Installation¶

Requirements¶

• Python >= 3.10
• beartype >= 0.20, installed automatically with bearshape

Install with pip¶

pip install bearshape

The distribution name and import package are both bearshape:

import bearshape

bearshape intentionally does not use extras such as bearshape[numpy]. Install bearshape and your backend packages explicitly.

pip install bearshape numpy

pip install bearshape numpy torch

pip install bearshape numpy jax

pip install bearshape numpy cupy

pip install bearshape numpy optree  # or install jax and use bearshape.jax.Tree

!!! note bearshape.jax, bearshape.torch, and bearshape.cupy require numpy

alongside the backend. The lightweight root import import bearshape does not.

Install with uv¶

uv add bearshape

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 bearshape.optree.Tree |

Import boundaries¶

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

import bearshape

print(bearshape.__version__)
print(bearshape.N, bearshape.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:

• bearshape.numpy needs numpy
• bearshape.jax needs jax and numpy
• bearshape.torch needs torch and numpy
• bearshape.cupy needs cupy and numpy
• bearshape.optree needs optree

Verify installation¶

import bearshape
print(bearshape.__version__)

Then verify the backend you actually plan to use:

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