Add new Relax function to the batched model for evaluating query tokens over multiple time steps in parallel

Add new Relax function to the batched model for evaluating query tokens over multiple time steps in parallel #247