[core] use kernels
to support _flash_3_hub
attention backend
#33177
Loading
kernels
to support _flash_3_hub
attention backend
#33177