Skip to content

Conversation

@jkaniecki
Copy link

Cherry-pick of
6e1be4e


Cherry-pick of
vllm-project@6e1be4e

---------

Signed-off-by: Jan Kaniecki <[email protected]>
Signed-off-by: Jan Kaniecki <[email protected]>
Co-authored-by: Copilot <[email protected]>
Copilot AI review requested due to automatic review settings December 4, 2025 10:51
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR adds support for chunked attention in the vLLM-Gaudi implementation, cherry-picked from the vllm-gaudi repository. Chunked attention is an optimization technique that processes attention in smaller chunks rather than all at once.

Key changes:

  • Added new methods to compute attention bias and block mappings for chunked attention
  • Extended attention metadata structures to include chunked attention fields
  • Integrated chunked attention support into the model execution pipeline

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File Description
vllm_gaudi/v1/worker/hpu_model_runner.py Core implementation of chunked attention logic including bias computation, block mapping, and model runner integration
vllm_gaudi/v1/attention/backends/hpu_attn.py Extended decode metadata factory method to accept chunked attention parameters
vllm_gaudi/attention/backends/hpu_attn.py Added chunked attention fields to metadata class and updated forward pass to use chunked attention blocks

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Signed-off-by: Jan Kaniecki <[email protected]>
@github-actions
Copy link

github-actions bot commented Dec 4, 2025

✅ CI Passed

All checks passed successfully against the following vllm commit:
4fd9d6a85c00ac0186aa9abbeff73fc2ac6c721e

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