This PR makes several changes to how dependencies are installed so that TensorFlow and PyTorch work with GPU support. In particular:
On Windows-native, TensorFlow 2.10.1 is installed to ensure GPU support. On all other platforms, the latest version (2.15) is installed.
PyTorch is pinned to 2.0.1 and is only automatically installed on macOS. For GPU support with CUDA 11.8, users now use pipx inject to install version 2.0.1 compiled with CUDA 11.8.
Users are now instructed to install CUDA 11.8 and cuDNN 9.0.0, and the documentation now links to resources to assist with issues installing CUDA.
This PR makes several changes to how dependencies are installed so that TensorFlow and PyTorch work with GPU support. In particular:
pipx injectto install version 2.0.1 compiled with CUDA 11.8.