Dataloader segmentation fault

[mosharaf]

mosharaf@Mosharafs-MacBook-Pro:~/Work/FedScale/dataset|master ⇒  python
Python 3.7.12 | packaged by conda-forge | (default, Oct 26 2021, 05:57:50)
[Clang 11.1.0 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from torch.utils.data import DataLoader
[1]    26092 segmentation fault  python

[AmberLJC]

it is torch library, how can I fix it?

(fedscale) jiachen_liu2@jiachen-liu2-xin:~/FedScale$ python
Python 3.7.12 | packaged by conda-forge | (default, Oct 26 2021, 06:08:53) 
[GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from torch.utils.data import DataLoader
>>> 

[fanlai0990]

Can not reproduce and it is Pytorch issue.

[mosharaf]

Installing torch through pip and then import torch in python 3.9 doesn't give any error. The same in the fedscale environment segfaults. There is a torch/python versioning problem in our environment.

[AmberLJC]

It also may be some issues with Mac, because it works fine on the server in current fedscale environment

[mosharaf]

Yes, I'm talking about Mac.

Installing torch through pip and then import torch in python 3.9 doesn't give any error. The same in the fedscale environment segfaults. There is a torch/python versioning problem in our environment.

[mosharaf]

FWIW, submitting jobs through works in the fedscale environment as well. Only running python in the interactive mode doesn't