cuequivariance-ops-cu12

0.10.0 verified Mon Apr 27 auth: no python

GPU-accelerated operations for equivariant primitives, part of the cuequivariance ecosystem. Version 0.10.0, actively maintained. Provides CUDA 12.x kernels for spherical tensor operations.

pip install cuequivariance-ops-cu12

Common errors

error ModuleNotFoundError: No module named 'cuequivariance_ops_cu12' ↓

cause Incorrect import: the package installs as cuequivariance-ops-cu12 but the importable module is cuequivariance_ops.

fix

Use 'import cuequivariance_ops' instead.

error RuntimeError: CUDA error: no kernel image is available for execution on the device ↓

cause Trying to run on a GPU that requires a different CUDA version (e.g., CUDA 11.x).

fix

Install the correct wheel for your CUDA version (e.g., cuequivariance-ops-cu11).

Warnings

breaking CUDA 12.x required: This package provides kernels compiled for CUDA 12.x only. Older CUDA versions will cause runtime errors. ↓

fix Install cu11 variant instead or update to CUDA 12.x.

gotcha Import name mismatch: Install as cuequivariance-ops-cu12 but import as cuequivariance_ops. ↓

fix Use 'import cuequivariance_ops as ceops' in code.

gotcha Requires Python >=3.10. Using older Python will cause installation failure. ↓

fix Upgrade Python to 3.10 or higher.

Imports

cuequivariance_ops
wrong
```
import cuequivariance_ops_cu12
```
correct
```
import cuequivariance_ops as ceops
```
The package is installed as cuequivariance-ops-cu12 but the import name is cuequivariance_ops. Trying to import 'cuequivariance_ops_cu12' will fail with ModuleNotFoundError.

Quickstart

Basic import and environment check for cuequivariance-ops.

import torch
import cuequivariance_ops as ceops

device = 'cuda' if torch.cuda.is_available() else 'cpu'
# Example: spherical harmonic transform
irreps_out, irreps_in = "1x0o", "1x1e"  # adapted to actual ops
# Check available ops
print(dir(ceops))