other.md

• Other notes
  • Torch tensors

Other notes

Torch tensors

Create torch tensor from python data

some_integers = torch.tensor((2, 3, 5, 7, 11, 13, 17, 19))
print(some_integers)

This does not copy, it is just a label

a = torch.ones(2, 2)
b = a

To copy:

a = torch.ones(2, 2)
b = a.clone()

Moving to CPU, GPU

• To do computing tensors must be in the same device

if torch.cuda.is_available():
    print('We have a GPU!')
else:
    print('Sorry, CPU only.')

device = torch.device("cpu")

if torch.cuda.is_available():
    my_device = torch.device('cuda')
else:
    my_device = torch.device('cpu')
print('Device: {}'.format(my_device))

x = torch.rand(2, 2, device=my_device)
print(x)

Moving to another device

y = torch.rand(2, 2)
y = y.to(my_device)

Devices:

• meta device: Tensor without any data attached to it