diff options
| author | Laurent Mazare <laurent.mazare@gmail.com> | 2024-07-15 20:37:36 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-07-15 20:37:36 +0200 |
| commit | 30cdd769f9404035235830e602ae01d50f782fb5 (patch) | |
| tree | dd9d8adcfce61fe40678f2967bbb25cecb1f679a /candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu | |
| parent | d74fbed3341f875fa81112e2f59565c464cd59d8 (diff) | |
| download | candle-30cdd769f9404035235830e602ae01d50f782fb5.tar.gz candle-30cdd769f9404035235830e602ae01d50f782fb5.tar.bz2 candle-30cdd769f9404035235830e602ae01d50f782fb5.zip | |
Update the flash attn kernels. (#2333)
Diffstat (limited to 'candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu')
| -rw-r--r-- | candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu b/candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu index 19b005ad..6df16b2c 100644 --- a/candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu +++ b/candle-flash-attn/kernels/flash_fwd_hdim128_fp16_sm80.cu @@ -5,6 +5,6 @@ #include "flash_fwd_launch_template.h" template<> -void run_mha_fwd_<cutlass::half_t, 128>(Flash_fwd_params ¶ms, cudaStream_t stream) { - run_mha_fwd_hdim128<cutlass::half_t>(params, stream); +void run_mha_fwd_<cutlass::half_t, 128, false>(Flash_fwd_params ¶ms, cudaStream_t stream) { + run_mha_fwd_hdim128<cutlass::half_t, false>(params, stream); } |