Fine-Tuning Language Models with Advantage-Induced Policy Alignment

TL;DR

This paper introduces Advantage-Induced Policy Alignment (APA), a novel reinforcement learning algorithm that improves the stability, sample efficiency, and performance of language models trained with human feedback, outperforming PPO.

Contribution

The paper proposes a new algorithm, APA, that addresses PPO's limitations by using advantage-based squared error loss, with both empirical and theoretical validation.

Findings

APA outperforms PPO in language tasks

APA provides more stable control over policy deviation

APA enhances sample efficiency and model performance

Abstract

Reinforcement learning from human feedback (RLHF) has emerged as a reliable approach to aligning large language models (LLMs) to human preferences. Among the plethora of RLHF techniques, proximal policy optimization (PPO) is of the most widely used methods. Despite its popularity, however, PPO may suffer from mode collapse, instability, and poor sample efficiency. We show that these issues can be alleviated by a novel algorithm that we refer to as Advantage-Induced Policy Alignment (APA), which leverages a squared error loss function based on the estimated advantages. We demonstrate empirically that APA consistently outperforms PPO in language tasks by a large margin, when a separate reward model is employed as the evaluator. In addition, compared with PPO, APA offers a more stable form of control over the deviation from the model's initial policy, ensuring that the model improves its performance without collapsing to deterministic output. In addition to empirical results, we also provide a theoretical justification supporting the design of our loss function.