xtrax 0.2.0

High-performance composable JAX library

xtrax

A set of composable building blocks for JAX/Equinox training loops — engine + trainer orchestration, safety-checked steps, axis tiling strategies, inference-time sparsification, distributed/sharding helpers, streaming output callbacks, and orbax checkpointing — extracted from the author's research code.

Status

xtrax is alpha, experimental software built primarily for the author's personal research use. APIs may change without notice between releases; no backward-compatibility guarantees pre-1.0. Issues and pull requests are welcome, but support is best-effort — the project exists first to serve the author's own JAX training workflows.

Why xtrax?

• Composable architecture: Modular training pipeline designed for flexibility and extensibility
• Distributed training: Built-in support for multi-device and distributed JAX training
• Sparse operations: Efficient sparse model support with automatic sparsification
• Tiling and batching: Advanced tiling strategies for complex distributed scenarios
• Production-ready: Designed for real-world training workflows with checkpointing and safety features

Install

pip install xtrax

Requires Python 3.13+.

Quick Start

from xtrax import Trainer, Engine

# Create a simple training configuration
trainer = Trainer()

# Create an engine for distributed execution
engine = Engine()

# Define your loss and training logic
trainer.train(engine=engine)

For more details, see the documentation.

Features

Streaming outputs

Use BoundedCallbackHandler to capture and stream training outputs during execution. Integrate custom callbacks to monitor metrics and log results in real time.

Checkpointing

Save and restore training state with orbax checkpoints via save_checkpoint() and load_checkpoint(). Ensures resumable training across interruptions.

Documentation

Full API documentation and tutorials are available at https://xtrax.readthedocs.io.

License

Licensed under the Apache License 2.0.

Citation

If you use xtrax in research, please cite it:

@software{xtrax,
  title = {xtrax: High-Performance Composable JAX Training},
  author = {Russo, Marielle},
  year = {2026},
  url = {https://github.com/maraxen/xtrax}
}