If you don't need cuda kernels acceleration simply execute
python setup.py install
First you need to install nvcc and conda.
conda create -n e3nn python=3.8
To activate the environement execute
conda activate e3nn
pytorch need to be compatible with you version of nvcc.
Check your version of nvcc
nvcc --version
If like me you have Cuda compilation tools, release 9.2, V9.2.148 install pytorch with
pip install torch==1.5.0+cu92 torchvision==0.6.0+cu92 -f https://download.pytorch.org/whl/torch_stable.html
The version of cuda needs to match between pytorch and nvcc!
Now we can check if cuda works
python -c "import torch; print(torch.randn((), device='cuda'))"
Clone this repository
git clone [email protected]:e3nn/e3nn.git
cd e3nn
Remove previous attempts and cache
rm -rf build dist e3nn.egg-info __pycache__
rm -rf .cache/pip
Execute installation script
python setup_cuda.py install
Check compilation success
python -c "import torch; import e3nn.cuda_rsh"