Project scenario:
When running python for neural network training, the scatter_max operator is not suitable and an error is reported: RuntimeError: Not compiled with CUDA support
Problem Description
RuntimeError: Not compiled with CUDA support
Cause Analysis:
torch_scatter package version and CUDA version do not match
solution:
- Delete the current version torch_scatter
pip uninstall torch-scatter
- Install matching version torch_scatter
pip install torch-scatter -f https://data.pyg.org/whl/torch-2.0.1+cu118.html
Among them, 2.0.1 means that the current pytorch version number is 2.0.1 and the current CUDA version number is 11.8. These two version numbers need to match the corresponding version numbers installed on your computer.