Self-Exploring Language Models: Active Preference Elicitation for Online Alignment

TL;DR

This paper introduces Self-Exploring Language Models (SELM), an active exploration method for online alignment of LLMs that improves reward modeling and instruction-following performance through iterative, biased exploration.

Contribution

SELM presents a bilevel optimization approach that enhances exploration efficiency and reduces the need for separate reward models in online LLM alignment.

Findings

SELM outperforms DPO in exploration and alignment quality.

Fine-tuning on Zephyr-7B-SFT and Llama-3-8B-Instruct improves benchmark scores.

Significant gains on MT-Bench, AlpacaEval 2.0, and academic benchmarks.

Abstract

Preference optimization, particularly through Reinforcement Learning from Human Feedback (RLHF), has achieved significant success in aligning Large Language Models (LLMs) to adhere to human intentions. Unlike offline alignment with a fixed dataset, online feedback collection from humans or AI on model generations typically leads to more capable reward models and better-aligned LLMs through an iterative process. However, achieving a globally accurate reward model requires systematic exploration to generate diverse responses that span the vast space of natural language. Random sampling from standard reward-maximizing LLMs alone is insufficient to fulfill this requirement. To address this issue, we propose a bilevel objective optimistically biased towards potentially high-reward responses to actively explore out-of-distribution regions. By solving the inner-level problem with the reparameterized reward function, the resulting algorithm, named Self-Exploring Language Models (SELM), eliminates the need for a separate RM and iteratively updates the LLM with a straightforward objective. Compared to Direct Preference Optimization (DPO), the SELM objective reduces indiscriminate favor of unseen extrapolations and enhances exploration efficiency. Our experimental results demonstrate that when fine-tuned on Zephyr-7B-SFT and Llama-3-8B-Instruct models, SELM significantly boosts the performance on instruction-following benchmarks such as MT-Bench and AlpacaEval 2.0, as well as various standard academic benchmarks in different settings. Our code and models are available at https://github.com/shenao-zhang/SELM.