Skip to content

adding row-split caching#216

Open
azrael417 wants to merge 5 commits into
mainfrom
tkurth/distributed-attention-row-split-caching
Open

adding row-split caching#216
azrael417 wants to merge 5 commits into
mainfrom
tkurth/distributed-attention-row-split-caching

Conversation

@azrael417

Copy link
Copy Markdown
Collaborator

for performance reasons, the kernel re-computes the dense vs sparse splits with every pass. this is not necessary since this split is constant. This MR moves this into the constructor, improving kernel performance and removing some D2H syncs in the process.

@azrael417 azrael417 self-assigned this Jun 12, 2026
@azrael417 azrael417 requested a review from bonevbs June 12, 2026 06:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant