flash-mla / flash_mla /flash_fwd_mla_fp16_sm90.cu
drbh
Migrated from kernels-community/flash-mla
0b53d3a unverified
raw
history blame contribute delete
147 Bytes
#include "flash_fwd_mla_kernel.h"
template void run_mha_fwd_splitkv_mla<cutlass::half_t, 576>(Flash_fwd_mla_params &params, cudaStream_t stream);