onnxruntime
Support bidirectional GroupQueryAttention on CPU and CUDA
#31704

Merged

Support bidirectional GroupQueryAttention on CPU and CUDA #31704

tianleiwu merged 9 commits into main from copilot/add-causal-attribute-to-groupqueryattention

Add GroupQueryAttention causal attribute

059a58ff

Test GroupQueryAttention causal masks

f6ec0132

Reject bidirectional GQA in JS WebGPU

4047b614

copilot-swe-agent assigned

copilot-swe-agent 68 days ago

copilot-swe-agent assigned

tianleiwu 68 days ago

copilot-swe-agent requested a review from

tianleiwu 68 days ago

Update GroupQueryAttention operator docs

f1e4f2f1

Expand GQA causal interaction tests

aa079300

tianleiwu marked this pull request as ready for review 67 days ago

fix(cuda): guard bidirectional GQA fallbacks

7ec28472

tianleiwu requested a review from

hariharans29 61 days ago

tianleiwu requested a review from

kunal-vaishnavi 61 days ago

tianleiwu requested a review from

titaiwangms 61 days ago

tianleiwu approved these changes on 2026-08-13

titaiwangms requested a review from

copilot-pull-request-reviewer 61 days ago

copilot-pull-request-reviewer commented on 2026-08-13

hariharans29 dismissed these changes on 2026-08-13

fix(gqa): validate bidirectional attention

9eab39cd

tianleiwu dismissed their stale review via 9eab39cd 61 days ago

fix(gqa): update renamed causal member references

db0a602f

fix(gqa): update JS causal member reference

44c5dd77

titaiwangms approved these changes on 2026-08-14

tianleiwu merged 125ea215 into main 60 days ago

tianleiwu deleted the copilot/add-causal-attribute-to-groupqueryattention branch 60 days ago

Reviewers

titaiwangms

tianleiwu

hariharans29

copilot-pull-request-reviewer

kunal-vaishnavi

Assignees

tianleiwu

Copilot

Labels

None yet

Milestone

No milestone

onnxruntime Support bidirectional GroupQueryAttention on CPU and CUDA #31704 Merged

Support bidirectional GroupQueryAttention on CPU and CUDA #31704

onnxruntime
Support bidirectional GroupQueryAttention on CPU and CUDA
#31704

Merged