Function conv1d_update 

pub fn conv1d_update(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    conv_state: DevicePtr,
    input: DevicePtr,
    weight: &DenseWeight,
    output: DevicePtr,
    d_inner: u32,
    d_conv: u32,
    batch_size: u32,
    stream: u64,
) -> Result<()>

Causal conv1d update (decode step, supports batched sequences).

Kernel: causal_conv1d_update(conv_state, new_input, weight, bias, output, batch, dim, d_conv) Grid: (ceil(dim/256), batch, 1) Block: (256, 1, 1)

For batch > 1, conv_state and input must be contiguous [batch, …].