An asynchronous distributed reinforcement-learning engine for post-training text, vision, and audio models.