Historique des commits

Auteur SHA1 Message Date
  Tri Dao 5458c78e6d Remove sink token il y a 4 jours
  Tri Dao 6752d62aa4 Add dynamic splits il y a 1 semaine
  Ted Zadouri 06e34f62d1 Enable MLA flag in FA3 (rope=64, latent=512) (#1504) il y a 1 semaine
  Tri Dao 15cf7ee435 Rename collective_mainloop -> mainloop, move tile_scheduler variable il y a 3 semaines
  Tri Dao 4e8496a781 Support Qv il y a 3 semaines
  Tri Dao ed53b5fc4c Fix compilation for FP8 when kHeadDimV != kHeadDim il y a 3 semaines
  Tri Dao 4c8819d8c6 Rename Mma0->MmaQK, Mma1->MmaPV, use Cluster only if hdimV >= 192 il y a 3 semaines
  Tri Dao 6d199aa207 Fix shape_O in epilogue params when kHeadDimV != kHeadDim il y a 1 mois
  Tri Dao 2a204125ae Implement HeadDim_V != HeadDim_QK, support hdimQK=192, hdimV=128 il y a 1 mois
  Tri Dao 454ce31594 [FA3] Compile with nvcc 12.8 instead of 12.3 il y a 1 mois
  Tri Dao 1e3208566a Tune tile sizes for compilation il y a 1 mois
  Tri Dao df5fe55264 Change tile sizes for Sm8x to reduce stack frame il y a 1 mois
  Tri Dao 180ff782dd Template for Sm86 il y a 1 mois
  Tri Dao 7bc3f031a4 Compile for both Sm80 and Sm90 il y a 1 mois
  Tri Dao 7a802796e1 Big refactor and update il y a 1 mois
  Kai Londenberg b443207c1f Paged Attention support for FA3 (#1268) il y a 3 mois
  jayhshah a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) il y a 4 mois
  Ying Zhang 1c9717d699 address comments il y a 5 mois
  Ying Zhang dff976a84a fixes il y a 6 mois
  Ying Zhang 7b4e68e04f hopper local attention il y a 6 mois
  Ying Zhang db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. il y a 6 mois
  jayhshah c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) il y a 6 mois
  Tri Dao bafe253042 [FA3] Bwd il y a 7 mois
  jayhshah 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) il y a 7 mois
  Tri Dao 3aae9c18c1 Revert "Changes For FP8 (#1075)" il y a 7 mois
  ganeshcolfax 1899c970c8 Changes For FP8 (#1075) il y a 7 mois
  Ying Zhang dfe1a59e4b Add var-seq-len to FA3 fp16 / bf16 fwd (#1072) il y a 7 mois
  Cameron Shinn cb516f855b Remove torchlib dependency from cpp files (#1083) il y a 7 mois
  Tri Dao 74b0761ff7 [FA3] BF16 forward il y a 7 mois
  Tri Dao 7f67966cc7 FA3 initial code release il y a 7 mois