<ul data-eligibleForWebStory="true"><li>Model merging aims to integrate task-specific expert models into a unified architecture while maintaining multi-task generalization capabilities.</li><li>Parameter interference between models often leads to reduced performance.</li><li>Resolving interference without extra data or computations during testing is a challenge.</li><li>The paper suggests minimizing interference by utilizing task vectors in the linear layer.</li><li>A method called WUDI-Merging is proposed, focusing on eliminating interference without additional data or rescaling coefficients.</li><li>Empirical evaluations across vision and language benchmarks show the effectiveness of the method in data-free model merging.</li><li>WUDI-Merging surpasses baseline methods by an average improvement of 10.9% and even outperforms mainstream test-time adaptation approaches by 3.3%.</li><li>The method exhibits superior performance while requiring minimal computing resources.</li><li>The code for WUDI-Merging will be made publicly available soon.</li></ul>

Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectors

Discover more