MACA trains language models to be more consistent reasoners through multi-agent debate and consensus-based reinforcement learning. - View it on GitHub