support grouped query attention(MQA & GQA) for flash_attn#22

Merged

iclementine merged 2 commits intoFlagOpen:mainfrom iclementine:mqa

May 27, 2024

+165-81

Commits on May 13, 2024

support grouped query attention(GQA) for flash_attn(fwd, bwd, split_kv, total_attention)
iclementine
committed

Commits on May 27, 2024

add mqa/gqa into feature list; update documentations and testings for flash attention.
iclementine
committed