无法将 torch.Tensor 传输到 torch.cuda.Tensor

Question

I use the following code to turn the torch.Tensor into torch.cuda.Tensor

import torch
import numpy as np

a = np.random.randn(1, 1, 2, 3)

t2 = torch.tensor(a)
t2 = t2.cuda()

t3 = torch.tensor(a, device=torch.device('cuda'))
print(type(t3), t3.device, type(t2), t2.device)

And its output is <class 'torch.Tensor'> cuda:0 <class 'torch.Tensor'> cuda:0

I expect the class is 'torch.cuda.Tensor'. I don't know why this happens. My torch version is 1.6.0 and the corresponding cuda version is 10.2

Answer 1

That is because there is no such class as torch.cuda.Tensor in PyTorch 1.6.0.

The t3.device and t2.device in print(type(t3), t3.device, type(t2), t2.device) prints cuda:0 which means your tensors are already on GPU.