Whenever I call autograds backward my script never terminates. backwardis not per se blocking, all lines after it are still executed, the script just does not terminate. It appears that there is some sort of working thread in the background which hangs, but I was not able to find any information about it.
I originally encountered the problem while training neural networks, however I eventually found a very short example with the same behavior:
import torch
x = torch.randn(3, requires_grad=True)
y = x * 2
print(y)
gradients = torch.tensor([0.1, 1.0, 0.0001], dtype=torch.float)
y.backward(gradients)
print(x.grad)
print("all done")
When I remove the backward line, the script finishes as expected. Otherwise I see a process call python in the task manager, if I terminate it by hand, the script execution also terminates.
I installed pytorch on Windows 7 using conda (conda create --name grad_test pytorch -c pytorch) in the most recent, stable version (python 3.7, pytorch 1.2.0).
