2025/02/14 by Yu-Ang Lee, Ching-Yun Ko, Lee, Yu-Ang +9 · 1 citation
Computer Science · Decision Sciences · #Advanced Computational Techniques and Applications #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Model-Driven Software Engineering Techniques #Scientific Computing and Data Management
paper · pdf · doi:10.48550/arxiv.2502.10339
openalex publication_date 2025/02/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Model merging is an efficient way of obtaining a multi-task model from several pretrained models without further fine-tuning, and it has gained attention in various domains, including natural language processing (NLP). Despite the efficiency, a key challenge in model merging is the seemingly inevitable decrease in task performance as the number of models increases. In this paper, we propose Spectral Truncation And Rescale (STAR) that aims at mitigating ``merging conflicts'' by truncating small components in the respective spectral spaces, which is followed by an automatic parameter rescaling scheme to retain the nuclear norm of the original matrix. STAR requires no additional inference on original training data and is robust to hyperparamater choice. We demonstrate the effectiveness of STAR through extensive model merging cases on diverse NLP tasks. Specifically, STAR works robustly across varying model sizes, and can outperform baselines by 4.2% when merging 12 models on Flan-T5. Our code is publicly available at https://github.com/IBM/STAR.