Commit Verlauf

Autor SHA1 Nachricht Datum
  Tri Dao 0890032358 Implement backward pass for Sm80 vor 1 Monat
  Tri Dao 14894c5717 Make BwdPostprocessKernel work with Sm80 vor 2 Monaten
  Tri Dao 2c996ca25f Use SeqlenInfo for bwd and epilogue vor 2 Monaten
  Tri Dao 29cdfedd80 Use Bulk reduce instead of TMA for dQaccum, split across WGs vor 3 Monaten
  Tri Dao 82dc825759 Don't use the unsafe convert_type function vor 3 Monaten
  Tri Dao 6e8b25e426 Refactor vor 5 Monaten
  Ying Zhang db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. vor 6 Monaten
  Cameron Shinn 3cea2fb6ee Add ArchTag to pre/postprocess bwd kernels (#1180) vor 6 Monaten
  Tri Dao bafe253042 [FA3] Bwd vor 7 Monaten