Function moe_weighted_sum_blend_fp8_grouped 

pub fn moe_weighted_sum_blend_fp8_grouped(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    output: DevicePtr,
    expert_out: DevicePtr,
    expert_weights: DevicePtr,
    token_to_perm: DevicePtr,
    shared_out: DevicePtr,
    input: DevicePtr,
    gate_weight: DevicePtr,
    hidden: u32,
    top_k: u32,
    k: u32,
    num_tokens: u32,
    stream: u64,
) -> Result<()>

2026-09-25: Grouped blend, one blockIdx.y per token. expert_out rows are sorted positions; token_to_perm[token * top_k + k] maps a slot to its row. gate_weight may be NULL (ungated shared expert).