torch.Tensor.detach()

Creator

Creator

Seonglae Cho

Created

Created

2024 Sep 12 23:6

Editor

Editor

Seonglae Cho

Edited

Edited

2025 Jul 20 20:59

Refs

Refs

torch.no_grad()

Pytorch Autograd

In PyTorch, detach() is a method that blocks gradient flow

It saves memory by removing a tensor from the computation graph, allowing you to manipulate it without affecting gradients

torch.Tensor.detach — PyTorch 2.4 documentation

This method also affects forward mode AD gradients and the result will never have forward mode AD gradients.

https://pytorch.org/docs/stable/generated/torch.Tensor.detach.html

Recommendations

/////////