TensorFlow · Function

tf.ones

Creates a tensor of a specified shape whose elements are all one. The element type defaults to tf.float32 and can be selected with dtype.

How it works

tf.ones constructs a tensor from the requested shape, fills every element with one, and uses the requested element type. If no type is supplied, TensorFlow creates a float32 tensor; an optional name labels the operation, while layout can produce a DTensor with a specified layout.

Syntax

tf.ones(shape, dtype=tf.dtypes.float32, name=None, layout=None)

Parameters

Name	Type	Description
`shape`	`list, tuple, or 1-D int32 Tensor`	A list or tuple of integers, or a one-dimensional int32 Tensor describing the output dimensions.
`dtype`	`tf.DType`	Optional element type for the output; defaults to tf.float32.
`name`	`string`	Optional operation name.
`layout`	`tf.experimental.dtensor.Layout`	Optional DTensor layout; when supplied, the result is a DTensor with that layout.

Returns

A Tensor with all elements set to one (1).

Key points

Use tf.ones to construct a tensor uniformly filled with the scalar value one.
The output's element type is controlled by dtype, while its dimensions come from shape.
Every position in the returned tensor contains one rather than an uninitialized value.
name optionally assigns an operation name, which can help identify it in TensorFlow graphs.
The function returns a Tensor, not a Python list or scalar.

Common mistakes

shape must be a list or tuple of integers, or a one-dimensional int32 Tensor.
If dtype is omitted, the resulting tensor is float32.
Supplying layout changes the result to a DTensor using that layout.

Example: Create an integer tensor of ones

import tensorflow as tf

x = tf.ones([2, 3], dtype=tf.int32)
print(x)

Creates and prints a 2-by-3 integer tensor whose six elements are all one.

When to use it

Create a uniform tensor

Initialize a constant-valued input or buffer with known dimensions.

Choose an explicit dtype

Control numeric behavior for integer or floating-point computation.

Request a DTensor layout

Create distributed tensor data with an explicitly requested arrangement.

Interview questions

QWhat does tf.ones return, and what is its default dtype?

tf.ones creates a tensor with the supplied shape and sets every element to one. Its dtype defaults to tf.float32, but callers can request another dtype such as tf.int32.

Follow-up: How would you create ones matching an existing tensor? Use tf.ones_like, which derives the output shape from the input and can optionally use a new dtype.

QWhich forms can the shape argument take?

shape can be a list of integers, a tuple of integers, or a one-dimensional int32 Tensor. The returned Tensor has that shape and contains only ones.

Follow-up: What if the desired values are not all one? Use tf.fill with dimensions and a value, or tf.zeros when the value should be zero.

QWhat is the purpose of layout?

The layout argument is optional and accepts tf.experimental.dtensor.Layout. When provided, the result is a DTensor with that layout.

Follow-up: How can a layout be requested through DTensor APIs? DTensor documentation shows wrapping tf.ones with dtensor.call_with_layout and passing a layout and shape.

tf.ones_like
tf.zeros
tf.fill
tf.eye

Source: Ones. Checked 2026-10-15.