Added an example which runs the BoolQ benchmark against a model, outputting total accuracy. This can be used to quickly evaluate a model, and also demonstrates the use of grammars to constrain sampling.

Author: martindevans