onnxruntime
Skip redundant present_key/value copy when aliased to external KV cache (follow-up to #29715)
#31150

Merged

Skip redundant present_key/value copy when aliased to external KV cache (follow-up to #29715) #31150

titaiwangms merged 7 commits into main from copilot/cudnn-sdpa-decode-phase3

titaiwangms requested a review from

copilot-pull-request-reviewer 107 days ago

copilot-pull-request-reviewer commented on 2026-07-30

titaiwangms marked this pull request as ready for review 107 days ago

titaiwangms requested a review from

tianleiwu 106 days ago

titaiwangms added ep:CUDA

tianleiwu commented on 2026-07-31

titaiwangms requested a review from

tianleiwu 102 days ago

titaiwangms force pushed from f5ba8eac to bdc6871b 96 days ago

titaiwangms force pushed from bdc6871b to b75d33fe 74 days ago

tianleiwu requested changes on 2026-08-31

titaiwangms requested a review from

tianleiwu 73 days ago

titaiwangms requested a review from

copilot-pull-request-reviewer 65 days ago

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

titaiwangms requested a review from

copilot-pull-request-reviewer 65 days ago

copilot-pull-request-reviewer commented on 2026-09-10

Add NOTE documenting Phase 2 (internal-cache cuDNN decode) investigat…

5891c0df

Add NOTE documenting Phase 3 (cuDNN SDPA prefill chunking) investigat…

c5040c3f

Skip redundant present_key/value D2D copy when aliased to KV cache

6f557d17

Address review findings for present_kv copy-skip fix

760f851f

Address review findings on present_kv copy-skip PR (#31150)

b96f3e14

Fix CUDA plugin session logging for attention

55a954be

Address final review feedback for attention copy skip

5dc33812

titaiwangms force pushed from 1256fc86 to 5dc33812 58 days ago

titaiwangms enabled auto-merge (squash) 58 days ago

tianleiwu approved these changes on 2026-09-17

titaiwangms merged c7b64756 into main 58 days ago

titaiwangms deleted the copilot/cudnn-sdpa-decode-phase3 branch 58 days ago

Reviewers

tianleiwu

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

ep:CUDA

Milestone

No milestone

onnxruntime Skip redundant present_key/value copy when aliased to external KV cache (follow-up to #29715) #31150 Merged

Skip redundant present_key/value copy when aliased to external KV cache (follow-up to #29715) #31150

onnxruntime
Skip redundant present_key/value copy when aliased to external KV cache (follow-up to #29715)
#31150

Merged