Commit History

Автор SHA1 Съобщение Дата
  Tri Dao 02541ac9e8 [CE] Assert logit_scale > 0 преди 2 седмици
  Tri Dao 803f609aa1 Fix comment in assert преди 2 седмици
  Tri Dao 454ce31594 [FA3] Compile with nvcc 12.8 instead of 12.3 преди 2 седмици
  Tri Dao 5231d95fe1 Drop Pytorch 2.1 преди 2 седмици
  Tri Dao 979702c87a Bump to v2.7.4 преди 2 седмици
  Tri Dao bb135af07c Don't compile for CUDA 11, compile for official pytorch 2.6.0 преди 2 седмици
  Aman Karmani cd393e0ace [Build] Update version of setuptools used to generate core package (#1460) преди 2 седмици
  Michał Górny 6b1d059eda Support ROCM builds from source distribution, and improve error handling (#1446) преди 3 седмици
  Driss Guessous bc482cbf91 Add a macro for namespace (#1419) преди 1 месец
  Ying Zhang 0fcd405cfe Merge pull request #1442 from houseroad/replace_c10_optional преди 1 месец
  Lu Fang 74aed78373 Replace c10::optional with std::optional in flash_attn преди 1 месец
  Tri Dao 22c0358f4b Fix nvcc_from_env not found преди 1 месец
  Kirthi Shankar Sivamani 89c5a7dd4e Change version to 2.7.3 (#1437) преди 1 месец
  Tri Dao 2ac6c986be Fix Sm80 tile_count_semaphore, adjust test tolerance преди 1 месец
  Kirthi Shankar Sivamani 07bddf918a Blackwell support (#1436) преди 1 месец
  rocking 77ad12d24e [AMD ROCm] Support variable length of page attention (#1431) преди 1 месец
  Kirthi Shankar Sivamani d57f826835 Expose `zero_tensors` arg in varlen functions (#1433) преди 1 месец
  Tri Dao a93359a2bf If PackGQA, use producer threads instead of Mma threads to load Q преди 1 месец
  Tri Dao e94f7e89dc Always enable PackGQA is Split to reduce compilation and binary size преди 1 месец
  Tri Dao 40fa35acd8 Always enable PackGQA if PagedKV to reduce compilation and bin size преди 1 месец
  Tri Dao a84a237d2a Split bwd softcap compilation units for Sm80 преди 1 месец
  Tri Dao 518e919a60 Fix softcap compilation преди 1 месец
  Tri Dao ea8cd7fe7b Ungroup hdim and group softcap for Sm80 compilation преди 1 месец
  Tri Dao 1e3208566a Tune tile sizes for compilation преди 1 месец
  Tri Dao 84f1287e42 Rename bool_constant<true> to true_type, same w bool_constant<false> преди 1 месец
  Tri Dao df5fe55264 Change tile sizes for Sm8x to reduce stack frame преди 1 месец
  Tri Dao 8dd0b479d5 Always enable PackGQA for Sm8x to reduce compilation and binary size преди 1 месец
  Garrett Byrd a69fc21595 Added hipBLAS/cuBLAS distinction in benchmark_gemm.py (#1393) преди 1 месец
  Kevin Wang efbf19cd15 Fix incorrect torch dtype (#1399) преди 1 месец
  Cao Dong d525b38291 fix bug when is_grad is false (#1406) преди 1 месец