jaxtyping

IMPORTANT: this relaxes the python constraint to python3.8 to be installable with eztils, but will throw a RuntimeError if you actually try to run with python3.8.

Type annotations and runtime type-checking for:

1. shape and dtype of JAX arrays; (Now also supports PyTorch, NumPy, and TensorFlow!)
2. PyTrees.

For example:

from jaxtyping import Array, Float, PyTree

# Accepts floating-point 2D arrays with matching dimensions
def matrix_multiply(x: Float[Array, "dim1 dim2"],
                    y: Float[Array, "dim2 dim3"]
                  ) -> Float[Array, "dim1 dim3"]:
    ...

def accepts_pytree_of_ints(x: PyTree[int]):
    ...

def accepts_pytree_of_arrays(x: PyTree[Float[Array, "batch c1 c2"]]):
    ...

Installation

pip install jaxtyping

Requires Python 3.9+.

JAX is an optional dependency, required for a few JAX-specific types. If JAX is not installed then these will not be available, but you may still use jaxtyping to provide shape/dtype annotations for PyTorch/NumPy/TensorFlow/etc.

The annotations provided by jaxtyping are compatible with runtime type-checking packages, so it is common to also install one of these. The two most popular are typeguard (which exhaustively checks every argument) and beartype (which checks random pieces of arguments).

Documentation

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

Finally

See also: other libraries in the JAX ecosystem

Equinox: neural networks.

Optax: first-order gradient (SGD, Adam, ...) optimisers.

Diffrax: numerical differential equation solvers.

Lineax: linear solvers and linear least squares.

Eqxvision: computer vision models.

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

Levanter: scalable+reliable training of foundation models (e.g. LLMs).

Disclaimer

This is not an official Google product.