Repository for setting up the ABC-toolkit.

• Docker. For linux, I used this guide.
• nvidia-container-toolkit

On Windows, I recommend installing Windows Subsystem for Linux (WSL). This isn't required but I haven't tested without it and the commands below may be slightly different.

On Linux distributions, you will need to add the current user to the docker group with sudo usermod -a -G docker USERNAME.

1. Make a copy of .env-default and name it .env. Update parameters if needed. For testing, default values should work, just create the following directories: ./data/inputs & ./data/outputs.

   • INPUT_DIR is mounted to /data/inputs in all the containers. Your data should be (anywhere) in INPUT_DIR.
   • OUTPUT_DIR will contain all container outputs.

2. Make the directory backend/models. In practice, this will contain all ML models but for now, leave it empty.

3. Build the base image: ./build_base_image.sh

4. Run docker compose up. You should see output similar to:

If so, all containers are running and the backend can detect the GPU!!