Grouped Reward Policy Optimization, a reinforcement learning algorithm for training language models with group-based reward estimation.