Conda (Anaconda/Miniconda/Miniforge)

Basics

About Anaconda, Miniconda and Miniforge

The Conda installer is a package manager and environment manager for Python, R, and other languages. It comes in 3 flavors:

  • Anaconda: The Anaconda (company) driven Conda installer with hundreds of scientific packages included.

  • Miniconda: The Anaconda (company) driven minimalistic Conda installer including only Python by default.

  • Miniforge: The community (conda-forge) driven minimalistic Conda installer using the conda-forge channel by default.

Zurada only offers support for Miniforge as it works just like Anaconda or Miniconda but is lightweight and specifically built to support open-source and customizable environments. Compared to the other 2, it:

  • is smaller than Anaconda (no preinstalled packages).

  • does not rely on the default Anaconda channel, which can sometimes have licensing issues.

  • uses conda-forge as its default channel—a large, community-maintained repository of conda packages.

What is a Conda environment?

A Conda environment is a self-contained directory that includes:

  • A specific version of Python or R

  • Specific versions of packages (like NumPy, TensorFlow, etc.)

  • Any other dependencies

Each environment is isolated from others, which means packages in one environment don’t interfere with packages in another.

Why use multiple Conda environments?

  • Avoid Conflicts Between Packages. Different projects may need different package versions. Maintaining 1 or multiple environments per project can help in prototyping and or maintining different versions of your code. For example, Assume “Project A” requires TensorFlow 2.15 and “Project B” requires TensorFlow 1.14. If you installed both in the same environment, they would conflict. With Conda, you can do:

    conda create --name ProjectA python=3.10 tensorflow=2.15
conda create --name ProjectB python=3.7 tensorflow=1.14

  • Safe code prototyping and experimenting with new package versions. If you want to make sure your code works on the latest version of a package or you want to try a new exciting feature of the latest version of a package, you can create a new environment before updating.

  • Reproducibility. Environments can be exported and shared to others with conda env export > environment.yml and later imported by another party (or another machine) with conda env create -f environment.yml. For example, you could build an environment in your local workstation, then export it and rebuild it on the login node of the cluster.

Using Conda

Loading the miniforge3 module and the base environment

After logging into Zurada, you can start using Miniforge by loading the miniforge3 module with the command:

module load miniforge3/25.3.1-gcc-11.4.1

After executing this command, you should see your prompt change (it should now start with (base)).

Note

Sometimes the (base) environment does not activate automatically, to fix this simply use the command source activate base

Creating and activating an environment

Although you can install packages right after loading the miniforge3 module, which activates the base Miniforge environment for you, it is best to create and use named environments instead. These are isolated, easier to manage, and allow multiple setups under one base environment.

Miniforge supports both conda and mamba for managing environments. While both work the same way, mamba—written in C++—is faster and more memory-efficient than conda. We recommend using mamba for installing packages. However, if you have not run mamba init, you will need to use conda to activate or deactivate environments. For non-installation tasks like listing environments, both tools can be used interchangeably.

For example, to create an environment named my_env:

  1. Load the miniforge3 module by typing: module load miniforge3/25.3.1-gcc-11.4.1.

  2. Create the environment with mamba create --name my_env or conda create --name my_env.

Once you have created your environment, you can use it by executing conda activate environment_name_here. For example, conda activate my_env. The command prompt will then change from (base) to (my_env) to reflect this.

By default, your environments are stored at /home/user/.conda/envs/env_name. If you want to indicate a different prefix, you can use the --prefix option instead of the --name option. For example:

conda create --prefix /home/user/ProjectA/my_env

Installing packages with conda and mamba

Once you have activated an environment, you can install packages with the mamba install command or the conda install command. You can also control what version of a package you want installed based on the following constraints:

Conda installation constraint types

Constraint type

Syntax

Result

Fuzzy

python=3.11

3.11.0, 3.11.1, 3.11.2, etc.

Exact

python==3.11

3.11.0

Greater than or equal to

"python>=3.11"

3.11 or higher

OR

"python=3.11.1|3.11.3"

3.11.1, 3.11.3

AND

"python>=3.11,<3.13"

3.11, 3.12, not 3.13

For example,

mamba install python=3.11 # or conda install python=3.11
# NOTE: the quotation marks are necessary
# to protect the < and > symbols from the shell.
mamba install "python>=3.11.1" # or conda install "python>=3.11.1"

Installing packages with pip

Warning

If pip is not installed in your environment, either:

  • The shell won’t find pip, or

  • It will use a system-wide version, potentially failing due to lack of permissions to install packages system-wide.

Always check that pip is installed in the environment you activated with conda list -n env_name pip. The output should look similar to this:

# packages in environment at /home/user/.conda/envs/env_name:
#
# Name                    Version                   Build  Channel
pip                       25.1.1             pyh145f28c_0    conda-forge

We recommend using pip within an environment only if the package you need to install is not already available in conda-forge or another (open-source) conda channel. For example, newer versions of PyTorch are no longer installable via conda and thus must be installed via pip:

pip3 install torch torchvision torchaudio

Cloning an environment

Sometimes you may want to create a new conda environment based on an existing one. To do this:

  1. Activate the environment you want to clone.

  2. Export its package list to a YAML file with. mamba env export > environment.yml or conda env export > environment.yml

  3. Deactivate the environment.

  4. Create the new environment with mamba env create --name new_env -f environment.yml or conda env create --name new_env -f environment.yml.

Miscellaneous

Here are some useful conda commands users are encouraged to get familiar with:

Some useful Conda commands

Command

Meaning

conda create --name my_env PACKAGE1 PACKAGE2

Create a new environment named my_env, install packages PACKAGE1 and PACKAGE2

conda activate my_env

Activate environment my_env to use packages installed in it

conda deactivate

Deactivate a currently activated environment. You can only have one environment activated at once, so there is not need to name the environment you want to deactivate.

conda env list

Get a list of all available environments. Active environment is shown with *

conda env remove --name my_env

Remove environment my_env and everything in it

conda install PACKAGE

Install a package. You can also specify the version of the package by using the syntax PACKAGE=version. For example, conda install python=3.15.

conda update PACKAGE

Update a package

conda search PACKAGE

Search for package

conda config --add channels CHANNEL

Add a channel to your environment. This is especially helpful for those using bioconda (conda config --add channels bioconda)

As an example, a typical workflow for python looks like this:

module load miniforge3/25.3.1-gcc-11.4.1
# Create a custom environment. In this case, create
# an environment named "my_env" and install packages
# numpy and scipy
conda create --name my_env python=3.11 numpy scipy
# Activate the environment "my_env"
conda activate my_env
# Suppose you realized you need the pandas package.
# So, install it!
conda install pandas

Conda in a batch job

After creating a virtual environment, you can use it within a batch script. Any program you install within an environment will become available after said environment is activated. That means to effectively use a software installed via Conda in a batch job you need to load the miniforge3 module and then instruct the script to activate the desired environment.

For example, suppose you created an environment named projectA_pytorch, then you could use it as follows:

#!/bin/bash

#SBATCH --job-name=projectA_pytorch_job
#SBATCH --partition=cpu384g
#SBATCH --output=/home/user/slurm-%x-%j.out
#SBATCH --time=1:00:00
#SBATCH --nodes=1
#SBATCH --mem=12069M

module load miniforge3/25.3.1-gcc-11.4.1
conda activate projectA_pytorch

## Execute the python script that calls pytorch
python program.py

To see a more concrete example, see our documentation on how to launch jupyter through a batch job.

Conda in an interactive job

After creating a virtual environment, you can use it within an interactive job. Any program you install within an environment will become available after said environment is activated. That means to effectively use a software installed via Conda you need to load the miniforge3 module, activate the desired environment and run your program.

For example, suppose you created an environment named projectA_pytorch, then you could use it as follows:

srun --partition=cpusm01 --job-name=projectA_pytorch_job --time=1:00:00 --nodes=1 --mem=12069M --pty /bin/bash -i
module load miniforge3/25.3.1-gcc-11.4.1
conda activate projectA_pytorch

## Execute the python script that calls pytorch
python program.py

To see a more concrete example, see our documentation on how to launch jupyter through an interactive job.