Clone and detach in v0.4.0

2

We’ll provide a migration guide when 0.4.0 is officially released. Here are the answers to your questions:

  1. tensor.detach() creates a tensor that shares storage with tensor that does not require grad. tensor.clone()creates a copy of tensor that imitates the original tensor's requires_grad field.
    You should use detach() when attempting to remove a tensor from a computation graph, and clone as a way to copy the tensor while still keeping the copy as a part of the computation graph it came from.

  2. tensor.data returns a new tensor that shares storage with tensor. However, it always has requires_grad=False (even if the original tensor had requires_grad=True

  3. You should try not to call tensor.data in 0.4.0. What are your use cases for tensor.data?

  4. tensor.clone() makes a copy of tensor. variable.clone() and variable.detach() in 0.3.1 act the same as tensor.clone() and tensor.detach() in 0.4.0.

99 Likes