Change underlying gpu memory address of Tensor

Hi, here’s the thing I want to do:
When I have a tensor_1 on GPU memory of device_1, I want to manually move the underlying GPU memory (ptr_1) of tensor_1 to another GPU device, with new device pointer (ptr_2) by using cuda-python package. Then I can get a valid GPU memory pointer (ptr_2). My question is, how can I change the underlying device ptr of tensor_1? Is it possible that I detach from tensor_1, so that I can get a seperate new tensor_2, then somehow I change the memory pointer of tensor_2 from ptr_1 to ptr_2, but everything else keep unchanged? I work on Python level.