uv: Fast Alternative to conda and venv

10 of march of 2025

Disclaimer: This post has been translated to English using a machine translation model. Please, let me know if you find any mistakes.

So far I have been managing my environments with conda. But for a while now I've been reading a lot about poetry, but especially about uv. What are the advantages of uv? Speed. uv is implemented in Rust, so it manages environments and installs packages extremely quickly.

The following table shows the speed difference between different package managers. Source: LLMs-from-scratch/setup/01_optional-python-setup-preferences/native-uv.md

Command	Speed
`conda install <pkg>`	slow
`pip install <pkg>`	up to 10 times faster than the previous version
`uv pip install <pkg>`	between 5 and 10 times faster than the previous
`uv add <pkg>`	between 2 and 5 times faster than the previous

Looking at the table, it's definitely worth using uv. So let's see how to create an environment and install packages with uv.

Repository download

As I said, I am using LLMs-from-scratch/setup/01_optional-python-setup-preferences[native-uv.md as the source, so let's download the repository, install the proposed environment, and see how to run a script.

We use --depth 1 to download only the latest commit of the repository and make it clone faster, we are not interested in the history.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		git clone https://github.com/rasbt/LLMs-from-scratch.git --depth 1
	
	Copied

>_ Output

			
				Cloning into 'LLMs-from-scratch'...
remote: Enumerating objects: 260, done.
remote: Counting objects: 100% (260/260), done.
remote: Compressing objects: 100% (226/226), done.
remote: Total 260 (delta 61), reused 121 (delta 22), pack-reused 0 (from 0)
Receiving objects: 100% (260/260), 1.64 MiB | 6.94 MiB/s, done.
Resolving deltas: 100% (61/61), done.

Now we are going to the repository that we have downloaded

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		cd LLMs-from-scratch
	
	Copied

Install `uv`

If we are on macOS or Linux, we can install using the command

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

If we are on Windows

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

Create environment

If we do an ls we can see that there is a file called pyproject.toml, this will be the file that uv will use to create the environment.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		ls
	
	Copied

>_ Output

			
				2025-03-10-uv.ipynb appendix-D          ch04                pyproject.toml
CITATION.cff        appendix-E          ch05                requirements.txt
LICENSE.txt         ch01                ch06                setup
README.md           ch02                ch07
appendix-A          ch03                pixi.toml

So let's see what the file has

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		cat pyproject.toml
	
	Copied

>_ Output

			
				[project]
name = "llms-from-scratch"
version = "0.1.0"
description = "Implement a ChatGPT-like LLM in PyTorch from scratch, step by step"
readme = "README.md"
requires-python = "&gt;=3.10"
dependencies = [
    "torch&gt;=2.3.0",
    "jupyterlab&gt;=4.0",
    "tiktoken&gt;=0.5.1",
    "matplotlib&gt;=3.7.1",
    "tensorflow&gt;=2.18.0",
    "tqdm&gt;=4.66.1",
    "numpy&gt;=1.26,&lt;2.1",
    "pandas&gt;=2.2.1",
    "pip&gt;=25.0.1",
]
[tool.setuptools.packages]
find = {}
[tool.uv.sources]
llms-from-scratch = { workspace = true }
[dependency-groups]
dev = [
    "llms-from-scratch",
]
[tool.ruff]
line-length = 140
[tool.ruff.lint]
exclude = [".venv"]
# Ignored rules (W504 removed)
ignore = [
    "C406", "E226", "E402", "E702", "E703",
    "E722", "E731", "E741"
]

As can be seen, there are data such as the name, version, etc., and the dependencies, which are the packages we are going to install.

To create the environment, we use the command uv sync, and we add the --dev flag to also install development dependencies and the --python flag to specify the version of Python we want to use.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		uv sync --dev --python 3.11
	
	Copied

>_ Output

			
				Using CPython 3.11.11
Creating virtual environment at: .venv
Resolved 160 packages in 175ms
Installed 139 packages in 1.46s
+ absl-py==2.1.0
+ anyio==4.8.0
+ appnope==0.1.4
+ argon2-cffi==23.1.0
+ argon2-cffi-bindings==21.2.0
+ arrow==1.3.0
+ asttokens==3.0.0
+ astunparse==1.6.3
+ async-lru==2.0.4
+ attrs==25.1.0
+ babel==2.17.0
+ beautifulsoup4==4.13.3
+ bleach==6.2.0
+ certifi==2025.1.31
+ cffi==1.17.1
+ charset-normalizer==3.4.1
...
+ tzdata==2025.1
+ uri-template==1.3.0
+ urllib3==2.3.0
+ wcwidth==0.2.13
+ webcolors==24.11.1
+ webencodings==0.5.1
+ websocket-client==1.8.0
+ werkzeug==3.1.3
+ wheel==0.45.1
+ wrapt==1.17.2

It has created the environment and installed the packages in a lightning-fast way

Moreover, if we run ls again now we will see a new folder called .venv, that is the folder for the virtual environment.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		ls -a
	
	Copied

>_ Output

			
				.                   CITATION.cff        ch02                pyproject.toml
..                  LICENSE.txt         ch03                requirements.txt
.git                README.md           ch04                setup
.github             appendix-A          ch05                uv.lock
.gitignore          appendix-D          ch06
.venv               appendix-E          ch07
2025-03-10-uv.ipynb ch01                pixi.toml

Add packages

If we want to add packages to our environment that are not in the pyproject.toml file, we can do so with the command uv add <pkg>.

For example, if we run cat pyproject.toml | grep dotenv we will see that the package python-dotenv is not installed.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		cat pyproject.toml | grep dotenv
	
	Copied

So we add the package

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		uv add dotenv
	
	Copied

>_ Output

			
				Resolved 162 packages in 92ms
Installed 2 packages in 5ms                                      ░░░░░░░░░░░░░░░░░░░░ [0/0] Installing wheels...
+ dotenv==0.9.9
+ python-dotenv==1.0.1

If we now run cat pyproject.toml | grep dotenv again, we will see that it has been added to the file.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		cat pyproject.toml | grep dotenv
	
	Copied

>_ Output

			
				    "dotenv&gt;=0.9.9",

This is very good because now with this new pyproject.toml file we can recreate the environment with the command uv sync on any other computer.

Running a script

Now that we have an environment, we can run a script in two ways, the first with uv run python <script>.py, which will activate the .venv environment and run the script.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		uv run python setup/02_installing-python-libraries/python_environment_check.py
	
	Copied

>_ Output

			
				[OK] Your Python version is 3.11.11
[OK] torch 2.6.0
[OK] jupyterlab 4.3.5
[OK] tiktoken 0.9.0
[OK] matplotlib 3.10.1
[OK] tensorflow 2.18.0
[OK] tqdm 4.67.1
[OK] numpy 2.0.2
[OK] pandas 2.2.3
[OK] psutil 7.0.0

However, if what we want is to run the script directly with python <script>.py, we need to activate the environment manually first.

	
		
			< >
			Input
		
		
			Python
			
		
	
	
		source .venv/bin/activate && python setup/02_installing-python-libraries/python_environment_check.py
	
	Copied

>_ Output

			
				[OK] Your Python version is 3.11.11
[OK] torch 2.6.0
[OK] jupyterlab 4.3.5
[OK] tiktoken 0.9.0
[OK] matplotlib 3.10.1
[OK] tensorflow 2.18.0
[OK] tqdm 4.67.1
[OK] numpy 2.0.2
[OK] pandas 2.2.3
[OK] psutil 7.0.0

Frequently asked questions

What do the --dev and --python flags do in uv sync --dev --python 3.11?

The --python 3.11 flag tells uv sync which Python version to build the virtual environment with (in the example it creates .venv using CPython 3.11.11), while --dev also installs the dependencies listed under the [dependency-groups] dev = [...] section of pyproject.toml, on top of the regular project dependencies. The resolve/install step is extremely fast — in the post's example it resolved 160 packages in 175ms and installed 139 in 1.46s.

How do you add a package that isn't listed in pyproject.toml using uv?

Run uv add <package>, e.g. uv add dotenv, which resolves and installs the package (plus its dependencies, such as python-dotenv) straight into the .venv environment. uv add also automatically appends the matching entry (e.g. "dotenv>=0.9.9") to the dependencies section of pyproject.toml, so anyone else can recreate that exact environment elsewhere with just uv sync.

Continue reading

Conda: Python Environment Management

Conda vs Pip comparison: manage Python packages and environments for data science. Learn Conda and Mamba for faster dependency resolution.

ChromaDB Python Tutorial: Queries & Dimension Error Fix

ChromaDB in Python: create collections, query embeddings, and fix InvalidDimensionException. Hands-on guide with query_e...

Hugging Face Tokenizers: Fast Encoding

Guide to Hugging Face Tokenizers library: transform text into structured data for AI models. Practical Python examples f...

Last posts -->

Have you seen these projects?

Gymnasia

LangGraph Deep Researcher

Tau

View all projects -->

>_ Available for projects

Do you have an AI project?

Let's talk.

maximofn@gmail.com

Machine Learning and AI specialist. I develop solutions with generative AI, intelligent agents and custom models.

Write me LinkedIn

Do you want to watch any talk?

Tomorrow's Agents: Deciphering the Mysteries of Planning, UX and Memory

AI agents, powered by LLMs, promise to transform applications. But are they simple executors today or future intelligent collaborators? To reach their...

Create your own Apple intelligence

Learn to create an IA system to execute efficiently on a device

Last talks -->

Do you want to improve with these tips?

Best practices building agents with Claude Code

Technical talk: skills, subagents, slash commands and MCPs in Claude Code

o1 prompt engineering

Create better prompts for o1 following an example

Memory profiler

See the memory usage of a script

Last tips -->

Use this locally

Hugging Face spaces allow us to run models with very simple demos, but what if the demo breaks? Or if the user deletes it? That's why I've created docker containers with some interesting spaces, to be able to use them locally, whatever happens. In fact, if you click on any project view button, it may take you to a space that doesn't work.