Pytorch separar
When we don't need a tensor to be traced for the gradient computation,
we detach the tensor from the current computational graph.
coder
When we don't need a tensor to be traced for the gradient computation,
we detach the tensor from the current computational graph.