Commit History

Auteur SHA1 Bericht Datum
  Tri Dao 2c996ca25f Use SeqlenInfo for bwd and epilogue 1 maand geleden
  Tri Dao 29cdfedd80 Use Bulk reduce instead of TMA for dQaccum, split across WGs 2 maanden geleden
  Tri Dao 82dc825759 Don't use the unsafe convert_type function 2 maanden geleden
  Tri Dao df96486c31 Decode: varlen, paged KV, leftpad 2 maanden geleden
  Tri Dao 6e8b25e426 Refactor 4 maanden geleden
  Ying Zhang db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 5 maanden geleden
  Cameron Shinn 3cea2fb6ee Add ArchTag to pre/postprocess bwd kernels (#1180) 5 maanden geleden
  Tri Dao bafe253042 [FA3] Bwd 6 maanden geleden
  Tri Dao 7f67966cc7 FA3 initial code release 6 maanden geleden