Bone Soups: A Seek-and-Soup Model Merging Approach for Controllable Multi-Objective Generation

TL;DR

Bone Soup introduces a novel model merging approach that effectively balances multiple objectives in language generation, enabling controllable and Pareto optimal outputs tailored to diverse user demands.

Contribution

It proposes Bone Soup, a multi-objective reinforcement learning-based model merging method that considers objective impacts and uses a symmetric circulant matrix for flexible model combination.

Findings

Demonstrates strong controllability in multi-objective generation

Achieves Pareto optimality in diverse generation tasks

Outperforms existing merging approaches in experiments

Abstract

User information needs are often highly diverse and varied. A key challenge in current research is how to achieve controllable multi-objective generation while enabling rapid adaptation to accommodate diverse user demands during test time. Existing solutions, such as Rewarded Soup, focus on merging language models individually tuned on single objectives. While easy to implement and widely used, these approaches face limitations in achieving optimal performance due to their disregard for the impacts of competing objectives on model tuning. To address this issue, we propose Bone Soup, a novel model merging approach that first seeks a series of backbone models by considering the impacts of multiple objectives and then makes the soup (i.e., merge the backbone models). Specifically, Bone Soup begins by training multiple backbone models for different objectives using multi-objective reinforcement learning. Each backbone model is guided by a combination of backbone reward signals. To ensure that these models are optimal for the Pareto front, the backbone rewards are crafted by combining standard reward functions into basis vectors, which can then be modified through a rule-based construction method. Bone Soup leverages a symmetric circulant matrix mapping to generate the merging coefficients, which are used to merge the backbone models according to user preferences. Extensive experimental results demonstrate that Bone Soup exhibits strong controllability and Pareto optimality in controllable multi-objective generation, providing a more effective and efficient approach to addressing diverse user needs at test time.