Function gdn_decode_f32_strided_norm_snap 

pub fn gdn_decode_f32_strided_norm_snap(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    h_state: DevicePtr,
    query: DevicePtr,
    key: DevicePtr,
    value: DevicePtr,
    gate: DevicePtr,
    beta: DevicePtr,
    z_gate: DevicePtr,
    norm_weight: DevicePtr,
    output: DevicePtr,
    h_inter: DevicePtr,
    h_inter_seq_stride: u64,
    batch_size: u32,
    num_k_heads: u32,
    num_v_heads: u32,
    k_dim: u32,
    v_dim: u32,
    qk_stride: u32,
    v_stride: u32,
    gb_stride: u32,
    z_stride: u32,
    out_stride: u32,
    eps: f32,
    stream: u64,
) -> Result<()>

2026-09-25: super::gdn_decode_f32_strided_norm plus an inline h-state snapshot, for the batched-verify arm over batch_size sequences.

h_inter is the snapshot base for this token position, and sequences’ snapshots are h_inter_seq_stride FP32 elements apart. The caller measures that stride from the pool’s pointers; it does not follow from the head dims. NULL skips the snapshot.