pub fn moe_fp8_grouped_gemm_ptrtable_n128(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
a_fp8: DevicePtr,
b_packed_ptrs: DevicePtr,
b_scale_ptrs: DevicePtr,
scale2_vals: DevicePtr,
c: DevicePtr,
expert_offsets: DevicePtr,
sorted_token_ids: DevicePtr,
num_experts: u32,
n_out: u32,
k: u32,
max_m_tiles: u32,
stream: u64,
) -> Result<()>Expand description
FP8-A pointer-table grouped GEMM with transposed NVFP4 weights.
A must already be converted to FP8 E4M3. The launch shape mirrors
moe_w4a16_grouped_gemm_ptrtable_n128.