Iterative Length-Regularized Direct Preference Optimization: A Case Study on Improving 7B Language Models to GPT-4 Level

TL;DR

This paper introduces iterative length-regularized DPO (iLR-DPO), a method that improves 7B language models to GPT-4 level by balancing response quality and verbosity through length penalization.

Contribution

The paper proposes iLR-DPO, a novel extension of DPO that incorporates length regularization to prevent verbosity and enhance alignment with human preferences.

Findings

7B model achieves GPT-4 level performance on benchmarks

iLR-DPO improves response quality without increasing verbosity

Model outperforms GPT-4 in length-controlled win rate

Abstract

Direct Preference Optimization (DPO), a standard method for aligning language models with human preferences, is traditionally applied to offline preferences. Recent studies show that DPO benefits from iterative training with online preferences labeled by a trained reward model. In this work, we identify a pitfall of vanilla iterative DPO - improved response quality can lead to increased verbosity. To address this, we introduce iterative length-regularized DPO (iLR-DPO) to penalize response length. Our empirical results show that iLR-DPO can enhance a 7B model to perform on par with GPT-4 without increasing verbosity. Specifically, our 7B model achieves a $50.5\%$ length-controlled win rate against $\texttt{GPT-4 Preview}$ on AlpacaEval 2.0, and excels across standard benchmarks including MT-Bench, Arena-Hard and OpenLLM Leaderboard. These results demonstrate the effectiveness of iterative DPO in aligning language models with human feedback.