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

• URL: http://arxiv.org/abs/2404.10719v3
• Date: Thu, 10 Oct 2024 08:30:17 GMT
• Title: Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study
• Authors: Shusheng Xu, Wei Fu, Jiaxuan Gao, Wenjie Ye, Weilin Liu, Zhiyu Mei, Guangju Wang, Chao Yu, Yi Wu,
• Abstract summary: Reinforcement Learning from Human Feedback (RLHF) is currently the most widely used method to align large language models (LLMs) with human preferences. In academic benchmarks, state-of-the-art results are often achieved via reward-free methods, such as Direct Preference Optimization (DPO) We show that PPO is able to surpass other alignment methods in all cases and achieve state-of-the-art results in challenging code competitions.
• Score: 16.99550556866219
• License:
• 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.

Related papers

• MPPO: Multi Pair-wise Preference Optimization for LLMs with Arbitrary Negative Samples [22.521746860874305]
  This study introduces the MPPO algorithm, which leverages the average likelihood of model responses to fit the reward function. Through a comparison of Point-wise, Pair-wise, and List-wise implementations, we found that the Pair-wise approach achieves the best performance. Experimental results demonstrate MPPO's outstanding performance across various benchmarks.
  arXiv  Detail & Related papers  (2024-12-13T14:18:58Z)
• Accelerated Preference Optimization for Large Language Model Alignment [60.22606527763201]
  Reinforcement Learning from Human Feedback (RLHF) has emerged as a pivotal tool for aligning large language models (LLMs) with human preferences. Direct Preference Optimization (DPO) formulates RLHF as a policy optimization problem without explicitly estimating the reward function. We propose a general Accelerated Preference Optimization (APO) framework, which unifies many existing preference optimization algorithms.
  arXiv  Detail & Related papers  (2024-10-08T18:51:01Z)
• Understanding Reference Policies in Direct Preference Optimization [50.67309013764383]
  Direct Preference Optimization (DPO) has become a widely used training method for the instruction fine-tuning of large language models (LLMs) This work explores an under-investigated aspect of DPO - its dependency on the reference model or policy.
  arXiv  Detail & Related papers  (2024-07-18T17:08:10Z)
• DPO Meets PPO: Reinforced Token Optimization for RLHF [35.638723885233475]
  We introduce an algorithm that learns the token-wise reward function from preference data and performs policy optimization based on this learned token-wise reward signal. Experiments demonstrate that textttRTO performs better than PPO and other direct preference learning algorithms.
  arXiv  Detail & Related papers  (2024-04-29T17:58:30Z)
• From $r$ to $Q^*$: Your Language Model is Secretly a Q-Function [50.812404038684505]
  We show that we can derive DPO in the token-level MDP as a general inverse Q-learning algorithm, which satisfies the Bellman equation. We discuss applications of our work, including information elicitation in multi-turn dialogue, reasoning, agentic applications and end-to-end training of multi-model systems.
  arXiv  Detail & Related papers  (2024-04-18T17:37:02Z)
• A dynamical clipping approach with task feedback for Proximal Policy Optimization [29.855219523565786]
  There is no theoretical proof that the optimal PPO clipping bound remains consistent throughout the entire training process. Past studies have aimed to dynamically adjust PPO clipping bound to enhance PPO's performance. We propose Preference based Proximal Policy Optimization (Pb-PPO) to better reflect the preference (maximizing Return) of reinforcement learning tasks.
  arXiv  Detail & Related papers  (2023-12-12T06:35:56Z)
• Secrets of RLHF in Large Language Models Part I: PPO [81.01936993929127]
  Large language models (LLMs) have formulated a blueprint for the advancement of artificial general intelligence. reinforcement learning with human feedback (RLHF) emerges as the pivotal technological paradigm underpinning this pursuit. In this report, we dissect the framework of RLHF, re-evaluate the inner workings of PPO, and explore how the parts comprising PPO algorithms impact policy agent training.
  arXiv  Detail & Related papers  (2023-07-11T01:55:24Z)
• Fine-Tuning Language Models with Advantage-Induced Policy Alignment [80.96507425217472]
  We propose a novel algorithm for aligning large language models to human preferences. We show that it consistently outperforms PPO in language tasks by a large margin. We also provide a theoretical justification supporting the design of our loss function.
  arXiv  Detail & Related papers  (2023-06-04T01:59:40Z)
• You May Not Need Ratio Clipping in PPO [117.03368180633463]
  Proximal Policy Optimization (PPO) methods learn a policy by iteratively performing multiple mini-batch optimization epochs of a surrogate objective with one set of sampled data. Ratio clipping PPO is a popular variant that clips the probability ratios between the target policy and the policy used to collect samples. We show in this paper that such ratio clipping may not be a good option as it can fail to effectively bound the ratios. We show that ESPO can be easily scaled up to distributed training with many workers, delivering strong performance as well.
  arXiv  Detail & Related papers  (2022-01-31T20:26:56Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.