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.

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

[CPU][ARM] PagedAttention fixes #28017

Open
wants to merge 1 commit into
base: master
Choose a base branch
from

Conversation

alvoron
Copy link
Contributor

@alvoron alvoron commented Dec 11, 2024

Details:

  • Fix: do not infer non-executable node if it's in constant path
  • Fix: redefine the 2nd output memory of PagedAttention node if m_hasScore is false
  • These 2 fixes required to support fp16 PagedAttention: Aarch64 paged attention enablement #27841

Tickets:

  • ticket-id

@alvoron alvoron added the platform: arm OpenVINO on ARM / ARM64 label Dec 11, 2024
@alvoron alvoron requested review from a team as code owners December 11, 2024 13:02
@github-actions github-actions bot added the category: CPU OpenVINO CPU plugin label Dec 11, 2024
@@ -177,7 +177,7 @@ void PagedAttention::execute(dnnl::stream strm) {
VectorDims scoreDims{len};
redefineOutputMemory({outDims, scoreDims});
} else {
redefineOutputMemory(0, outDims);
redefineOutputMemory({outDims, {}});
Copy link
Contributor

Choose a reason for hiding this comment

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

{} implies a scalar e.g. one element tensor ({1} in the plugin's representation). Is it what we should expect here, or it should be an empty tensor {0}?

Copy link
Contributor

Choose a reason for hiding this comment

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

That actaully means empty output. So most likelly {0} suites better

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
category: CPU OpenVINO CPU plugin platform: arm OpenVINO on ARM / ARM64
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants