The instruction for docker is for running the docker instance. But when you are inside the docker, you can't train/eval. It's just like cloning the code, downloading all datasets, pre-trained models on your own computer. OS: macOS