cuequivariance

What it is and what it does

cuEquivariance is an NVIDIA library for constructing geometric neural networks that incorporate equivariance—the property of respecting symmetries like rotations and translations. It provides an API for describing segmented polynomials built from tensor products, optimized CUDA kernels for execution, and integration with PyTorch and JAX. The core idea is that AI models incorporating equivariance tend to be more data-efficient because they encode physical symmetries directly rather than learning them from data.

The package sits at the intersection of differential geometry and deep learning. It targets researchers and practitioners building physics-informed models, molecular simulations, or other domains where rotational and translational symmetry matter. Installation is straightforward for the core library, though GPU acceleration requires separate CUDA-specific packages. The library is in Beta, meaning its API and functionality may change.

Use it for:

• Building molecular property prediction models that respect 3D rotation and translation symmetries.
• Developing physics simulations where equivariance reduces data requirements for training.
• Constructing point-cloud neural networks for 3D shape analysis with built-in symmetry handling.
• Integrating equivariant operations into existing PyTorch or JAX workflows for geometric deep learning.
• Accelerating segmented polynomial computations on GPU for large-scale scientific computing tasks.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

cuEquivariance provides CUDA-accelerated operations for building geometric neural networks that respect symmetries, with segmented polynomial kernels and bindings for PyTorch and JAX.

Yes, if you are building geometric or physics-informed neural networks and need GPU acceleration for equivariant operations. The Apache 2.0 license is permissive, install friction is low, and maintenance is active. However, note that the package is in Beta, so expect potential API changes. Verify that the PyTorch or JAX bindings match your framework choice and that your CUDA version (12 or 13) is supported before committing.

Install

cuequivariance on PyPI

pip install cuequivariance

uv add cuequivariance

poetry add cuequivariance

Installing cuequivariance

Before you install

Low install friction with a pure-wheel distribution. Active maintenance with a release 7 days ago. Requires Python 3.10 or later and depends on common scientific libraries (numpy, scipy, sympy, networkx, opt-einsum).

License in practice

Apache 2.0 permissive license allows commercial and derivative use with attribution and license preservation requirements for distributed modifications.

Quickstart

pip install cuequivariance
import cuequivariance
# For PyTorch: pip install cuequivariance-torch cuequivariance-ops-torch-cu12
# For JAX: pip install cuequivariance-jax cuequivariance-ops-jax-cu12

CUDA kernels require separate installation (cuequivariance-ops-torch-cu12/cu13 or cuequivariance-ops-jax-cu12/cu13) and a compatible NVIDIA GPU.

Verify before relying

• Specific performance benchmarks or speedup claims compared to non-accelerated implementations.
• Data efficiency gains from equivariance in real-world training scenarios.
• Scope and completeness of PyTorch and JAX API coverage.
• Beta state stability and likelihood of breaking API changes in future releases.

Package facts

Evidence: cuequivariance-0.11.1-py3-none-any.whl

Tags

More Artificial Intelligence packages

Further reading