tyro 1.0.6

CLI interfaces & config objects, from types

tyro

Documentation   •   pip install tyro

tyro.cli() is a tool for generating CLI interfaces from type-annotated Python.

We can define configurable scripts using functions:

https://github.com/user-attachments/assets/6f884313-6111-40a1-b9c7-7cd83d737296

Or instantiate configs defined using tools like dataclasses, pydantic, and attrs:

https://github.com/user-attachments/assets/edec520d-0c05-4547-8dc5-c2e211aadfb2

Other features include helptext generation, nested structures, subcommands, and shell completion. For examples and the API reference, see our documentation.

Why tyro?

1. Define things once. Standard Python type annotations, docstrings, and default values are parsed to automatically generate command-line interfaces with nice helptext.

2. Static types. Unlike tools dependent on dictionaries, YAML, or dynamic namespaces, arguments populated by tyro are better undestood by IDEs and language servers, as well as static checking tools like pyright and mypy.

3. Modularity. tyro supports hierarchical configurations, which make it easy to decentralize definitions, defaults, and documentation.

In the wild

tyro is designed to be lightweight for throwaway scripts, while improving maintainability for larger projects. Examples:

nerfstudio-project/nerfstudio	Open-source tools for neural radiance fields.
Sea-Snell/JAXSeq	Train very large language models in Jax.
kevinzakka/obj2mjcf	Interface for processing OBJ files for Mujoco.
blurgyy/jaxngp	CUDA-accelerated implementation of instant-ngp, in JAX.
NVIDIAGameWorks/kaolin-wisp	PyTorch library for neural fields.
autonomousvision/sdfstudio	Unified framework for surface reconstruction.
openrlbenchmark/openrlbenchmark	Collection of tracked experiments for reinforcement learning.
vwxyzjn/cleanrl	Single-file implementation of deep RL algorithms.
pytorch-labs/LeanRL	Fork of CleanRL, optimized using PyTorch 2 features.
pytorch/torchtitan	PyTorch-native platform for training generative AI models.
KwaiVGI/LivePortrait	Stitching and retargeting for portraits.
Physical-Intelligence/openpi	Open-source models for robotics.
haosulab/ManiSkill	GPU-parallelized simulation + learning for robot manipulation.
mujocolab/mjlab	Lightweight, modular abstractions for RL and sim-to-real robotics.
amazon-far/holosoma	Humanoid robotics framework for RL training and deployment.
MalcolmMielle/bark_monitor	Show your neighbor that your dog doesn't bark!

Alternatives

tyro is opinionated. If any design decisions don't make sense, feel free to file an issue!

You might also consider one of many alternative libraries. Some that we like:

• cappa offers a similar core feature set but with very different ergonomics. It looks polished and well-maintained!
• cyclopts and defopt has very comprehensive type annotation support and a heavier emphasis on subcommand generation.
• simple-parsing and jsonargparse provide deeper integration with configuration file formats like YAML and JSON.
• clipstick, which focuses on simplicity + generating CLIs from Pydantic models.
• datargs provides a minimal API for dataclasses.
• fire and clize support arguments without type annotations.

There are also some options that directly extend tyro:

• mininterface simultaneously generates GUI, TUI, web, CLI, and file-based program configuration.
• manuscript generates CLI interfaces from a simple block of configuration variables.

We also have some notes on tyro's design goals and other alternatives in the docs here.