PyTorch .remainder()

In PyTorch, the .remainder() function computes the element-wise remainder of division between two tensors or between a tensor and a scalar. The result always has the same sign as the divisor, unlike .fmod(), which matches the sign of the dividend. This operation works with both integer and floating-point tensors.

Syntax

torch.remainder(input, other)

Parameters:

• input (Tensor): The input tensor containing dividend values.
• other (Tensor or Number): The divisor. It can be a scalar or another tensor of compatible shape.
• out (Tensor, optional): The output tensor to store the result.

Return value:

Returns a tensor containing the element-wise remainder of the division.

• If other is a scalar, the operation is applied using the same divisor for all elements.
• If other is a tensor, element-wise division is performed.

Example 1: Divide a 1D Tensor by an Integer

This example computes the remainder of each element in x when divided by 2, keeping the sign consistent with the divisor:

import torch

x = torch.tensor([-3, -4, -1, -6, 4, 7, 8])
print(torch.remainder(x, 3))

Copy to clipboard
Copy to clipboard

The output of this code is:

tensor([0, 2, 2, 0, 1, 1, 2])

Copy to clipboard
Copy to clipboard

Example 2: Divide a 2D Tensor by an Integer

In this example, each negative number wraps around within the range [0, 4) since the remainder must match the divisor’s sign:

import torch

A = torch.tensor([[ 1,  2,  3],
                  [-1, -2, -3]])
print(torch.remainder(A, 4))

Copy to clipboard
Copy to clipboard

The output of this code is:

tensor([[1, 2, 3],
        [3, 2, 1]])

Copy to clipboard
Copy to clipboard

Example 3: Divide a Tensor by Another Tensor

This example demonstrates element-wise remainder calculation between two tensors of the same shape:

import torch

num = torch.tensor([ 3, -3,  3, -3], dtype=torch.int32)
den = torch.tensor([ 2,  2, -2, -2], dtype=torch.int32)
print(torch.remainder(num, den))

Copy to clipboard
Copy to clipboard

The output of this code is:

tensor([ 1,  1, -1, -1])

Copy to clipboard
Copy to clipboard

Note: If input and other don’t share the same shape, PyTorch tries automatic size expansion (broadcasting). Dimensions match when they’re equal or one of them is 1 (aligned from the right). If no match is possible, a size-mismatch error is raised.

Example 4: Divide a Floating-Point Tensor by a Number

This example calculates remainders for floating-point values, preserving the sign of the divisor:

import torch

xf = torch.tensor([-7.5, 7.5, 5.0])
print(torch.remainder(xf, 4.0))

Copy to clipboard
Copy to clipboard

The output of this code is:

tensor([0.5, 3.5, 1.0])

Copy to clipboard
Copy to clipboard