I have done AI/ML dev on Arch and Fedora and it was pretty awarding not to mention that i got first prize for my AI model 2 times in a streak but i doubt if i can do AI/ML dev on NixOS since i cant use pip.
That’s not true, you can, you just have to patch it. Also, I suggest moving to uv as it’s leagues better than pip. Here’s an example project I’m using that with: https://github.com/Michael-C-Buckley/cacheguard.
I use direnv + uv to handle all my deps in a painless, streamlined manner. I did not commit the .envrc but I have what i used referenced in .config/README.md. Pip could be similarly patched.
4 Likes
You can also use pip by creating a venv ( venv — Creation of virtual environments — Python 3.14.2 documentation )
3 Likes
i tried that when i was gonna use the script to train my model but that failed brutally
my system crashed , this was like a few months ago
Hmmm, that sounds like another issue, perhaps OOM killer striking?
idk man idk what happened
I made a short video about using Python via Nix. uv seems cute, but Nix solves a much larger surface area
AI/ML dev can mean many things. Using python packages not packaged in nix but installed with pip or other managers is possible. I have a per project shell, which creates venv where i can use pip in the same way as on any other distro. Only friction was that the shell needs to export LD library paths
Nixpkgs contains many AI/ML libraries already. Including CUDA,NVIDIA drivers, etc. I am very happy with using NixOS for this purpose.
Using agentic tooling works very well to package the missing ones with Nix. You can often just ask to generate a flake.nix for a given repository and ensure that it works. The key is to always provide instructions to test the result, and iterate until the tests pass.
I am doing most of my development in Python, and have barely touched imperative package managers in the last years.
1 Like