ROCm / flash-attention

Fast and memory-efficient exact attention
BSD 3-Clause "New" or "Revised" License
142 stars 46 forks source link

Use same python as build flash-attn to generate ck kernel #66

Closed oraluben closed 4 months ago

oraluben commented 5 months ago

This fixes issues e.g. when python3 is from system and do not have the necessary packages to generate kernels.