Novel recurrent architecture based language model for faster inference when generating long sequences.
