Legitimate way to move module or tensor to GPU?

[jdhao]

I have seen two ways to move module or tensor to GPU:

• Use the cuda() method
• Use the to() method

Is there any difference between these two methods in terms of moving module or tensor to GPU? Which method is preferred?

[SimonW (Simon Wang)]

@jdhao

Is there any difference between these two methods in terms of moving module or tensor to GPU?

No

@jdhao

Which method is preferred?

.to is preferred because you can define a device object and use it everywhere.