david/flash-attention

cermin dari https://github.com/Dao-AILab/flash-attention

Pembuat	SHA1 Pesan	Tanggal
jayhshah	a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236)	2 bulan lalu
jayhshah	c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173)	3 bulan lalu
jayhshah	5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100)	4 bulan lalu
Tri Dao	3aae9c18c1 Revert "Changes For FP8 (#1075)"	4 bulan lalu
ganeshcolfax	1899c970c8 Changes For FP8 (#1075)	4 bulan lalu