request-reviewer 10 days ago

copilot-pull-request-reviewer commented on 2026-11-02

fix(cuda): restore QMoE scale prepack gate and reuse 12-token INT2 GE…

bbf432ca

apsonawane commented on 2026-11-02

apsonawane approved these changes on 2026-11-02

baijumeswani approved these changes on 2026-11-02

tianleiwu enabled auto-merge (squash) 9 days ago

tianleiwu merged f42c839f into main 9 days ago

tianleiwu deleted the tlwu/20261031/qmoe-cuda-int2-tc branch 9 days ago

Reviewers

baijumeswani

apsonawane

Assignees

No one assigned

Labels

model:qwen-3.8

Milestone

No milestone