Segmentation fault on import torch_geometric

[midannii]

🐛 Describe the bug

When I try to import torch_geometric I get a segmentation fault error.

I am just trying to import torch_geometric to get started and as such haven't even used any functions yet, since every time I try to import I get this error.

(I can import torch_scatter, torch_cluster, and torch_sparse)

Help me plz..

Environment

• PyG version: latest

• PyTorch version: 1.12.1

• OS: linux(centOS)

• Python version: 3.8.16

• CUDA/cuDNN version: 11.2

• How you installed PyTorch and PyG (conda, pip, source):
  - pytorch: conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=10.2 -c pytorch
  - pyg: conda install pyg -c pyg

• Any other relevant information (e.g., version of torch-scatter):
  - torch_scatter: 2.1.0
  - torch_cluster: 1.6.0
  - torch_sparse: 0.6.16
  - torch: 1.9.1 (checked in conda list)
  - torch_spline_conv: 1.2.1 (checked in conda list)

[EdisonLeeeee]

It appears that you are using different versions of PyTorch simultaneously, such as 1.9.1 and 1.12.1. Would you be able to uninstall one of them and try again?

[EulerPascal404]

PyTorch version: 1.12.1
torch: 1.9.1 (checked in conda list)

Yup, it seems you have 2 different versions of pytorch installed. You can uninstall pytorch with the following command: conda uninstall pytorch. Then try reinstalling pytorch - it should fix the error.