Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectors
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910999698735104 |
|---|---|
| author | Cheng, Runxi Xiong, Feng Wei, Yongxian Zhu, Wanyun Yuan, Chun |
| author_facet | Cheng, Runxi Xiong, Feng Wei, Yongxian Zhu, Wanyun Yuan, Chun |
| contents | Model merging seeks to integrate task-specific expert models into a unified architecture while preserving multi-task generalization capabilities, yet parameter interference between constituent models frequently induces performance degradation. Although prior work has explored many merging strategies, resolving interference without additional data for retraining or test-time computation remains challenging. In this paper, we theoretically demonstrate that the task vectors of the linear layer constitute an approximate linear subspace for its corresponding input. Therefore, we can minimize interference under the guidance of task vectors. Based on this insight, we propose \textbf{WUDI-Merging} (\textbf{W}hoever started the interference sho\textbf{U}ld en\textbf{D} \textbf{I}t), a simple yet effective model merging method that eliminates interference without any additional data or rescaling coefficients. Comprehensive empirical evaluations across vision and language benchmarks demonstrate our method's superiority, achieving state-of-the-art performance in data-free model merging scenarios (average 10.9\% improvement versus baseline methods) while even outperforming mainstream test-time adaptation approaches by 3.3\%, and only very few computing resources are required. The code will be publicly available soon. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_08099 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectors Cheng, Runxi Xiong, Feng Wei, Yongxian Zhu, Wanyun Yuan, Chun Machine Learning Model merging seeks to integrate task-specific expert models into a unified architecture while preserving multi-task generalization capabilities, yet parameter interference between constituent models frequently induces performance degradation. Although prior work has explored many merging strategies, resolving interference without additional data for retraining or test-time computation remains challenging. In this paper, we theoretically demonstrate that the task vectors of the linear layer constitute an approximate linear subspace for its corresponding input. Therefore, we can minimize interference under the guidance of task vectors. Based on this insight, we propose \textbf{WUDI-Merging} (\textbf{W}hoever started the interference sho\textbf{U}ld en\textbf{D} \textbf{I}t), a simple yet effective model merging method that eliminates interference without any additional data or rescaling coefficients. Comprehensive empirical evaluations across vision and language benchmarks demonstrate our method's superiority, achieving state-of-the-art performance in data-free model merging scenarios (average 10.9\% improvement versus baseline methods) while even outperforming mainstream test-time adaptation approaches by 3.3\%, and only very few computing resources are required. The code will be publicly available soon. |
| title | Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectors |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2503.08099 |