Skip to content

Updates to JAX flash attention kernel. - #4578

Open
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_952800018
Open

Updates to JAX flash attention kernel.#4578
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_952800018

Conversation

@copybara-service

@copybara-service copybara-service Bot commented Jul 23, 2026

Copy link
Copy Markdown
Contributor

Updates to JAX flash attention kernel.

  • Sink jnp.dynamic-slice instructions fetching blocks of q, k and v into the inner loop.
  • Forced final @v layout for better utilization
  • Forcing fusions with must_fuse_call
  • Removed calculations related to mask blocking when the mask is causal

FUTURE_COPYBARA_INTEGRATE_REVIEW=#4513 from Shuwen-Fang:sort_one_sparsecore cccf4ec

@copybara-service
copybara-service Bot force-pushed the test_952800018 branch 5 times, most recently from 0e10bb7 to 1cf8d0d Compare July 29, 2026 19:53
- Sink jnp.dynamic-slice instructions fetching blocks of q, k and v into the inner loop.
- Forced final @v layout for better utilization
- Forcing fusions with must_fuse_call
- Removed calculations related to mask blocking when the mask is causal

FUTURE_COPYBARA_INTEGRATE_REVIEW=#4513 from Shuwen-Fang:sort_one_sparsecore cccf4ec
PiperOrigin-RevId: 952800018
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