For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

build_block_scaled_configs

def build_block_scaled_configs[a_type: DType, b_type: DType, c_type: DType, sfa_dtype: DType, sfb_dtype: DType, N: Int, K: Int, transpose_b: Bool = True]() -> Set[BlockScaledMatmulConfig[a_type, b_type, c_type, sfa_dtype, sfb_dtype, transpose_b]]

Build a set of BlockScaledMatmulConfig instances by sweeping M from 8 to 8192.

Parameters:

a_type (DType): DType of the A (left) operand elements.
b_type (DType): DType of the B (right) operand elements.
c_type (DType): DType of the output matrix elements.
sfa_dtype (DType): DType of the A operand block scaling factors.
sfb_dtype (DType): DType of the B operand block scaling factors.
N (Int): The N dimension (output columns) of the matmul.
K (Int): The K dimension (contraction axis) of the matmul.
transpose_b (Bool): Whether the B operand is stored transposed (defaults to True).

Returns:

Set[BlockScaledMatmulConfig[a_type, b_type, c_type, sfa_dtype, sfb_dtype, transpose_b]]: A set of unique BlockScaledMatmulConfig instances covering the swept M range.