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

BeaverTails is a collection of datasets designed to facilitate research on safety alignment in large language models (LLMs).

[ACL2025 Best Paper] Language Models Resist Alignment

🤖 Enhance reinforcement learning stability and efficiency with advanced algorithms like TRPO, PPO, DPO, GRPO, DAPO, and GSPO for optimized policy training.