TRPO, PPO, 그리고 DPO를 통한 거대언어모델 강화학습 방법론 동향 연구

초록

As the utilization of reinforcement learning (RL) in training large language models (LLMs) becomes more prevalent, the necessity to identify optimal RL methodologies tailored for LLMs has emerged. The fields of LLMs andRL are continually evolving through the development of novel techniques that contribute to their mutual advancement. This paper addresses the current trends in reinforcement learning algorithms aimed at enhancing the performanceof large language models.

키워드