Installation with latest torch

[shenoynikhil]

Installing this package with recent versions of torch results in package conflicts. I am using the following yml file to install my environment.

name: custom_env
channels:
  - conda-forge
  - pytorch
  - nvidia
  - pyg

dependencies:
  # standard stuff
  - python >=3.8
  - pip

  # ML
  - e3nn
  - pytorch::pytorch
  - pytorch::pytorch-cuda==12.1
  - pyg::pyg

  # MD
  - openmm
  - openmm-ml
  - openmm-torch

I think it comes from torchani's incompatibility with recent torch versions. Any suggestions on how this could be potentially fixed?

[JMorado]

Hi @shenoynikhil, What happens if you use the following YAML file:

name: custom_env
channels:
  - conda-forge

dependencies:
  # standard stuff
  - python >=3.8
  - pip
  - cudatoolkit

  # ML
  - e3nn
  - pytorch

  # MD
  - openmm
  - openmm-ml
  - openmm-torch

Assuming you're on x86_64, this resolves to the latest packages on my end.

[JMorado]

It seems the package causing the conflicts was PyG. It works fine if you install it afterwards via pip, as described here.

[peastman]

Combining packages from conda-forge with any other channel often doesn't work correctly.

[shenoynikhil]

@JMorado That seems to resolve the issue!