pub fn gdn_decode_wy4_woa(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
h_table: DevicePtr,
query: DevicePtr,
key: DevicePtr,
value: DevicePtr,
gate: DevicePtr,
beta: DevicePtr,
output: DevicePtr,
stash: DevicePtr,
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,
stash_seq_floats: u32,
engaged_flag: DevicePtr,
stream: u64,
) -> Result<()>Expand description
Write-on-accept K=4 verify (gated_delta_rule_wy4_woa): byte-identical
output to gdn_decode_wy4, writes NO state, stashes the per-row update
terms for gdn_wy4_fold. Table form only. Grid (num_v_heads, batch), 128.