uv workspace

uvws is a way to get something like a conda base env, but for uv. uv is far from being a full replacement for conda, but if you only need python packages then with uvws you may find it sufficient.

Setup

First, install uv if you haven't already, using the official script:

curl -LsSf https://astral.sh/uv/install.sh | sh

If you've already installed uv, you might want to ensure it's up to date:

uv self update

Now you're ready to install uvws. First, fork the repo. Then clone your fork (replace YOURNAME below of course)--I clone it into my home dir for convenience--and create and activate the uvws environment:

cd
git clone [email protected]:YOURNAME/uvws.git
cd uvws
uv sync
source .venv/bin/activate

You will probably want to have the env auto-activated in all your shells, so run (modifying the location and shell rc file name as needed):

echo source ~/uvws/.venv/bin/activate >> ~/.bashrc

Usage

To pip install packages, just prepend the using pip install with pip, for instance:

uv pip install torch torchvision torchaudio

A big benefit of uvws is that you can automatically setup all your machines with the same packages. To do so, instead of using uv pip install, use uv add; this also does an install, but also adds the package to uvws's pyproject.toml, so when you clone your forked repo and uv sync on another machine, you'll get the same packages.

cd ~/uvws
uv add nbclassic # or whatever package you want

You also probably want to get an editable install of packages you're working on. To do so, first edit repos.txt to list each package you want to clone and editable install; each line should be in the format org_or_user/repo. Then to clone them all in parallel (auto skipping any you already have), run:

./clone.sh 

To add these as editable workspace packages, run:

./addnew.py 

At any point you can quickly see the git status for all repos, using:

./status.sh

And, you can do a git pull of all in parallel using:

./update.sh

If at any point you want to reset the env entirely, run:

cd ~/uvws
# uncomment to run this:
# rm -rf .venv uv.lock 

You can also delete you pyproject.toml if you wish to reset your git add and ./addnew.py changes. Replace it with pyproject.tmpl to get the fresh version (i.e cp pyproject.tmpl pyproject.toml)

Conda alternative

If you need conda (e.g because you'd like to be able to switch CUDA environments easily) but also want to use uv sync, check out this clever technique: warner-benjamin/uv-tools.