onnxruntime
[CPU] Support right-padded prefill chunks in GroupQueryAttention
#33505

Merged

[CPU] Support right-padded prefill chunks in GroupQueryAttention #33505

tianleiwu merged 5 commits into main from tlwu/gqa_cpu_right_padded_prefill_chunk_slide_window

tianleiwu requested a review from

copilot-pull-request-reviewer 19 days ago

copilot-pull-request-reviewer commented on 2026-10-21

tianleiwu marked this pull request as draft 19 days ago

tianleiwu requested a review from

apsonawane 17 days ago

tianleiwu requested a review from

kunal-vaishnavi 17 days ago

tianleiwu requested a review from

titaiwangms 17 days ago

tianleiwu requested a review from

copilot-pull-request-reviewer 9 days ago

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

tianleiwu requested a review from

copilot-pull-request-reviewer 8 days ago

copilot-pull-request-reviewer commented on 2026-11-01

tianleiwu force pushed from ee6254e1 to e845abec 8 days ago

tianleiwu changed the base branch from main to tlwu/qga_right_padded_prefill_chunk_slide_window 8 days ago

tianleiwu force pushed from e845abec to 8110e03e 8 days ago

tianleiwu marked this pull request as ready for review 7 days ago

tianleiwu force pushed from 8110e03e to 5a499a82 7 days ago

titaiwangms dismissed these changes on 2026-11-02

Base automatically changed from tlwu/qga_right_padded_prefill_chunk_slide_window to main 6 days ago

feat(cpu): support query_seqlens in GroupQueryAttention

a264c97d

fix(cpu): zero-fill unwritten qk_output columns

802297c6

fix(cpu): require present outputs whenever query_seqlens is set

59a90138

test(cpu): cover packed QKV chunks and WebGPU rejection of query_seqlens

666e9844

docs(cpu): update comments that say only CUDA implements query_seqlens

bcf9f62f

tianleiwu dismissed their stale review via bcf9f62f 6 days ago

tianleiwu force pushed from 5a499a82 to bcf9f62f 6 days ago

titaiwangms approved these changes on 2026-11-03

tianleiwu enabled auto-merge (squash) 6 days ago

tianleiwu merged 3fc465dc into main 6 days ago

tianleiwu deleted the tlwu/gqa_cpu_right_padded_prefill_chunk_slide_window branch 6 days ago

Reviewers

titaiwangms

copilot-pull-request-reviewer

apsonawane

kunal-vaishnavi

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime [CPU] Support right-padded prefill chunks in GroupQueryAttention #33505 Merged

[CPU] Support right-padded prefill chunks in GroupQueryAttention #33505

onnxruntime
[CPU] Support right-padded prefill chunks in GroupQueryAttention
#33505

Merged