Commit History

Autor SHA1 Mensaxe Data
  Ying Zhang 496fdc4f6c Add seqused_q in fwd / bwd and seqused_k in bwd. hai 5 meses
  jayhshah c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) hai 5 meses
  Tri Dao bafe253042 [FA3] Bwd hai 6 meses
  Ying Zhang dfe1a59e4b Add var-seq-len to FA3 fp16 / bf16 fwd (#1072) hai 6 meses
  Cameron Shinn cb516f855b Remove torchlib dependency from cpp files (#1083) hai 6 meses
  Tri Dao 74b0761ff7 [FA3] BF16 forward hai 7 meses
  Tri Dao 7f67966cc7 FA3 initial code release hai 7 meses