pytorch - use device inside 'with statement'

Question

Is there a way of running pytorch inside the context of a specific (GPU) device (without having to specify the device for each new tensor, such as the .to option)?

Something like an equivalent of the tensorflow with tf.device('/device:GPU:0'):..

It seems that the default device is the cpu (unless I'm doing it wrong):

with torch.cuda.device('0'):
   a = torch.zeros(1)
   print(a.device)

>>> cpu

Answer 1

Unfortunately in the current implementation the with-device statement doesn't work this way, it can just be used to switch between cuda devices.

---

You still will have to use the device parameter to specify which device is used (or .cuda() to move the tensor to the specified GPU), with a terminology like this when:

# allocates a tensor on GPU 1
a = torch.tensor([1., 2.], device=cuda)

So to access cuda:1:

cuda = torch.device('cuda')

with torch.cuda.device(1):
    # allocates a tensor on GPU 1
    a = torch.tensor([1., 2.], device=cuda)

And to access cuda:2:

cuda = torch.device('cuda')

with torch.cuda.device(2):
    # allocates a tensor on GPU 2
    a = torch.tensor([1., 2.], device=cuda)

However tensors without the device parameter will still be CPU tensors:

cuda = torch.device('cuda')

with torch.cuda.device(1):
    # allocates a tensor on CPU
    a = torch.tensor([1., 2.])

To sum it up:

No - unfortunately it is in the current implementation of the with-device statement not possible to use in a way you described in your question.

---

Here are some more examples from the documentation:

cuda = torch.device('cuda')     # Default CUDA device
cuda0 = torch.device('cuda:0')
cuda2 = torch.device('cuda:2')  # GPU 2 (these are 0-indexed)

x = torch.tensor([1., 2.], device=cuda0)
# x.device is device(type='cuda', index=0)
y = torch.tensor([1., 2.]).cuda()
# y.device is device(type='cuda', index=0)

with torch.cuda.device(1):
    # allocates a tensor on GPU 1
    a = torch.tensor([1., 2.], device=cuda)

    # transfers a tensor from CPU to GPU 1
    b = torch.tensor([1., 2.]).cuda()
    # a.device and b.device are device(type='cuda', index=1)

    # You can also use ``Tensor.to`` to transfer a tensor:
    b2 = torch.tensor([1., 2.]).to(device=cuda)
    # b.device and b2.device are device(type='cuda', index=1)

    c = a + b
    # c.device is device(type='cuda', index=1)

    z = x + y
    # z.device is device(type='cuda', index=0)

    # even within a context, you can specify the device
    # (or give a GPU index to the .cuda call)
    d = torch.randn(2, device=cuda2)
    e = torch.randn(2).to(cuda2)
    f = torch.randn(2).cuda(cuda2)
    # d.device, e.device, and f.device are all device(type='cuda', index=2)