Repository navigation

CUDA backend: lower fused attention variants - #24455

Closed

pytorchbot wants to merge 33 commits into

release/1.6from

cherry-pick-24426-by-pytorch_bot_bot_

Closed

CUDA backend: lower fused attention variants#24455
pytorchbot wants to merge 33 commits into
release/1.6from
cherry-pick-24426-by-pytorch_bot_bot_

Commits on Oct 29, 2026
- Fix crashes, wrong outputs and lowering failures in several backends and in exir (#24222)
  
  shoumikhin
  authored and
  pytorchbot
  committed
  6207bdf
- Merge release/1.6 into the #24222 cherry-pick and restore three files
  shoumikhin
  committed
  8127c0f
Commits on Oct 30, 2026
Commits on Oct 31, 2026
Commits on Nov 2, 2026
- CUDA backend: delete each output handle once when a program returns one tensor twice (#24420)
  
  shoumikhin
  authored and
  pytorchbot
  committed
  4521c2c
- CUDA backend: lower fused attention variants (#24426)
  
  shoumikhin
  authored and
  pytorchbot
  committed
  13abd1d