Guardado en:
Detalles Bibliográficos
Autores principales: Xu, Shusheng, Fu, Wei, Gao, Jiaxuan, Ye, Wenjie, Liu, Weilin, Mei, Zhiyu, Wang, Guangju, Yu, Chao, Wu, Yi
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:https://arxiv.org/abs/2404.10719
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912065776517120
author Xu, Shusheng
Fu, Wei
Gao, Jiaxuan
Ye, Wenjie
Liu, Weilin
Mei, Zhiyu
Wang, Guangju
Yu, Chao
Wu, Yi
author_facet Xu, Shusheng
Fu, Wei
Gao, Jiaxuan
Ye, Wenjie
Liu, Weilin
Mei, Zhiyu
Wang, Guangju
Yu, Chao
Wu, Yi
contents Reinforcement Learning from Human Feedback (RLHF) is currently the most widely used method to align large language models (LLMs) with human preferences. Existing RLHF methods can be roughly categorized as either reward-based or reward-free. Novel applications such as ChatGPT and Claude leverage reward-based methods that first learn a reward model and apply actor-critic algorithms, such as Proximal Policy Optimization (PPO). However, in academic benchmarks, state-of-the-art results are often achieved via reward-free methods, such as Direct Preference Optimization (DPO). Is DPO truly superior to PPO? Why does PPO perform poorly on these benchmarks? In this paper, we first conduct both theoretical and empirical studies on the algorithmic properties of DPO and show that DPO may have fundamental limitations. Moreover, we also comprehensively examine PPO and reveal the key factors for the best performances of PPO in fine-tuning LLMs. Finally, we benchmark DPO and PPO across a collection of RLHF testbeds, ranging from dialogue to code generation. Experiment results demonstrate that PPO is able to surpass other alignment methods in all cases and achieve state-of-the-art results in challenging code competitions. Our code is publicly available at https://github.com/openpsi-project/ReaLHF.
format Preprint
id arxiv_https___arxiv_org_abs_2404_10719
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study
Xu, Shusheng
Fu, Wei
Gao, Jiaxuan
Ye, Wenjie
Liu, Weilin
Mei, Zhiyu
Wang, Guangju
Yu, Chao
Wu, Yi
Computation and Language
Reinforcement Learning from Human Feedback (RLHF) is currently the most widely used method to align large language models (LLMs) with human preferences. Existing RLHF methods can be roughly categorized as either reward-based or reward-free. Novel applications such as ChatGPT and Claude leverage reward-based methods that first learn a reward model and apply actor-critic algorithms, such as Proximal Policy Optimization (PPO). However, in academic benchmarks, state-of-the-art results are often achieved via reward-free methods, such as Direct Preference Optimization (DPO). Is DPO truly superior to PPO? Why does PPO perform poorly on these benchmarks? In this paper, we first conduct both theoretical and empirical studies on the algorithmic properties of DPO and show that DPO may have fundamental limitations. Moreover, we also comprehensively examine PPO and reveal the key factors for the best performances of PPO in fine-tuning LLMs. Finally, we benchmark DPO and PPO across a collection of RLHF testbeds, ranging from dialogue to code generation. Experiment results demonstrate that PPO is able to surpass other alignment methods in all cases and achieve state-of-the-art results in challenging code competitions. Our code is publicly available at https://github.com/openpsi-project/ReaLHF.
title Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study
topic Computation and Language
url https://arxiv.org/abs/2404.10719