CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutSign UpSign In
pytorch

Real-time collaboration for Jupyter Notebooks, Linux Terminals, LaTeX, VS Code, R IDE, and more,
all in one place.

GitHub Repository: pytorch/tutorials
 Path: blob/main/recipes_source/recipes/changing_default_device.py
Views: 713
1
"""

2
Changing default device

3
=======================

4


5
It is common practice to write PyTorch code in a device-agnostic way,

6
and then switch between CPU and CUDA depending on what hardware is available.

7
Typically, to do this you might have used if-statements and ``cuda()`` calls

8
to do this:

9


10
.. note::

11
   This recipe requires PyTorch 2.0.0 or later.

12


13
"""

14
import torch

15


16
USE_CUDA = False

17


18
mod = torch.nn.Linear(20, 30)

19
if USE_CUDA:

20
    mod.cuda()

21


22
device = 'cpu'

23
if USE_CUDA:

24
    device = 'cuda'

25
inp = torch.randn(128, 20, device=device)

26
print(mod(inp).device)

27


28
###################################################################

29
# PyTorch now also has a context manager which can take care of the

30
# device transfer automatically. Here is an example:

31


32
with torch.device('cuda'):

33
    mod = torch.nn.Linear(20, 30)

34
    print(mod.weight.device)

35
    print(mod(torch.randn(128, 20)).device)

36


37
#########################################

38
# You can also set it globally like this: 

39


40
torch.set_default_device('cuda')

41


42
mod = torch.nn.Linear(20, 30)

43
print(mod.weight.device)

44
print(mod(torch.randn(128, 20)).device)

45


46
################################################################

47
# This function imposes a slight performance cost on every Python

48
# call to the torch API (not just factory functions). If this

49
# is causing problems for you, please comment on

50
# `this issue <https://github.com/pytorch/pytorch/issues/92701>`__

51


52