Skip to content

Benchmark Fix and Improvement#84

Merged
DiyouS merged 13 commits intomainfrom
fix/benchmark-improve
Apr 29, 2026
Merged

Benchmark Fix and Improvement#84
DiyouS merged 13 commits intomainfrom
fix/benchmark-improve

Conversation

@DiyouS
Copy link
Copy Markdown
Collaborator

@DiyouS DiyouS commented Apr 24, 2026

This PR consists of two main parts:

  1. Fix a problem that data generation may fail on 16b or 8b kernels if the server does not have GPU-version pytorch.
  2. Add a sparse-attention gemv kernel. The kernel will first search the non-zeros in the activation vector, then use the dense vector to perform GEMV.

@DiyouS DiyouS self-assigned this Apr 24, 2026
@DiyouS DiyouS added bug Something isn't working enhancement New feature or request labels Apr 24, 2026
@DiyouS DiyouS marked this pull request as draft April 24, 2026 11:25
@DiyouS DiyouS linked an issue Apr 24, 2026 that may be closed by this pull request
@DiyouS
Copy link
Copy Markdown
Collaborator Author

DiyouS commented Apr 24, 2026

The current CI failed because of the new gen_data.py does not generate the correct results, which further triggers two other bugs in Spatz SW and CI setup.

I will work on fixing them

@DiyouS DiyouS marked this pull request as ready for review April 27, 2026 16:16
Copy link
Copy Markdown
Contributor

@Navaneeth-KunhiPurayil Navaneeth-KunhiPurayil left a comment

Choose a reason for hiding this comment

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

Thanks for the fixes!
LGTM!

@DiyouS DiyouS force-pushed the fix/benchmark-improve branch from eeab9e2 to fb610d7 Compare April 28, 2026 14:44
@DiyouS DiyouS merged commit 7b2b863 into main Apr 29, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Benchmark Kernel Issues

2 participants