Mapping tensors without offloading to cpu

[amaleki (Amir Maleki)]

what if a is something else, like a=torch.tensor([20,10,30,40,0])?