Function paged_decode_attn_nvfp4 

pub fn paged_decode_attn_nvfp4(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    q: DevicePtr,
    k_cache: DevicePtr,
    v_cache: DevicePtr,
    output: DevicePtr,
    block_tables: DevicePtr,
    seq_lens: DevicePtr,
    max_blocks_per_seq: u32,
    num_seqs: u32,
    num_q_heads: u32,
    num_kv_heads: u32,
    head_dim: u32,
    block_size: u32,
    inv_sqrt_d: f32,
    q_stride: u32,
    block_stride_bytes: u64,
    data_section_bytes: u64,
    stream: u64,
) -> Result<()>

Paged decode attention (NVFP4 KV cache, single/multi sequence).

Kernel: paged_decode_attn_nvfp4(Q, K_cache, V_cache, O, block_tables, seq_lens, max_blocks_per_seq, num_q_heads, num_kv_heads, head_dim, block_size, inv_sqrt_d, q_stride, block_stride_bytes, data_section_bytes) Grid: (num_q_heads, num_seqs, 1) Block: (256, 1, 1)