david/flash-attention

огледало от https://github.com/Dao-AILab/flash-attention

Автор	SHA1 Съобщение	Дата
jayhshah	c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173)	преди 3 месеца
jayhshah	5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100)	преди 4 месеца
Tri Dao	3aae9c18c1 Revert "Changes For FP8 (#1075)"	преди 4 месеца
ganeshcolfax	1899c970c8 Changes For FP8 (#1075)	преди 4 месеца