LUFFY is a reinforcement learning framework that bridges the gap between zero-RL and imitation learning by incorporating off-policy reasoning traces into the training process. Built upon GRPO, LUFFY ...