flash_fwd_hdim96_bf16_causal_sm80.cu 444 B

1234567891011
  1. // Copyright (c) 2023, Tri Dao.
  2. // Splitting the different head dimensions to different files to speed up
  3. // compilation. This file is auto-generated. See "generate_kernels.py"
  4. #include "flash_fwd_launch_template.h"
  5. template <>
  6. void run_mha_fwd_<cutlass::bfloat16_t, 96, true>(Flash_fwd_params& params,
  7. cudaStream_t stream) {
  8. run_mha_fwd_hdim96<cutlass::bfloat16_t, true>(params, stream);
  9. }