onnxruntime
[GQA] Make present_key/present_value outputs optional and add Gemma4 support
#28242

Merged

[GQA] Make present_key/present_value outputs optional and add Gemma4 support #28242

apsonawane merged 36 commits into main from asonawane/gemma4

Update GQA to support Gemma4

0db95787

Fix the op

ec041dbb

Merge branch 'main' into asonawane/gemma4

bcd82436

github-actions commented on 2026-04-27

Fix lint error

7139053b

Fix webgpu build

0bf63941

apsonawane requested a review from

copilot-pull-request-reviewer 121 days ago

fix webgpu build

3c07e1dd

copilot-pull-request-reviewer commented on 2026-04-27

Address copilot comments

f9051967

tianleiwu requested changes on 2026-04-27

Make GQA present_key/present_value outputs optional for KV-shared layers

d2ead38c

Fix tests

aad74efa

Update the docs

64005dd5

apsonawane requested a review from

copilot-pull-request-reviewer 119 days ago

apsonawane requested a review from

tianleiwu 119 days ago

copilot-pull-request-reviewer commented on 2026-04-29

tianleiwu requested changes on 2026-05-01

Address comments

2b3a2ce3

Address copilot comments

9a14803a

apsonawane requested a review from

copilot-pull-request-reviewer 117 days ago

apsonawane requested a review from

tianleiwu 117 days ago

copilot-pull-request-reviewer commented on 2026-05-01

tianleiwu commented on 2026-05-01

Address comments

2ef269cd

Fix unit tests

6cbe62cb

tianleiwu commented on 2026-05-01

Fix comments

3db82c6b

apsonawane changed the title ~~Add external_key/external_value inputs to GroupQueryAttention for KV-shared layers~~ [GQA] Make present_key/present_value outputs optional and add Gemma4 support 117 days ago

tianleiwu requested changes on 2026-05-02

tianleiwu commented on 2026-05-02

kunal-vaishnavi commented on 2026-05-02

Address comments

bd023f13

Fix unit tests

f0035aab

fix cuda tests

0afa1c91

address comments

ab0ddfba

tianleiwu requested changes on 2026-05-04

tianleiwu requested a review from

copilot-pull-request-reviewer 114 days ago

copilot-pull-request-reviewer commented on 2026-05-04

Support KV-shared decode with separate Q/KV sequence lengths

ec507310

apsonawane requested a review from

copilot-pull-request-reviewer 114 days ago

copilot-pull-request-reviewer commented on 2026-05-05

Address copilot comment:

1069d550

fix cuda pipeline

b1c6271d

[GQA] Support KV-shared layers with empty K/V inputs (kv_sequence_len…

e4281555

apsonawane requested a review from

copilot-pull-request-reviewer 112 days ago

copilot-pull-request-reviewer commented on 2026-05-06

Add unit test and fix documentation

9d4e8c68

Merge branch 'main' into asonawane/gemma4

17ec2437

tianleiwu commented on 2026-05-08

Merge branch 'main' into asonawane/gemma4

4ca15f95

Fix comments

17a155a4

Fix

2dd66c41

apsonawane requested a review from

copilot-pull-request-reviewer 110 days ago

copilot-pull-request-reviewer commented on 2026-05-08

Apply copilot comments

01a1ef6f

tianleiwu commented on 2026-05-08

revert docs

24afd1e7

tianleiwu commented on 2026-05-08

Address comments

29bd4cb9

Merge branch 'main' into asonawane/gemma4

04810935

tianleiwu requested changes on 2026-05-11

Fix comments

85209d6c

apsonawane requested a review from

copilot-pull-request-reviewer 107 days ago

apsonawane requested a review from

tianleiwu 107 days ago

copilot-pull-request-reviewer commented on 2026-05-11

address copilot comments

d819b56c

tianleiwu requested changes on 2026-05-11

Improve

3b996e42

tianleiwu dismissed these changes on 2026-05-11

kunal-vaishnavi commented on 2026-05-11

tianleiwu commented on 2026-05-11

address comments

42af4d76

apsonawane dismissed their stale review via 42af4d76 107 days ago

apsonawane enabled auto-merge (squash) 107 days ago

tianleiwu approved these changes on 2026-05-11

kunal-vaishnavi approved these changes on 2026-05-12

apsonawane merged 83e402e0 into main 107 days ago

apsonawane deleted the asonawane/gemma4 branch 107 days ago

Reviewers

kunal-vaishnavi

tianleiwu

github-actions

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime [GQA] Make present_key/present_value outputs optional and add Gemma4 support #28242 Merged

[GQA] Make present_key/present_value outputs optional and add Gemma4 support #28242

onnxruntime
[GQA] Make present_key/present_value outputs optional and add Gemma4 support
#28242

Merged