Skip to content

persistent kernel version of the flash attention forward + FLOP calculation fix when seqlen_q != seqlen_k #316

persistent kernel version of the flash attention forward + FLOP calculation fix when seqlen_q != seqlen_k

persistent kernel version of the flash attention forward + FLOP calculation fix when seqlen_q != seqlen_k #316

Annotations

2 errors

Integration-Tests-AMD (self-hosted, gfx942)

cancelled Dec 18, 2024 in 48s