diff options
Diffstat (limited to 'candle-flash-attn/kernels/flash_fwd_hdim64_bf16_sm80.cu')
| -rw-r--r-- | candle-flash-attn/kernels/flash_fwd_hdim64_bf16_sm80.cu | 13 |
1 files changed, 2 insertions, 11 deletions
diff --git a/candle-flash-attn/kernels/flash_fwd_hdim64_bf16_sm80.cu b/candle-flash-attn/kernels/flash_fwd_hdim64_bf16_sm80.cu index fffcbebb..22eac878 100644 --- a/candle-flash-attn/kernels/flash_fwd_hdim64_bf16_sm80.cu +++ b/candle-flash-attn/kernels/flash_fwd_hdim64_bf16_sm80.cu @@ -1,19 +1,10 @@ // Copyright (c) 2023, Tri Dao. - // Splitting the different head dimensions to different files to speed up compilation. +// This file is auto-generated. See "generate_kernels.py" #include "flash_fwd_launch_template.h" -// template<> -// void run_mha_fwd_<cutlass::bfloat16_t, 64>(Flash_fwd_params ¶ms, cudaStream_t stream) { -// using elem_type = cutlass::bfloat16_t; -// if (params.p_dropout == 1.f) { -// run_flash_fwd<Flash_fwd_kernel_traits<64, 128, 64, 4, true, false, elem_type>, false>(params, stream); -// } else { -// run_flash_fwd<Flash_fwd_kernel_traits<64, 128, 64, 4, false, false, elem_type>, true>(params, stream); -// } -// } template<> void run_mha_fwd_<cutlass::bfloat16_t, 64>(Flash_fwd_params ¶ms, cudaStream_t stream) { run_mha_fwd_hdim64<cutlass::bfloat16_t>(params, stream); -}
\ No newline at end of file +} |