DPO Meets PPO: Reinforced Token Optimization for RLHF | Read Paper on Bytez