Graft: Integrating the Domain Knowledge via Efficient Parameter Synergy for MLLMs

TL;DR

This paper introduces Graft, a framework for integrating domain-specific knowledge into multimodal large language models through efficient parameter fusion guided by compatibility scores, enhancing their adaptability and performance.

Contribution

The paper proposes a novel Compatibility-Aware Parameter Splicing (CAPS) strategy and a domain compatibility scoring mechanism for modular, efficient integration of heterogeneous domain expertise in MLLMs.

Findings

Effective knowledge sharing across domain-specific MLLMs demonstrated.

Significant performance improvements on diverse multimodal benchmarks.

Minimal inference overhead achieved through low-rank adaptation layers.

Abstract

Multimodal Large Language Models (MLLMs) have achieved success across various domains. However, their applicability tends to degrade when confronted with different types of data inputs, especially for MLLMs that have been fine-tuned for specific tasks. Despite its importance, the study of knowledge sharing among domain-specific MLLMs--such as those trained for mathematics or code--remains largely underexplored. To address the fragmentation of knowledge across domain-specialized MLLMs, we propose a unified parameter integration framework that enables modular composition of expert capabilities. Our method is grounded in a novel Compatibility-Aware Parameter Splicing (CAPS) strategy, which leverages both local functional attribution and global information-theoretic signals to guide selective parameter fusion. By extending this mechanism to the low-rank adaptation layer granularity, we ensure efficient integration with minimal inference overhead. Furthermore, we introduce a domain compatibility scoring mechanism that quantifies inter-expert alignment at the activation level and correlates with downstream task utility. This principled fusion protocol allows the final model to synergize heterogeneous expertise while preserving structural modularity. Extensive evaluations across diverse multimodal benchmarks validate the effectiveness of our framework, offering a scalable path toward compositional, domain-adaptive MLLMs.