Request for a more detailed and stable environment.yml

[BBluetail]

Dear Dr Acciarini.I 'm having installation issues on Windows with the current environment.yml file. The PyTorch installation fails with DLL errors (fbgemm.dll not found). Could you provide a more detailed and stable environment configuration?

The current file lacks version specifications for key dependencies, which causes compatibility issues on Windows. A properly version-pinned environment.yml with explicit Python, PyTorch, and CUDA versions would greatly help.

Specifically, it would be helpful to have:

Python version pinned (3.10 is more stable than 3.12)

PyTorch version and CUDA version explicitly specified

All major dependencies version-pinned

Clear separation between conda and pip packages

Windows-specific dependencies included

This would make installation much more reliable across different systems.

[Sceki]

Hello @SoMBluetail ,

thank you for the detailed report and for outlining concrete suggestions.

Unfortunately, we have not extensively tested the installation on Windows, and we currently do not have CI/CD or systematic cross-platform testing in place. As a result, Windows-specific issues (such as the PyTorch DLL errors you mention) may indeed arise, and at the moment, we do not have the bandwidth to thoroughly investigate and maintain platform-specific environment configurations.

That said, your proposed solution (pinning Python, PyTorch, CUDA, and major dependencies, and clearly separating conda and pip packages) sounds reasonable and would likely improve installation stability. If you are willing to prepare a PR with an updated, version-pinned environment.yml, especially one that works reliably on Windows, I would be very happy to review it and merge it if appropriate.

If you do encounter additional concrete error messages or logs that you think are generally useful to document, feel free to share them as well.