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 &quot;SIMD-group VSA kernel is unavailable&quot;)
return _launch_simd_block_sparse(kernel, q_tiled, k_tiled, v_tiled, block_idx, block_num, geometry, scale)</code></pre></details>