When I run this code, I found this behavior.
import torch
x = torch.Tensor([1, 2, 3])
x = x.cuda(1) # some memory on GPU id 1 is allocated
x = x.tolist() # some memory on GPU id 0 is allocated
Is this expected or not ?
When I run this code, I found this behavior.
import torch
x = torch.Tensor([1, 2, 3])
x = x.cuda(1) # some memory on GPU id 1 is allocated
x = x.tolist() # some memory on GPU id 0 is allocated
Is this expected or not ?