Function quantize_weight_q4k 

pub fn quantize_weight_q4k(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input_bf16: DevicePtr,
    out_q4k: DevicePtr,
    nrows: u32,
    n_per_row: u32,
    stream: u64,
) -> Result<()>

Quantize bf16 weights [nrows, n_per_row] -> GGML block_q4_K (at model load).