Parameter Competition Balancing for Model Merging

TL;DR

This paper presents PCB-Merging, a lightweight, training-free method for effectively merging multiple fine-tuned models by balancing parameter competition, leading to improved multitasking performance across diverse scenarios.

Contribution

Introduces PCB-Merging, a novel parameter balancing technique that adjusts parameter importance for better model merging without retraining.

Findings

Outperforms existing model merging methods across various tasks and domains.

Enhances out-of-domain generalization and multitasking capabilities.

Effective across different model sizes and modalities.

Abstract

While fine-tuning pretrained models has become common practice, these models often underperform outside their specific domains. Recently developed model merging techniques enable the direct integration of multiple models, each fine-tuned for distinct tasks, into a single model. This strategy promotes multitasking capabilities without requiring retraining on the original datasets. However, existing methods fall short in addressing potential conflicts and complex correlations between tasks, especially in parameter-level adjustments, posing a challenge in effectively balancing parameter competition across various tasks. This paper introduces an innovative technique named PCB-Merging (Parameter Competition Balancing), a lightweight and training-free technique that adjusts the coefficients of each parameter for effective model merging. PCB-Merging employs intra-balancing to gauge parameter significance within individual tasks and inter-balancing to assess parameter similarities across different tasks. Parameters with low importance scores are dropped, and the remaining ones are rescaled to form the final merged model. We assessed our approach in diverse merging scenarios, including cross-task, cross-domain, and cross-training configurations, as well as out-of-domain generalization. The experimental results reveal that our approach achieves substantial performance enhancements across multiple modalities, domains, model sizes, number of tasks, fine-tuning forms, and large language models, outperforming existing model merging methods. The code is publicly available at: \url{https://github.com/duguodong7/pcb-merging}.