Best way to make dependencies optional for an specific arch?

[JorgeRuizITCL]

Whenever I try to use Poetry with an Nvidia Jetson device, I end up modifying the local pyproject.toml file to prevent Poetry from breaking the default python installation.

The main issue with Poetry on these devices is that the docker runtime images come with preinstalled libraries typically compiled for ARM devices with CUDA support. These libraries are not published on PyPI and often need to be downloaded and installed manually from an specific repo.

To avoid this problem, I remove the dependency from the pyproject.toml file and then I install everything else in the default environment. By default, Poetry would uninstall the pre-installed dependency with CUDA support and install the locked version that runs on CPU.

Is there a way to skip the installation of this specific dependency based on the architecture or Linux version?