requirements.txt only accounts for packages >= and does not impose a limitation

[guilhermehge]

I've been working with pyannote for a while and from time to time when I build a new image, some packages get updated and my code stops working due to incompatibilities. It just happened with pytorch. Pytorch just updated to version 2.1.0 (4h ago) and, when running my code, I got the following message: RuntimeError: cuDNN version incompatibility: PyTorch was compiled against (8, 9, 2) but found runtime version (8, 5, 0). PyTorch already comes bundled with cuDNN. One option to resolving this error is to ensure PyTorch can find the bundled cuDNN.one possibility is that there is a conflicting cuDNN in LD_LIBRARY_PATH.

The docker image I'm using is this one: nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu20.04

As a suggestion, I recommend that pyannote's requirements have a >= and < restrictions, just like faster_whisper does, e.g:

ctranslate2>=3.17,<4
tokenizers>=0.13,<0.15
onnxruntime>=1.14,<2

In this case, for torch, I believe the requirement should be:
`torch>=2.0.0,<=2.0.1, since 2.1 can break stuff.

I know that I can lock my environment so the versions do not change, but I believe it would be beneficial to the library in general to fix the versions that it was tested on.

[github-actions]

Thank you for your issue.
We found the following entry in the FAQ which you may find helpful:

• How can I improve performance?

Feel free to close this issue if you found an answer in the FAQ.

If your issue is a feature request, please read this first and update your request accordingly, if needed.

If your issue is a bug report, please provide a minimum reproducible example as a link to a self-contained Google Colab notebook containing everthing needed to reproduce the bug:

• installation
• data preparation
• model download
• etc.

Providing an MRE will increase your chance of getting an answer from the community (either maintainers or other power users).

Companies relying on pyannote.audio in production may contact me via email regarding:

• paid scientific consulting around speaker diarization and speech processing in general;
• custom models and tailored features (via the local tech transfer office).

This is an automated reply, generated by FAQtory

[hbredin]

Well, let's say that, in the past, people have complained about pyannote being too restrictive on dependency versions... so I am not sure what to do here to content everyone :)

One solution would be to improve continuous integration and unit tests.
Would you like to contribute this to the project?
Especially since you've been working with pyannote for a while (your words)... maybe it is time to give back :)

[guilhermehge]

Fair enough. It was just a suggestion anyway. I can just do a pip freeze on my side.

"For a while" I meant like a month working with it, not much though, I'm not quite familiar with the code, but yes, I can try contributing.