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

get_scaling_kind

def get_scaling_kind[a_type: DType, scales_dtype: DType, SF_VECTOR_SIZE: Int]() -> UMMAKind

Selects the SM100 UMMA kind matching the operand and scale-factor types.

Maps the combination of operand dtype, scale-factor dtype, and scale-factor vector size to the corresponding UMMAKind used by SM100 block-scaled matmul instructions.

Parameters:

a_type (DType): Operand element type (uint8 for MXFP4/NVFP4, float8_e4m3fn for MXFP8).
scales_dtype (DType): Scale-factor element type.
SF_VECTOR_SIZE (Int): Number of elements each scale factor covers.

Returns:

UMMAKind: The UMMAKind matching the provided type combination.