diff options
Diffstat (limited to 'candle-flash-attn/kernels/flash_fwd_hdim256_bf16_sm80.cu')
| -rw-r--r-- | candle-flash-attn/kernels/flash_fwd_hdim256_bf16_sm80.cu | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/candle-flash-attn/kernels/flash_fwd_hdim256_bf16_sm80.cu b/candle-flash-attn/kernels/flash_fwd_hdim256_bf16_sm80.cu index cb074a95..1b370141 100644 --- a/candle-flash-attn/kernels/flash_fwd_hdim256_bf16_sm80.cu +++ b/candle-flash-attn/kernels/flash_fwd_hdim256_bf16_sm80.cu @@ -1,9 +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, 256>(Flash_fwd_params ¶ms, cudaStream_t stream) { +template<> +void run_mha_fwd_<cutlass::bfloat16_t, 256>(Flash_fwd_params ¶ms, cudaStream_t stream) { run_mha_fwd_hdim256<cutlass::bfloat16_t>(params, stream); } |