TensorFlow · Function

tf.reshape

tf.reshape returns a tensor with a requested shape while preserving all values and their order. The total number of elements must stay unchanged; one shape dimension may be inferred with -1.

How it works

tf.reshape creates a new tensor view with the requested shape while preserving the input values and their order. The requested dimensions must contain the same total number of elements; one dimension may be -1 so TensorFlow can infer it, and [] converts a one-element tensor to a scalar.

Syntax

tf.reshape(tensor, shape, name=None)

Parameters

Name	Type	Description
`tensor`	`Tensor`	Input Tensor to reshape.
`shape`	`Tensor`	Output shape; must use int32 or int64 values.
`name`	`string`	Optional name for the operation.

Returns

A Tensor with the same type as tensor.

Key points

Use tf.reshape to give a tensor a different shape.
The result preserves every input value and its sequence.
The shape parameter determines the result dimensions.
The operation is efficient because it can reuse the input data buffer.
Passing [-1] is a concise way to flatten a tensor.
The returned Tensor keeps the input tensor's data type.

Common mistakes

The new shape must preserve both the element count and value order; reshape is not an axis-reordering operation.
Only one output dimension may be inferred with -1.
An empty shape is valid only when the tensor contains exactly one element.

Example: Reshape a tensor

import tensorflow as tf

x = tf.constant([[1, 2, 3], [4, 5, 6]])
y = tf.reshape(x, [3, 2])
print(y.numpy())
# [[1 2]
#  [3 4]
#  [5 6]]

Creates a two-dimensional tensor with the same six values in their original order.

When to use it

Prepare model inputs

Convert a flat batch or feature vector into the dimensions expected by a model layer.

Combine adjacent axes

Collapse several compatible dimensions into one while retaining row-major value order.

Flatten data

Turn a tensor into a single vector for downstream processing or inspection.

Create a scalar

Represent a one-element tensor as a scalar using an empty shape.

Interview questions

QWhat does tf.reshape do?

tf.reshape changes a tensor's shape without changing its values or their order. The output has the same element type as the input, and the total element count must remain constant.

Follow-up: How would you exchange two axes? Use tf.transpose; reshape itself does not rearrange axes.

QHow does -1 work in the shape argument?

The shape argument defines the output dimensions and may be an int32 or int64 tensor. One dimension can be -1, in which case TensorFlow computes it from the unchanged total size.

Follow-up: What does shape [-1] produce? It flattens the tensor into one dimension.

QWhat does tf.reshape return and why is it efficient?

tf.reshape returns a Tensor with the same type as tensor. It can reuse the underlying data buffer because it preserves element order and count, making the operation fast regardless of tensor size.

Follow-up: Does that mean no tensor object is created? No; the operation returns a new Tensor while potentially reusing the underlying storage.

tf.shape
tf.transpose
tf.expand_dims
tf.squeeze

Source: Reshape. Checked 2026-10-15.