Skip to content

Latest commit

 

History

History
7 lines (6 loc) · 481 Bytes

File metadata and controls

7 lines (6 loc) · 481 Bytes

Documentation

  • Benchmarks — measured M2 performance and tiled-versus-naive matmul results.
  • Correctness — CPU-to-Metal parity and end-to-end GPT-2 validation.
  • Contributing — local checks to run before proposing a change.
  • Test command reference — the smallest useful validation command for each task.
  • Troubleshooting — platform, model-asset, and test setup notes.