pytorch debugging timeout with PyCharm

Question

I have a frustrating problem, where I cannot debug my pytorch code while in Pycharm.
While trying to inspect (breakpoint, then print e.g.) the code below, I receive a "Loading time out"

import torch

    tensors = []
    num_tensors = 16
    shape = (1, 3, 512, 512)
    for i in range(num_tensors):
        tensors.append(torch.zeros(shape))

I saw this[1,2] posts, set variable loading policy to syncronyous, disabled Qt debugger options, and all the options specified, But I believe there is something basic I'm missing.
pycharm 2019.2.5, happens both in python2 and python3.

I have the same problem. It does run fine, but the problem occurs when we want to debug. — Emile D., Mar 13 '20 at 13:06

score 5 · Answer 1 · answered Apr 07 '20 at 11:47

5

Are you using DataLoader? If yes, you can try reducing the num_workers to 0.

answered Apr 07 '20 at 11:47

JasonWayne

1,724
1
19
16

score 0 · Answer 2 · answered Mar 13 '20 at 14:02

0

I had this problem too. Both with the Pycharm version 2019.3.3, 2019.3.4 and 2019.2.6. I had to revert back to 2018.3.7 where it was working. I may have some parameters that caused the conflicts. It doesn't explain the underlying reason though...

answered Mar 13 '20 at 14:02

Emile D.

602
2
11
20

pytorch debugging timeout with PyCharm

2 Answers2