TensorFlow is back at Google I/O on May 14! Register now

tf.keras.distribution.TensorLayout

A layout to apply to a tensor.

tf.keras.distribution.TensorLayout(
    axes, device_mesh=None
)

This API is aligned with jax.sharding.NamedSharding and tf.dtensor.Layout.

Args
`axes`	tuple of strings that should map to the `axis_names` in a `DeviceMesh`. For any dimensions that doesn't need any sharding, A `None` can be used a placeholder.
`device_mesh`	Optional `DeviceMesh` that will be used to create the layout. The actual mapping of tensor to physical device is not known until the mesh is specified.

Attributes
`axes`
`device_mesh`

Attributes

axes

device_mesh

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates. Some content is licensed under the numpy license.

Last updated 2024-04-26 UTC.