服务器 python 安装 - 本地或系统范围

Question

We have an Ubuntu server and our team is discussing whether to install a system-wide Python version with PyTorch etc or to create local installations per user (such as Conda). One of the problems is HDD space given to each user, as many local installations with all the dependencies might take some space.

Did you have similar dilemmas in your cases? Any opinions on this?

Answer 1

The Conda package cache location ( pkgs_dirs ) can be shared across users while still maintaining separate user-specific envs ( envs_dirs ). As long as the package cache and environment locations are on the same physical disk, then Conda will use hardlinks to minimize redundancy. See conda config --describe pkgs_dirs .