How to Fix AttributeError: module ‘torch’ has no attribute ‘permute’

The AttributeError: module ‘torch’ has no attribute ‘permute’ error occurs because the torch module does not have an attribute called permute in the older version of the PyTorch library.

To fix the AttributeError: module ‘torch’ has no attribute ‘permute’ error, upgrade the PyTorch library to the latest version using the pip install –upgrade torch torchvision command.

To upgrade to a specific version of PyTorch, you can specify the version number in the command.

pip install --upgrade torch==1.7.0 torchvision==0.8.1

If you are using Anaconda, use the conda package manager to upgrade PyTorch:

conda install pytorch torchvision -c pytorch

Let’s write a simple PyTorch program in which “torch” has a “permute” attribute.

import torch

batch_size, length, dim = 32, 5, 200
inputs = torch.randn(batch_size, length, dim)
inputs = torch.permute(inputs, (1, 0, 2))
print(inputs)

Output

tensor([[[ 1.0026, -2.7712, -0.7772, ..., -0.6685, -0.4177, 0.6504],
 [ 0.2645, -0.4712, -1.5182, ..., 1.4168, 0.9946, -0.8884],
 [-0.6747, 0.4766, -0.3585, ..., 0.3867, 0.2957, -2.0025],
 ...,
 [ 0.5598, -0.5110, 0.8959, ..., 0.7500, 0.3673, -1.3500],
 [-1.2054, 0.4063, 0.0761, ..., 0.0068, -1.3447, -1.1213],
 [ 1.1985, 0.2248, 0.5502, ..., -1.3683, 1.4266, -0.7689]],

 [[-0.8764, 0.3507, -0.5025, ..., -1.1899, 1.2798, -0.1717],
 [ 1.2368, -0.1195, -0.1074, ..., -1.0518, 0.9270, -0.6185],
 [ 0.6959, -1.4283, 0.9314, ..., 1.2142, 2.2323, 0.2987],
 ...,
 [-1.0119, -0.7466, -1.4464, ..., 0.6538, 1.1523, -0.2886],
 [ 0.1741, 0.1881, 0.1870, ..., 0.0029, 0.7537, -0.0494],
 [ 0.3870, 0.1019, 0.1006, ..., -0.8216, 0.7249, 0.9346]],
 ...,
 [-1.2947, -0.6852, -0.6406, ..., -0.2920, -2.4572, 0.2402],
 [ 1.5241, -0.0057, -0.5739, ..., -1.5625, 1.0062, -0.3826],
 [-0.8617, -0.9933, 0.0583, ..., -1.8369, 0.0189, -0.6340]]]) 

Using an older PyTorch version, you can use the transpose() function instead of permute() to resolve this AttributeError.

If you are using the latest version of PyTorch, please check the spelling and capitalization of permute() function, and make sure that you have imported the correct module.

PyTorch implementation of permute attribute

import torch

x = torch.randn(2, 3, 5)
inputs = torch.permute(x, (2, 0, 1)).size()
print(inputs)

Output

torch.Size([5, 2, 3])

The torch.Tensor.permute() function reorders the dimensions of a tensor. It returns a view of the original tensor input with its dimensions permuted.

The permute() function accepts a sequence of integers as the argument, specifying the new order of the dimensions.

The “permute” function reorders the dimensions of a tensor, and it was added in PyTorch version 1.0.

That’s it for this troubleshooting article.