Terminate called after throwing an instance of 'std::bad_alloc' from importing torch_geometric

Question

I am writing in python and getting the error:

"terminate called after throwing an instance of 'std::bad_alloc'.
what(): std::bad_alloc.
Aborted (core dumped)"

After lots of debugging, I found out the source of the issue is:

import torch_geometric

I even created a file with just this line of code, and I still get the error.
I am running in a conda environment (4.10.3) I made sure that I installed torch_geometric while I was in the conda environment. I tried deleting and reinstalling, but this did not work.
I also tried deleting and reinstalling torch/cuda.
I googled the error, but only seemed to come up with issues in data allocation, but I'm not sure how this would be an issue, since I am just importing torch_geometric.

Any ideas?

score 5 · Answer 1 · answered Apr 28 '22 at 15:08

5

This problem is because of mismatched versions of pytorch. The current pytorch being used is 1.11.0, but when scatter and sparse were installed installed scatter and sparse, 1.10.1 were used:

pip install torch-scatter -f https://data.pyg.org/whl/torch-1.10.1+cu113.html.
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.10.1+cu113.html

So,torch-1.10.1 was used to install scatter and sparse, but torch-1.11.0 was the true version.

Simply doing:

pip uninstall torch-scatter
pip uninstall torch-sparse
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.11.0+cu113.html.
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.11.0+cu113.html

Resolves the issue.

answered Apr 28 '22 at 15:08

Carter Karl Falkenberg

113
7

works for torch_scatter but `import torch_sparse` results in `OSError: libtorch_cuda_cu.so: cannot open shared object file: No such file or directory`. – Jesse W. Collins Apr 29 '22 at 14:36
1

I had a different torch version, that was all. – Jesse W. Collins Apr 29 '22 at 16:39
Yes, you just need to make sure torch versions match. – Carter Karl Falkenberg Apr 29 '22 at 19:34
I encountered the same behavior by mixing conda-forge with other conda channels. The docs for [pyg](https://pytorch-geometric.readthedocs.io/en/latest/install/installation.html) and [torch](https://pytorch.org/get-started/locally/) specify other channels, not conda-forge. – eric_kernfeld Mar 31 '23 at 19:58

Terminate called after throwing an instance of 'std::bad_alloc' from importing torch_geometric

1 Answers1