get_mxfp8_scale_offset

nvmath.linalg.advanced.helpers.matmul.get_mxfp8_scale_offset(

operand_or_shape: torch.Tensor | tuple[int, ...],

index: tuple[int, ...] | tuple[torch.Tensor, ...],

axis: Literal[-1, -2] | None = None,

) → int | torch.Tensor

Computes offset of a scale in the 1D interleaved scales tensor, applied to element operand[index].

Deprecated since version 0.9.0: Please use get_block_scale_offset() instead.