Commit History

Autor SHA1 Mensaxe Data
  Tri Dao a901c7eeda Make Sm80 forward pass work with persistent scheduler hai 1 mes
  Tri Dao ba2061dfe8 Support cu_seqlens_k_new in flash_attn_with_kvcache hai 1 mes
  Tri Dao 6807b1ea37 Longest-processing-time-first scheduler for causal hai 1 mes
  Tri Dao df96486c31 Decode: varlen, paged KV, leftpad hai 2 meses
  Tri Dao 6e8b25e426 Refactor hai 4 meses
  Ying Zhang dff976a84a fixes hai 5 meses
  jayhshah 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) hai 6 meses
  Tri Dao 74b0761ff7 [FA3] BF16 forward hai 6 meses
  Tri Dao 7f67966cc7 FA3 initial code release hai 6 meses