Function hc_head_lowrank 

pub fn hc_head_lowrank(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    streams: DevicePtr,
    w: &HcLowRank,
    y_out: DevicePtr,
    scratch: DevicePtr,
    num_tokens: u32,
    hidden_size: u32,
    hc_mult: u32,
    norm_eps: f32,
    stream: u64,
) -> Result<()>

2026-09-25: The model-level mixer (use_combine=False in the reference bench/qwen4_exp/ref/modeling_qwen4_exp.py): the same collapse with no injection vector, on the same path choice as hc_pre_lowrank. It is also the model’s final normalization: the config parser sets final_norm_identity (config/src/parsers/qwen4_exp.rs).