Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Flash Decode GQA (and MQA) Improvements (Round 1) #12739

Merged
merged 7 commits into from
Sep 20, 2024
Merged

Flash Decode GQA (and MQA) Improvements (Round 1) #12739

merged 7 commits into from
Sep 20, 2024

Conversation

caixunshiren
Copy link
Contributor

@caixunshiren caixunshiren commented Sep 16, 2024 •
edited by sraizada-tt
Loading

Ticket

This PR contains the round 1 improvements outlined in #12330 :

  • Support transpose_q, which both q of shape [1 x qh x b x h] and [1 x b x qh x h] are supported as input
  • Support GQA on a shared cache, which KV of shape [1 x kh x s x h] is supported
  • Add support for tensor indices for GQA

FYI @sraizada-tt @cglagovichTT

Post commit pipeline: https://github.com/tenstorrent/tt-metal/actions/runs/10956618476/job/30423043917

@caixunshiren caixunshiren added kernels kernels, such as hlks or llks or below P1 LLM_feature llama3 labels Sep 16, 2024
@caixunshiren caixunshiren self-assigned this Sep 16, 2024
@caixunshiren caixunshiren marked this pull request as ready for review September 17, 2024 18:23
@caixunshiren
Copy link
Contributor Author

caixunshiren commented Sep 18, 2024 •
edited
Loading

all post commit: https://github.com/tenstorrent/tt-metal/actions/runs/10951786445

@caixunshiren caixunshiren changed the title Flash Decode GQA (and MQA) Improvements Flash Decode GQA (and MQA) Improvements (Round 1) Sep 18, 2024
@sraizada-tt sraizada-tt merged commit 72f5ccd into main Sep 20, 2024
105 checks passed
@sraizada-tt sraizada-tt deleted the page-attention-gqa branch September 20, 2024 10:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@sraizada-tt sraizada-tt sraizada-tt approved these changes

@uaydonat uaydonat uaydonat approved these changes

@cfjchu cfjchu cfjchu approved these changes

@cglagovichTT cglagovichTT cglagovichTT approved these changes

@TT-BrianLiu TT-BrianLiu TT-BrianLiu approved these changes

@eyonland eyonland Awaiting requested review from eyonland

@patrickroberts patrickroberts Awaiting requested review from patrickroberts

@yan-zaretskiy yan-zaretskiy Awaiting requested review from yan-zaretskiy

@xanderchin xanderchin Awaiting requested review from xanderchin

@ayerofieiev-tt ayerofieiev-tt Awaiting requested review from ayerofieiev-tt ayerofieiev-tt is a code owner

@dmakoviichuk-tt dmakoviichuk-tt Awaiting requested review from dmakoviichuk-tt dmakoviichuk-tt is a code owner

@yieldthought yieldthought Awaiting requested review from yieldthought

@mtairum mtairum Awaiting requested review from mtairum

Assignees

@caixunshiren caixunshiren

Labels
kernels kernels, such as hlks or llks or below llama3 LLM_feature P1
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
6 participants
@caixunshiren @TT-BrianLiu @cglagovichTT @cfjchu @uaydonat @sraizada-tt