OptMerge: Unifying Multimodal LLM Capabilities and Modalities via Model Merging

TL;DR

This paper introduces a benchmark and new methods for merging multimodal large language models, demonstrating improved performance and modality complementarity without additional training data.

Contribution

It presents a comprehensive benchmark for MLLM merging, explores merging across modalities, and proposes a noise-removal and optimization technique for better model merging.

Findings

Model merging improves MLLM performance by 2.48% on average.

Multimodal merging outperforms individual modalities.

Proposed method enhances robustness by noise removal from task vectors.

Abstract

Foundation models update slowly due to resource-intensive training, whereas domain-specific models evolve rapidly between releases. Model merging seeks to combine multiple expert models into a single, more capable model, reducing storage and serving costs while supporting decentralized development. Despite its potential, previous studies have primarily focused on merging visual classification models or Large Language Models (LLMs) for code and math tasks. Recently, Multimodal LLMs (MLLMs) that extend LLMs through large-scale multimodal training have gained traction. However, there lacks a benchmark for model merging research that clearly divides the tasks for MLLM training and evaluation. In this paper, $\textbf{(i)}$ we introduce a model merging benchmark for MLLMs, which includes multiple tasks such as VQA, Geometry, Chart, OCR, and Grounding, studying both LoRA and full fine-tuning models. Moreover, we explore how model merging can combine different modalities (e.g., vision-language, audio-language, and video-language models), moving toward the Omni-language model. $\textbf{(ii)}$ We implement 10 model merging algorithms on the benchmark. Furthermore, we propose a novel method that removes noise from task vectors and robustly optimizes the merged vector based on a loss defined over task vector interactions, achieving an average performance gain of 2.48%. $\textbf{(iii)}$ We find that model merging offers a promising way for building improved MLLMs without requiring training data. Our results also demonstrate that the complementarity among multiple modalities outperforms individual modalities.