Historique des commits

Auteur SHA1 Message Date
  Tri Dao 6807b1ea37 Longest-processing-time-first scheduler for causal il y a 1 mois
  Tri Dao 29cdfedd80 Use Bulk reduce instead of TMA for dQaccum, split across WGs il y a 1 mois
  Tri Dao 314b9edfc0 Don't need to link to cuda lib anymore il y a 1 mois
  Tri Dao f11624b746 Disable --split-compile due to ptxas register allocation failure il y a 1 mois
  Tri Dao e8a1edbeb2 Clean up some #include il y a 1 mois
  Tri Dao 8ae77ea17c Download nvcc 12.3 to compile for best perf il y a 1 mois
  Tri Dao 42fc4962f0 Uncomment tanh softcapping il y a 1 mois
  Tri Dao 6bc55b571c Use --split-compile to speed up compilation il y a 1 mois
  Tri Dao 9553b2728f More env vars to disable features il y a 1 mois
  Tri Dao 3248babb9e QOL: Use env var to selectively disable features il y a 1 mois
  Tri Dao 94657af3e8 Add option for not doing intra-WG overlapping of gemm and softmax il y a 1 mois
  Tri Dao fc2fd95a18 Renable FP8 kernels il y a 1 mois
  Tri Dao 64d92bce53 Split PagedKV into separate .cu files to speed up compilation il y a 1 mois
  Tri Dao 018b9af683 Move .cu files to instantiations, use generate_kernels.py il y a 1 mois
  Tri Dao 82c1aa3514 Move PackGQA epilogue code to pack_gqa.h il y a 2 mois
  Tri Dao df96486c31 Decode: varlen, paged KV, leftpad il y a 2 mois
  Tri Dao 6e8b25e426 Refactor il y a 3 mois
  hlky 8476986721 Fix FAv3 compilation with MSVC (#1240) il y a 4 mois
  Ying Zhang dff976a84a fixes il y a 4 mois
  Tri Dao bafe253042 [FA3] Bwd il y a 5 mois
  jayhshah 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) il y a 5 mois
  Tri Dao 3aae9c18c1 Revert "Changes For FP8 (#1075)" il y a 5 mois
  ganeshcolfax 1899c970c8 Changes For FP8 (#1075) il y a 5 mois
  janEbert 3c4053b75c Make FA3 externally importable (#1053) il y a 5 mois
  Tri Dao 74b0761ff7 [FA3] BF16 forward il y a 6 mois
  Tri Dao 7f67966cc7 FA3 initial code release il y a 6 mois