Skip to content

Klax¤

A lightweight machine learning package for computational mechanics built on JAX.

Warning

Klax is still in early development and will likely see significant API changes in the near future. Likewise, the documentation is still under heavy development.

Overview¤

Klax provides specialized machine learning architectures, constraints, and training utilities for mechanics and physics applications. Built on top of JAX, Equinox, and Optax, it offers:

  • Special Neural Networks: Implementations of, e.g., Input Convex Neural Networks (ICNNs), matrix-valued neural networks, MLPs with custom initialization, and more.
  • JAX Compatibility: Seamless integration with JAX's automatic differentiation and acceleration
  • Parameter Constraints: Differentiable and non-differentiable parameter constraints through klax.Unwrappable and klax.Constraint
  • Customizable Training: Methods and APIs for customized calibrations on arbitrary PyTree data structures through klax.fit, klax.Loss, and klax.Callback.

Klax is designed to be minimally intrusive - all models inherit directly from equinox.Module without additional abstraction layers. This ensures full compatibility with the JAX/Equinox ecosystem.

The constraint system is derived from Paramax's paramax.AbstractUnwrappable, extending it to support non-differentiable/zero-gradient parameter constraints such as ReLU-based non-negativity constraints.

The provided calibration utilities (klax.fit, klax.Loss, klax.Callback) are designed to operate on arbitrarily shaped PyTrees of data, fully utilizing the flexibility of JAX and Equinox. While they cover most common machine learning use cases, as well as our specialized requirements, they remain entirely optional. The core building blocks of Klax work seamlessly in custom training loops.

Currently Klax's training utilities are built around Optax, but different optimization libraries could be supported in the future if desired.

Installation¤

Klax can be installed via pip using

pip install klax

If you want to add the latest release of klax to your Python uv project run

uv add klax

or get the most recent changes from the main branch via

uv add "klax @ git+https://github.com/Drenderer/klax.git@main"

Getting Started¤

If you're new to the JAX ecosystem, we recommend looking at Quickstart, which provides a concise overview of JAX's core functionality.

As the contents of the documentation are still rudimentary, we recommend checking out Equinox and Paramax and taking a look at our examples.

Citation¤

Acknowledgement¤

Klax is built on top of several powerful frameworks:

JAX - For automatic differentiation and acceleration
Equinox - For neural network primitives
Optax - For optimization utilities
Paramax - For constraints (We decided to embed Paramax directly into Klax due to the need for non-differentiable constraints).