Skip to main content

PyTorch-like neural networks in JAX

Project description

Equinox

Equinox is a JAX library based around a simple idea: represent parameterised functions (such as neural networks) as PyTrees.

In doing so:

  • We get a PyTorch-like API...
  • ...that's fully compatible with native JAX transformations...
  • ...with no new concepts you have to learn. (It's all just PyTrees.)

The elegance of Equinox is its selling point in a world that already has Haiku, Flax and so on.

(In other words, why should you care? Because Equinox is really simple to learn, and really simple to use.)

Installation

pip install equinox

Requires Python 3.7+ and JAX 0.3.4+.

Documentation

Available at https://docs.kidger.site/equinox.

Quick example

Models are defined using PyTorch-like syntax:

import equinox as eqx
import jax

class Linear(eqx.Module):
    weight: jax.numpy.ndarray
    bias: jax.numpy.ndarray

    def __init__(self, in_size, out_size, key):
        wkey, bkey = jax.random.split(key)
        self.weight = jax.random.normal(wkey, (out_size, in_size))
        self.bias = jax.random.normal(bkey, (out_size,))

    def __call__(self, x):
        return self.weight @ x + self.bias

and fully compatible with normal JAX operations:

@jax.jit
@jax.grad
def loss_fn(model, x, y):
    pred_y = jax.vmap(model)(x)
    return jax.numpy.mean((y - pred_y) ** 2)

batch_size, in_size, out_size = 32, 2, 3
model = Linear(in_size, out_size, key=jax.random.PRNGKey(0))
x = jax.numpy.zeros((batch_size, in_size))
y = jax.numpy.zeros((batch_size, out_size))
grads = loss_fn(model, x, y)

Finally, there's no magic behind the scenes. All eqx.Module does is register your class as a PyTree. From that point onwards, JAX already knows how to work with PyTrees.

Citation

If you found this library to be useful in academic work, then please cite: (arXiv link)

@article{kidger2021equinox,
    author={Patrick Kidger and Cristian Garcia},
    title={{E}quinox: neural networks in {JAX} via callable {P}y{T}rees and filtered transformations},
    year={2021},
    journal={Differentiable Programming workshop at Neural Information Processing Systems 2021}
}

(Also consider starring the project on GitHub.)

See also

Numerical differential equation solvers: Diffrax.

Type annotations and runtime checking for PyTrees and shape/dtype of JAX arrays: jaxtyping.

SymPy<->JAX conversion; train symbolic expressions via gradient descent: sympy2jax.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

equinox-0.6.0.tar.gz (55.5 kB view hashes)

Uploaded Source

Built Distribution

equinox-0.6.0-py3-none-any.whl (66.6 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page