cuequivariance-ops-cu13

0.10.0 verified Fri May 01 auth: no python

GPU-accelerated operations for equivariant primitives, specifically compiled with CUDA 13 support. Version 0.10.0 is the latest, with active development.

pip install cuequivariance-ops-cu13

Common errors

error ModuleNotFoundError: No module named 'cuequivariance_ops_cu13' ↓

cause Package not installed or wrong Python environment.

fix

Run 'pip install cuequivariance-ops-cu13' in the correct environment.

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

cause The installed package was compiled for a different CUDA version than what is available on the system.

fix

Install the appropriate package version matching your CUDA toolkit, e.g., cuequivariance-ops-cu12 for CUDA 12.

error RuntimeError: Expected tensor to be on GPU, but found tensor on CPU ↓

cause Operations require GPU tensors but input is on CPU.

fix

Move tensors to GPU with .cuda() before passing to equivariant operations.

Warnings

breaking In version 0.10.0, the package name changed from 'cuequivariance-ops' to 'cuequivariance-ops-cu13' for CUDA 13 specific build. Import paths also changed accordingly. ↓

fix Update imports to use 'cuequivariance_ops_cu13' instead of 'cuequivariance.ops'.

gotcha The package is compiled for specific CUDA versions (13). Attempting to install on a system with different CUDA toolkit may result in runtime errors or missing symbols. ↓

fix Ensure your system has CUDA 13 installed or use a different version of the package matching your CUDA version (e.g., cuequivariance-ops-cu12).

deprecated Older versions of cuequivariance-ops (pre-0.9) used different function signatures and are not backward compatible. ↓

fix Refer to migration guide in the documentation when upgrading from older versions.

Imports

EquivariantOp
wrong
```
from cuequivariance.ops import EquivariantOp
```
correct
```
from cuequivariance_ops_cu13 import EquivariantOp
```
The subpackage name differs between versions; older code may use 'cuequivariance.ops'.
SO3_linear
wrong
```
from cuequivariance import SO3_linear
```
correct
```
from cuequivariance_ops_cu13 import SO3_linear
```
SO3_linear and other ops are in the ops subpackage, not directly in cuequivariance.

Quickstart

Basic import and usage example for cuequivariance-ops-cu13. Ensure torch is installed and CUDA 13 is available.

import torch
from cuequivariance_ops_cu13 import EquivariantOp, SO3_linear

# Create a simple equivariant linear layer
batch = 4
irreps_in = 5
irreps_out = 3

x = torch.randn(batch, irreps_in * 2)  # shape (batch, 2*irreps_in) for complex representations

# Note: Actual usage may require specific shapes and irrep specs.
# The following is an example of calling a function.
# For real usage, see the documentation.
print('cuequivariance-ops-cu13 imported successfully')