PR #33176 Add FP16 and BF16 H512 GQA decode support

Add FP16 and BF16 H512 GQA decode support #33176

hanbitmyths merged 6 commits into main from sunghcho/gemma-4-h512-gqa

Add FP16 and BF16 H512 GQA decode support

8a5438b3

Merge main and retain H512 GQA eligibility

e4dedff7

Merge remote-tracking branch 'origin/main' into sunghcho/gemma-4-h512…

215b709d

hanbitmyths requested a review from

copilot-pull-request-reviewer 13 days ago

copilot-pull-request-reviewer commented on 2026-10-08

hanbitmyths marked this pull request as ready for review 12 days ago

hanbitmyths requested a review from

tianleiwu 12 days ago

hanbitmyths requested a review from

apsonawane 12 days ago

tianleiwu commented on 2026-10-08

apsonawane commented on 2026-10-08

Merge main into H512 GQA branch

d73ed2d1

Keep native H512 GQA preference decode-only

b39dd845

hanbitmyths requested a review from

copilot-pull-request-reviewer 12 days ago

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

Keep MEA eligible for one-token H512 first prompts

c5f8eab4

hanbitmyths requested a review from

copilot-pull-request-reviewer 12 days ago

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

apsonawane approved these changes on 2026-10-09

tianleiwu approved these changes on 2026-10-09

hanbitmyths merged 426d9b35 into main 11 days ago

hanbitmyths deleted the sunghcho/gemma-4-h512-gqa branch 11 days ago

Reviewers

tianleiwu

apsonawane

Assignees

No one assigned

Labels

None yet

Milestone

No milestone