Secrets of RLHF in Large Language Models Part II: Reward Modeling