Installation¶

Requirements¶

Python >= 3.10
beartype >= 0.20, installed automatically with shapix-rt

Install with pip¶

pip install shapix-rt

The distribution name is shapix-rt, where rt means runtime. The import path stays:

import shapix

Shapix intentionally does not use extras such as shapix-rt[numpy]. Install shapix-rt and your backend packages explicitly.

NumPyPyTorchJAXCuPyNumPy + OpTree

pip install shapix-rt numpy

pip install shapix-rt numpy torch

pip install shapix-rt numpy jax

pip install shapix-rt numpy cupy

pip install shapix-rt numpy optree  # or install jax and use shapix.jax.Tree

Note

shapix.jax, shapix.torch, and shapix.cupy require numpy alongside the backend. The lightweight root import import shapix does not.

Install with uv¶

uv add shapix-rt

Optional dependencies¶

Package	Purpose
`numpy`	NumPy array aliases, `ScalarLike`, and backend dtype helpers
`torch`	PyTorch tensor aliases and Torch `Like` types
`jax`	JAX array aliases, JAX `Like` types, and JAX `Tree`
`cupy`	CuPy array aliases and CuPy `Like` types
`optree`	Explicit OpTree backend via `shapix.optree.Tree`

Import boundaries¶

The root package is designed to stay optional-dependency-safe:

import shapix

print(shapix.__version__)
print(shapix.N, shapix.C)

That works even in a plain source checkout without installed package metadata. In that case __version__ falls back to a non-empty string such as 0+unknown.

Backend modules are stricter:

shapix.numpy needs numpy
shapix.jax needs jax and numpy
shapix.torch needs torch and numpy
shapix.cupy needs cupy and numpy
shapix.optree needs optree

Verify installation¶

import shapix
print(shapix.__version__)

Then verify the backend you actually plan to use:

from shapix import N, C
from shapix.numpy import F32