Including all of Github in your training dataset seems like a good idea