Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study

TL;DR

This study compares DPO and PPO in aligning large language models with human preferences, revealing PPO's superior performance across various benchmarks and its ability to achieve state-of-the-art results.

Contribution

The paper provides a comprehensive analysis of DPO and PPO, highlighting PPO's advantages and establishing its effectiveness in LLM alignment tasks.

Findings

PPO outperforms DPO in multiple RLHF benchmarks.

PPO achieves state-of-the-art results in code generation tasks.

Theoretical limitations of DPO are identified.

Abstract

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.