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).

Python function

as_layout

as_layout()

max.experimental.sharding.as_layout(layout)

source

Normalizes a declared layout.

Takes a declaration, never a value: a live tensor’s dims are whatever it currently holds, so reading a layout off one would fix every dimension to that. Anything holding a layout of its own is refused, and told to pass it.

A single-device max.graph type is accepted and coerced, so the graph types stay usable without appearing in any signature. A BufferType coerces to a BufferLayout, carrying its declaration across.

Parameters:

layout (object) – A TensorLayout or BufferLayout to take as given, or a single-device TensorType or BufferType.

Returns:

The equivalent layout.

Raises:

TypeError – If layout is a value rather than a declaration, or describes no layout at all.

Return type:

TensorLayout