Rewarded soups: towards Pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards

TL;DR

This paper introduces rewarded soup, a method that interpolates weights of multiple models fine-tuned on diverse rewards to achieve Pareto-optimal alignment across various tasks and preferences.

Contribution

It proposes a novel multi-policy interpolation approach that maintains linear connectivity of weights fine-tuned on different rewards, enabling better alignment with diverse objectives.

Findings

Effective across text, image, and control tasks

Weights remain linearly connected after diverse reward fine-tuning

Improves alignment by interpolating multiple specialized models

Abstract

Foundation models are first pre-trained on vast unsupervised datasets and then fine-tuned on labeled data. Reinforcement learning, notably from human feedback (RLHF), can further align the network with the intended usage. Yet the imperfections in the proxy reward may hinder the training and lead to suboptimal results; the diversity of objectives in real-world tasks and human opinions exacerbate the issue. This paper proposes embracing the heterogeneity of diverse rewards by following a multi-policy strategy. Rather than focusing on a single a priori reward, we aim for Pareto-optimal generalization across the entire space of preferences. To this end, we propose rewarded soup, first specializing multiple networks independently (one for each proxy reward) and then interpolating their weights linearly. This succeeds empirically because we show that the weights remain linearly connected when fine-tuned on diverse rewards from a shared pre-trained initialization. We demonstrate the effectiveness of our approach for text-to-text (summarization, Q&A, helpful assistant, review), text-image (image captioning, text-to-image generation, visual grounding, VQA), and control (locomotion) tasks. We hope to enhance the alignment of deep models, and how they interact with the world in all its diversity.