No Task Left Behind: Isotropic Model Merging with Common and Task-Specific Subspaces

TL;DR

This paper introduces an isotropic merging framework that aligns task-specific matrices to effectively combine multiple models, significantly reducing performance gaps and achieving state-of-the-art results in multi-task learning.

Contribution

It proposes a novel isotropic merging method that flattens singular value spectra and incorporates common and task-specific subspaces for improved model merging.

Findings

Achieves state-of-the-art performance on vision and language tasks.

Enhances alignment between task matrices improves merging effectiveness.

Reduces performance gap without additional training.

Abstract

Model merging integrates the weights of multiple task-specific models into a single multi-task model. Despite recent interest in the problem, a significant performance gap between the combined and single-task models remains. In this paper, we investigate the key characteristics of task matrices -- weight update matrices applied to a pre-trained model -- that enable effective merging. We show that alignment between singular components of task-specific and merged matrices strongly correlates with performance improvement over the pre-trained model. Based on this, we propose an isotropic merging framework that flattens the singular value spectrum of task matrices, enhances alignment, and reduces the performance gap. Additionally, we incorporate both common and task-specific subspaces to further improve alignment and performance. Our proposed approach achieves state-of-the-art performance on vision and language tasks across various sets of tasks and model scales. This work advances the understanding of model merging dynamics, offering an effective methodology to merge models without requiring additional training. Code is available at https://github.com/danielm1405/iso-merging .