Aligning Language Models with Preferences through f-divergence Minimization

TL;DR

This paper introduces f-DPG, a flexible framework for aligning language models with preferences using any f-divergence, unifying existing methods and demonstrating that divergence choice impacts alignment and diversity trade-offs.

Contribution

Proposes f-DPG, a novel approach that generalizes existing alignment methods by allowing any f-divergence to approximate target distributions, improving model alignment performance.

Findings

Jensen-Shannon divergence balances alignment and diversity effectively.

Different divergences offer distinct trade-offs in model alignment.

Jensen-Shannon divergence outperforms forward KL in experiments.

Abstract

Aligning language models with preferences can be posed as approximating a target distribution representing some desired behavior. Existing approaches differ both in the functional form of the target distribution and the algorithm used to approximate it. For instance, Reinforcement Learning from Human Feedback (RLHF) corresponds to minimizing a reverse KL from an implicit target distribution arising from a KL penalty in the objective. On the other hand, Generative Distributional Control (GDC) has an explicit target distribution and minimizes a forward KL from it using the Distributional Policy Gradient (DPG) algorithm. In this paper, we propose a new approach, f-DPG, which allows the use of any f-divergence to approximate any target distribution that can be evaluated. f-DPG unifies both frameworks (RLHF, GDC) and the approximation methods (DPG, RL with KL penalties). We show the practical benefits of various choices of divergence objectives and demonstrate that there is no universally optimal objective but that different divergences present different alignment and diversity trade-offs. We show that Jensen-Shannon divergence strikes a good balance between these objectives, and frequently outperforms forward KL divergence by a wide margin, leading to significant improvements over prior work. These distinguishing characteristics between divergences persist as the model size increases, highlighting the importance of selecting appropriate divergence objectives.