Cronologia Commit

Autore SHA1 Messaggio Data
  jayhshah a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 mesi fa
  Ying Zhang 7b4e68e04f hopper local attention 3 mesi fa
  Ying Zhang db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 3 mesi fa
  jayhshah c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) 3 mesi fa
  Tri Dao bafe253042 [FA3] Bwd 4 mesi fa
  Ying Zhang dfe1a59e4b Add var-seq-len to FA3 fp16 / bf16 fwd (#1072) 4 mesi fa
  Cameron Shinn cb516f855b Remove torchlib dependency from cpp files (#1083) 4 mesi fa
  Tri Dao 74b0761ff7 [FA3] BF16 forward 5 mesi fa
  Tri Dao 7f67966cc7 FA3 initial code release 5 mesi fa