Add to Chrome
Add to Firefox
Add to EdgeChatGLM-RLHF: Practices of Aligning Large Language Models with Human Feedback
Apr 03, 2024
ChatGLM is a free-to-use AI service powered by the ChatGLM family of large language models (LLMs). In this paper, we present the ChatGLM-RLHF pipeline -- a reinforcement learning from human feedback (RLHF) system -- designed to enhance ChatGLM's alignment with human preferences. ChatGLM-RLHF encompasses three major components: the collection of human preference data, the training of the reward model, and the optimization of policies. Throughout the process of integrating ChatGLM-RLHF into production, we encountered and addressed several unprecedented challenges. We introduce the strategies to mitigate reward variance for stabilized large-scale training, implement model parallelism with fused gradient-descent, and design regularization constraints to avoid catastrophic forgetting in LLMs. Experiments show that ChatGLM-RLHF brings significant improvements in alignment tasks compared to the supervised fine-tuned (SFT) version of ChatGLM. For instance, it achieves on average 15\% more wins against ChatGLM-SFT in Chinese alignment tasks. The work presents our practices of aligning LLMs with human preferences, offering insights into the challenges and solutions in RLHF implementations.
