BOND: Aligning LLMs with Best-of-N Distillation

TL;DR

BOND is a novel RLHF algorithm that efficiently mimics Best-of-N sampling by aligning model distributions, leading to improved summarization and model performance without high inference costs.

Contribution

BOND introduces a distribution matching approach using Jeffreys divergence to emulate Best-of-N sampling efficiently in RLHF.

Findings

Outperforms other RLHF methods on benchmarks

Improves summarization quality in experiments

Efficiently aligns model distributions with Best-of-N

Abstract

Reinforcement learning from human feedback (RLHF) is a key driver of quality and safety in state-of-the-art large language models. Yet, a surprisingly simple and strong inference-time strategy is Best-of-N sampling that selects the best generation among N candidates. In this paper, we propose Best-of-N Distillation (BOND), a novel RLHF algorithm that seeks to emulate Best-of-N but without its significant computational overhead at inference time. Specifically, BOND is a distribution matching algorithm that forces the distribution of generations from the policy to get closer to the Best-of-N distribution. We use the Jeffreys divergence (a linear combination of forward and backward KL) to balance between mode-covering and mode-seeking behavior, and derive an iterative formulation that utilizes a moving anchor for efficiency. We demonstrate the effectiveness of our approach and several design choices through experiments on abstractive summarization and Gemma models. Aligning Gemma policies with BOND outperforms other RLHF algorithms by improving results on several benchmarks.