Aten: read a CPU or CUDA Tensor that was allocated in Pytorch

Hi,
It it possible to use aten C++11 code to access a Tensor that was allocated on the GPU /CPU using PyTorch?
I am aware the opposite direction id possible.

Thanks,