david/flash-attention

Autor	SHA1 Nachricht	Datum
Tri Dao	6807b1ea37 Longest-processing-time-first scheduler for causal	vor 1 Tag
Tri Dao	6293008748 Add option for Mma0_is_RS and Mma1_is_RS in attn fwd	vor 1 Woche
Tri Dao	2c996ca25f Use SeqlenInfo for bwd and epilogue	vor 1 Woche
Tri Dao	9c954f7021 Use num_split_heuristics in fwd and fwd_varlen	vor 1 Woche
Tri Dao	f6e165becf Change tile_size and local to avoid wgmma being serialized	vor 1 Woche
Tri Dao	94657af3e8 Add option for not doing intra-WG overlapping of gemm and softmax	vor 3 Wochen
Tri Dao	fc2fd95a18 Renable FP8 kernels	vor 3 Wochen
Tri Dao	586ba914bb Move fwd tile size to a separate file	vor 3 Wochen