MARS: Harmonizing Multimodal Convergence via Adaptive Rank Search

TL;DR

MARS introduces an automated, adaptive method for optimizing the fine-tuning of multimodal large language models by balancing training dynamics across modalities, leading to improved performance and efficiency.

Contribution

The paper proposes MARS, a novel framework that uses dual scaling laws to automatically discover optimal rank pairs for balanced multimodal model fine-tuning.

Findings

MARS outperforms baseline methods in multimodal model fine-tuning.

The dual scaling laws effectively predict convergence time and task performance.

MARS provides a robust, automated strategy for optimizing multimodal learning.

Abstract

Fine-tuning Multimodal Large Language Models (MLLMs) with parameter-efficient methods like Low-Rank Adaptation (LoRA) is crucial for task adaptation. However, imbalanced training dynamics across modalities often lead to suboptimal accuracy due to negative interference, a challenge typically addressed with inefficient heuristic methods such as manually tuning separate learning rates. To overcome this, we introduce MARS (Multimodal Adaptive Rank Search), an approach to discover optimal rank pairs that balance training dynamics while maximizing performance. Our key innovation, a proposed framework of dual scaling laws, enables this search: one law models module-specific convergence time to prune the search space to candidates with aligned dynamics, while the other predicts final task performance to select the optimal pair from the pruned set. By re-purposing the LoRA rank as a controller for modality-specific convergence speed, MARS outperforms baseline methods and provides a robust, automated strategy for optimizing MLLM fine-tuning.