Repository navigation

Parallelize CPU MultiHeadAttention QKV transposes - #33887

Merged

David Fan (jiafatom) merged 4 commits into

mainfrom

copilot/optimize-cpu-multattention-performance

Nov 9, 2026

Merged

Parallelize CPU MultiHeadAttention QKV transposes#33887
David Fan (jiafatom) merged 4 commits into
mainfrom
copilot/optimize-cpu-multattention-performance

Commits on Nov 4, 2026
- Initial plan
  Copilot
  authored
  06bfcca
- Add CPU attention bias regression coverage with threaded QKV transpose
  
  Copilot
  and
  Tianlei Wu (tianleiwu)
  authored
  fa2d72d
- Use the CPU thread pool for MultiHeadAttention QKV transposes
  
  Copilot
  and
  Tianlei Wu (tianleiwu)
  authored
  20a5962
Commits on Nov 8, 2026
- fix(cpu): avoid tiny QKV transpose scheduling
  
  Tianlei Wu (tianleiwu)
  and
  Copilot
  committed
  87aaa48