História revízii

Autor SHA1 Správa Dátum
  Tri Dao 20b84d6363 Don't use IntraWGOverlap for hdim 64,512 3 dní pred
  Tri Dao 6752d62aa4 Add dynamic splits 1 týždeň pred
  Tri Dao 74dfa43c8d Fix divide by 0 in causal tile_scheduler for large seqlen 2 týždňov pred
  Tri Dao 7bc3f031a4 Compile for both Sm80 and Sm90 1 mesiac pred
  Tri Dao 7a802796e1 Big refactor and update 1 mesiac pred
  Son Nguyen 478ee666cc Make namespace comment consistent (#1305) 4 mesiacov pred
  jayhshah a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 4 mesiacov pred
  Ying Zhang dff976a84a fixes 6 mesiacov pred
  jayhshah 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 7 mesiacov pred
  Tri Dao 74b0761ff7 [FA3] BF16 forward 7 mesiacov pred
  Tri Dao 7f67966cc7 FA3 initial code release 7 mesiacov pred