Why Use Miniconda?

Conda is an extremely popular environment management system in data science. However, the de-facto installation method for Windows is Anaconda, which adds a lot of bloat. You could use conda from the CLI with Anaconda installed, but I’d prefer to not have the extra packages and GUI installed.

Luckily, Anaconda also includes a minimal version of conda, which is what we’ll be installing today.

1. Install Miniconda

Follow instructions from the official Miniconda installation guide.

I used the silent install instructions for Windows Powershell (reproduced below):

wget "https://repo.anaconda.com/miniconda/Miniconda3-latest-Windows-x86_64.exe" -outfile ".\miniconda.exe"
Start-Process -FilePath ".\miniconda.exe" -ArgumentList "/S" -Wait
del .\miniconda.exe

2. Add the following lines to the .bashrc file

The miniconda3/bin folder is called "Scripts" on Windows.

Miniconda should have overwritten or appended to your ~/.bashrc file. I appended these lines to add the miniconda Scripts directory to your path and source conda.sh.

export PATH="/c/Users/<WINDOWS_USERNAME>/miniconda3/Scripts:$PATH"
source /c/Users/<WINDOWS_USERNAME>/miniconda3/etc/profile.d/conda.sh

Extra Setup

Store environments on an external drive

Add the following to your ~/.condarc file (or make it if it doesn't exist):

envs_dirs:
  - E:/conda/envs # Replace this with whichever path you want!

Install Pytorch on Miniconda with CUDA support

1. Install the standalone CUDA Toolkit

Look up your GPU to determine which version(s) of CUDA are supported. At the time of writing, the CUDA Toolkit download link is here.

Install the toolkit and restart your computer.

2. Install Pytorch

As of October 2024, Pytorch no longer supports conda, so we need to install Pytorch and its sibling libraries with pip instead.

Ensure you've activated the conda environment you want to install, then you can use Pytorch's handy tool which can generate the command you need.

3. Test CUDA support

With your environment activated, start a python shell and run:

import torch

print(torch.cuda.is_available())