Safe RLHF: Constrained Value Alignment via Safe Reinforcement Learning from Human Feedback
PKU-Alignment
