ACE-Merging: Data-Free Model Merging with Adaptive Covariance Estimation

TL;DR

ACE-Merging introduces a data-free, theoretically grounded method for model merging that estimates input covariance from parameter differences, effectively reducing interference among task-specific models.

Contribution

It presents ACE-Merging, a novel adaptive covariance estimation framework with a closed-form solution that outperforms prior data-free merging methods on vision and language benchmarks.

Findings

ACE-Merging achieves a 4% average improvement over previous methods across seven tasks.

It provides a state-of-the-art data-free model merging technique for vision and language models.

The approach is computationally efficient due to its closed-form solution.

Abstract

Model merging aims to combine multiple task-specific expert models into a single model while preserving generalization across diverse tasks. However, interference among experts, especially when they are trained on different objectives, often leads to significant performance degradation. Despite recent progress, resolving this interference without data access, retraining, or architectural modification remains a fundamental challenge. This paper provides a theoretical analysis demonstrating that the input covariance of each task, which is a key factor for optimal merging, can be implicitly estimated from the parameter differences of its fine-tuned model, even in a fully data-free setting. Building on this insight, we introduce \acem, an Adaptive Covariance Estimation framework that effectively mitigates inter-task interference. Our approach features a principled, closed-form solution that contrasts with prior iterative or heuristic methods. Extensive experiments on both vision and language benchmarks demonstrate that \acem sets a new state-of-the-art among data-free methods. It consistently outperforms existing baselines; for example, \acem achieves an average absolute improvement of 4\% over the previous methods across seven tasks on GPT-2. Owing to its efficient closed-form formulation, \acem delivers superior performance with a modest computational cost, providing a practical and theoretically grounded solution for model merging.