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

Mojo function

scale_align_elems

def scale_align_elems[dtype: DType]() -> Int

Scales per 16-byte unit, and so the granularity a flat scale TMA may START a copy at.

A TMA's global start address must be 16-byte aligned. In a flat 1 x N scale view a key index IS the innermost coordinate, so that requirement lands on the key index itself -- unlike a K tile, whose row is a whole depth-wide key and is therefore always aligned.

Parameters:

• ​dtype (DType): Scale element type.

Returns:

Int: Number of scales per 16-byte unit.