In reply to @maupind:matrix.org
Any academics using NixOS as a method for reproducibility of analysis? I'm interested to see what your processes are and curious if there are any funding opportunities that can be used to develop Nix in this area

I'm making some progress, now I can build llama-cpp-python with cuda, but unfortunately at runtime, the python app doesn't see the GPU 😭
these lines from llama.cpp derivation caught my attention :

  # It's necessary to consistently use backendStdenv when building with CUDA support,
  # otherwise we get libstdc++ errors downstream.
  # cuda imposes an upper bound on the gcc version, e.g. the latest gcc compatible with cudaPackages_11 is gcc11
  effectiveStdenv = if cudaSupport then cudaPackages.backendStdenv else stdenv;
....
effectiveStdenv.mkDerivation (finalAttrs: {
....

in my case I'm not making a derivation but I'm building a python package using pkgs.python311.pkgs.buildPythonPackage
could it be the reason why gpu/cuda is not used at runtime ?

not making a derivation b

buildPythonPackage is a wrapper around mkDerivation