On which Linux distribution, and which version did you install this? [question]

[UnixJunkie]

Hello,

Alternatively, if a docker container or some VM image could
be made available, it would be useful since installation from sources
looks challenging.

Regards,
Francois.

[kentoabeywardane]

Hi Francois,

Our system uses the Ubuntu 22.04.5 (Jammy Jellyfish) distribution.
We do not currently have a docker container or VM image. Could you please try the installation process as suggested and let us know if you run into issues, first?

It can be roughly laid out into these steps:

1. Create conda environment with python 3.8
2. Install pytorch with the applicable cuda toolkit (to match your hardware) with conda
3. Install pytorch geometric with conda
4. Install e3nn, pytorch lightning, rdkit, xtb, open3d, h5py with pip

Best,
Kento

[UnixJunkie]

Since there are so many dependencies, some being non Python things, some systems run into glibc compatibility problems.
The conda .yml file just doesn't work for us.

[kentoabeywardane]

That conda .yml is very specific to our cluster, so I suggest using the installation lines if possible.

Nevertheless, could you expand on what your system is? In order to potentially make an image/container, it would require us to build it on a similar system.

Thanks,
Kento

[UnixJunkie]

A docker container based on Ubuntu would be fine.
I.e. start from the Linux distro where your installation is working.

[kentoabeywardane]

Apologies for the delay. I've added a Dockerfile and corresponding minimal shepherd_env.yml file to the docker/ directory. Please try this out and let me know if you run into any issues.