This was ran on Ubuntu Server 24.0.1 hosted using a Virtual Machine. MacOS (version higher than Sierra) might also work, but has not been tested. Avoid using Windows, one of the dependencies (jaxlib) will fail to install.
Most modern computers should work. The main requirement, that I experienced, is that your CPU must be able to run A7X instructions.
For more details, look at Tensorflow's specifications.
Install python version 3.8.20 (any 3.8.x versions may also work).
Clone this repository:
git clone https://github.com/butterman0423/SGN_patching.gitEnter the repository:
cd /path/tp/repositoryCreate a .venv virtual environment and connect to it:
source .venv/bin/activateRun the following pip commands to install required dependencies:
pip install --upgrade pip setuptools
pip install jaxlib==0.1.76 -f https://storage.googleapis.com/jax-releases/jax_releases.html
pip install ./uncertainty-baselines[jax,models,datasets]
pip install --no-deps dm-haiku==0.0.6You are now able to run the tests. See the original SGN repository (https://github.com/ErikEnglesson/SGN) for directions on how to run each test.
Enabling GPU is a complicated process. Since Tensorflow 2.8.1 is used, CUDA 11.2 and cuDNN 8.1 are required. See https://www.tensorflow.org/install/source for details.
Follow this link for instructions on how to set up these packages properly: https://gist.github.com/adwellj/2769957e7fc8c7e9372e5f4b06f72b80. On Windows WSL2 also check out NVIDIA's help page for this https://docs.nvidia.com/cuda/wsl-user-guide/index.html#getting-started-with-cuda-on-wsl. Specifically, the step about removing the GPG key
Lastly run the following command in your shell to ensure that tensorflow recognizes the GPU
python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"