david/flash-attention

réplica de https://github.com/Dao-AILab/flash-attention

Autor	SHA1 Mensaxe	Data
Tri Dao	76f14c61c9 Tune fwd tile sizes for Sm86 and Sm89	hai 3 semanas
Tri Dao	51484a7b56 Make backward epilogue work for Sm80	hai 3 semanas
Tri Dao	659a631f4c Rename bwd classes to include Sm90 suffix	hai 3 semanas
Tri Dao	a901c7eeda Make Sm80 forward pass work with persistent scheduler	hai 4 semanas
Tri Dao	5171269dab Implement forward pass for Sm80	hai 1 mes