Morph MORPH
SIGN IN SIGN UP
tejas / flash-attention

Fast and memory-efficient exact attention

0 0 208 Python

[Cute,Bwd,Sm100] Try direct atomicadd rmem -> gmem

T
Tri Dao committed
21876951ef2aa3ae7cc94c6bc79428fd7b4ce8c0
Parent: 7254904