Repository navigation

[bugfix] Preserve masked FlashAttention mask and scale - #2080

Open

bingdongni wants to merge 4 commits into

hao-ai-lab:mainfrom

bingdongni:fix/masked-flash-attention-20261108

Open

[bugfix] Preserve masked FlashAttention mask and scale#2080
bingdongni wants to merge 4 commits into
hao-ai-lab:mainfrom
bingdongni:fix/masked-flash-attention-20261108

Commits on Nov 8, 2026
- [bugfix] Preserve masked FlashAttention semantics
  bingdongni
  committed
  b03d2c6
- [test] Guard unequal-length attention on FA3
  bingdongni
  committed
  23e89e9
- [test] Cover supported padding mask ranks
  bingdongni
  committed
  8e43b54
- [test] Remove unnecessary inline comment
  bingdongni
  committed
  4731dfa