--- title: minimax_h3_vsa_simd ---

SIMD-group block-sparse attention for MiniMax H3 VSA.

This backend supports tile sizes 64 and 128 and head dimension 128, evaluates the tile map selected by the reference router, and leaves unsupported shapes to the reference implementation.

fastvideo.mlx_runtime.minimax_h3_vsa_simd.simd_kernel_error

simd_kernel_error() -> str | None
Source
def simd_kernel_error() -> str | None:
    _simd_kernel()
    return _SIMD_KERNEL_ERROR

fastvideo.mlx_runtime.minimax_h3_vsa_simd.disable_simd_kernel

disable_simd_kernel(error: Exception) -> None

Remember a failed compile or execution so subsequent blocks use reference.

Source
def disable_simd_kernel(error: Exception) -> None:
    """Remember a failed compile or execution so subsequent blocks use reference."""
    global _SIMD_KERNEL, _SIMD_KERNEL_ERROR
    _SIMD_KERNEL_ERROR = str(error)
    _SIMD_KERNEL = None

fastvideo.mlx_runtime.minimax_h3_vsa_simd.simd_kernel_available

simd_kernel_available() -> bool
Source
def simd_kernel_available() -> bool:
    return _simd_kernel() is not None

fastvideo.mlx_runtime.minimax_h3_vsa_simd.simd_block_sparse

simd_block_sparse(q_tiled, k_tiled, v_tiled, block_idx, block_num, geometry, scale: float)

Block-sparse attention over tiled [S, H, D] using the reference tile map.

Source
def simd_block_sparse(
    q_tiled,
    k_tiled,
    v_tiled,
    block_idx,
    block_num,
    geometry,
    scale: float,
):
    """Block-sparse attention over tiled ``[S, H, D]`` using the reference tile map."""

    kernel = _simd_kernel()
    if kernel is None:
        raise RuntimeError(_SIMD_KERNEL_ERROR or "SIMD-group VSA kernel is unavailable")
    return _launch_simd_block_sparse(kernel, q_tiled, k_tiled, v_tiled, block_idx, block_num, geometry, scale)