clone and detach

[wangjiawen2013]

Hi,
It is said that "We don't need to be worried about memory overhead because with cloning, the tensor's buffer isn't copied, and only a reference to it is increased." in burn's book, which is different from pytorch. In pytorch, clone will copy the tensor's buffer.

In pytorch, detach will share the buffer with the original tensor. what is it in burn ? Does detach share or copy the tensor buffer in burn ?

There are no detailed info on clone and detach in burn's book, so we want to know are they the same between pytorch and burn.

[laggui]

tensor.detach is used to detach the tensor from the current autodiff graph and creates a new leaf tensor.

The tensor is not copied.