david/flash-attention

mirror da https://github.com/Dao-AILab/flash-attention

Autore	SHA1 Messaggio	Data
jayhshah	c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173)	5 mesi fa
Ying Zhang	a3a257c71d Fix out-of-bound writes for var-seq-len zero-length KVs	5 mesi fa
jayhshah	5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100)	6 mesi fa
Ying Zhang	dfe1a59e4b Add var-seq-len to FA3 fp16 / bf16 fwd (#1072)	6 mesi fa
Tri Dao	74b0761ff7 [FA3] BF16 forward	6 mesi fa
Tri Dao	7f67966cc7 FA3 initial code release	6 mesi fa