Where are the models stored and what script should be run to predict supertags for a collection of sentences?