pub fn moe_expert_down_act_nvfp4_grouped(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
geometry: Fp8GroupedGeometry,
act: DevicePtr,
down: Nvfp4ExpertTables,
output: DevicePtr,
expert_offsets: DevicePtr,
active_experts: DevicePtr,
active_count: DevicePtr,
sh_act: DevicePtr,
sh_down: &QuantizedWeight,
sh_down_out: DevicePtr,
n: u32,
k: u32,
cap: u32,
num_tokens: u32,
stream: u64,
) -> Result<()>