Installation error

[Evert-Homan]

Hi, on WSL2 Debian I followed

1. Setup script (recommended)

At the end of the installation I see the following error:

Testing the installation... Generating SMILES for P29275 (A2AR_HUMAN) using the pretrained model. Traceback (most recent call last): File "/home/evehom/pcmol/pcmol/generate.py", line 2, in <module> from pcmol.models.runner import Runner File "/home/evehom/pcmol/pcmol/__init__.py", line 1, in <module> from .models.runner import Runner File "/home/evehom/pcmol/pcmol/models/__init__.py", line 1, in <module> from .transformer import AF2SmilesTransformer, count_parameters File "/home/evehom/pcmol/pcmol/models/transformer.py", line 1, in <module> import torch File "/home/evehom/miniconda3/envs/pcmol/lib/python3.10/site-packages/torch/__init__.py", line 367, in <module> from torch._C import * # noqa: F403 ImportError: /home/evehom/miniconda3/envs/pcmol/lib/python3.10/site-packages/torch/lib/../../nvidia/cusparse/lib/libcusparse.so.12: undefined symbol: __nvJitLinkComplete_12_4, version libnvJitLink.so.12 **************

Please advise on how to resolve this. Thanks/Evert

[andriusbern]

Hi!
This seems like an error with your GPU, it might be not compatible or outdated. To test whether you can run the model at all on this device try running it on the CPU by:

python pcmol/generate.py --target P29275 --device cpu

[Evert-Homan]

Doing so gives me the same error:

(pcmol) evehom@everthodell:~/pcmol$ python pcmol/generate.py --target P29275 --device cpu
Traceback (most recent call last):
  File "/home/evehom/pcmol/pcmol/generate.py", line 2, in <module>
    from pcmol.models.runner import Runner
  File "/home/evehom/pcmol/pcmol/__init__.py", line 1, in <module>
    from .models.runner import Runner
  File "/home/evehom/pcmol/pcmol/models/__init__.py", line 1, in <module>
    from .transformer import AF2SmilesTransformer, count_parameters
  File "/home/evehom/pcmol/pcmol/models/transformer.py", line 1, in <module>
    import torch
  File "/home/evehom/miniconda3/envs/pcmol/lib/python3.10/site-packages/torch/__init__.py", line 367, in <module>
    from torch._C import *  # noqa: F403
ImportError: /home/evehom/miniconda3/envs/pcmol/lib/python3.10/site-packages/torch/lib/../../nvidia/cusparse/lib/libcusparse.so.12: undefined symbol: __nvJitLinkComplete_12_4, version libnvJitLink.so.12

This is on WSL2 Debian 12. NVIDIA RTX A500 GPU, NVIDIA driver 552.74 , Cuda 12.2

[andriusbern]

I updated the conda environment spec to be more suited for windows systems as well. I cannot test it myself, however you can try to remove the pcmol conda environment and then rerun the installation instructions (after pull/clone) to see if this is resolved.