Miniforge
The default installation of Miniconda may have restrictive licensing on some packages (see https://stackoverflow.com/a/78766306 for some explanation). The drop-in replacement "miniforge" will set up a similar minimal conda environment that defaults to the "conda-forge" channel of open packages.
As the packages in Python distributions can change rapidly, it is difficult to have a single, system-wide installation that is useful to everyone. We will continue to install some basic ones in /sopt, but they are not likely to change after initial build.
To let users customize a minimal Python environment that won't disturb others, we suggest installing the "Miniforge" version of the Conda Python environment. (See https://conda-forge.org/download/ for additional info.)
Example Install 1) Python3-based conda installed in the default $HOME/miniforge3/ directory:
cd
wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
chmod u+x Miniforge3-Linux-x86_64.sh
./Miniforge3-Linux-x86_64.sh -b
Example Install A) Same as option1, but installed to /DFS-L/DATA/$group/$user/miniforge3/ directory, with link in home directory (allows installing very large numbers of packages that would make your home directory go over quota):
cd
wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
chmod u+x Miniforge3-Linux-x86_64.sh
mkdir /DFS-L/DATA/$(id -gn)/$USER/miniforge3
ln -s /DFS-L/DATA/$(id -gn)/$USER/miniforge3 miniforge3
./Miniforge3-Linux-x86_64.sh -bu
As of 23-January-2025, this installs Conda 24.11.3, which uses Python 3.12.8.
This uses about 169MB of disk space.
Once installed, you can set up the environment paths to your private version (with either install option) using:
ml miniforge/3/own
Updating will upgrade conda to 25.1.0
conda update --all
Installing new packages (e.g. numpy) within your miniforge3 directory is as simple as:
conda install numpy
Create a new environment for CPU-based PyTorch
conda create --name torch-CPU pytorch torchvision torchaudio cpuonly -c pytorch
Only one python module (miniforge/miniconda/anaconda/python/Intel-python) can be loaded at a time.