发表机构
Wuhan University of Technology; Technical University of Munich; National University of Singapore（武汉理工大学; 慕尼黑工业大学; 新加坡国立大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究医学LVLMs的模型合并，提出涵盖多种成像模态和临床任务类型的MergeMedBench基准，评估现有合并方法，提出赢家通吃方法，该方法简单且无超参数，优于现有方法，为LoRA合并提供新视角和实用基线。
AI 中文摘要
大型视觉语言模型（LVLMs）可通过低秩适应（LoRA）等参数高效微调方法应用于专业医学成像任务，产生了针对特定成像模态和临床场景的专家模型生态系统。然而，在实践中部署多个专家LVLMs会带来大量计算和操作开销。模型合并通过将多个专家模型整合为一个无需重新训练的单一模型提供了一个有前景的解决方案，但在医学领域仍未得到充分探索。在这项工作中，我们首次对医学LVLMs的模型合并进行了系统研究。我们引入了MergeMedBench，这是一个涵盖八种成像模态和多种临床任务类型的综合基准，包括基于两种主流架构构建的16个LoRA微调模型。我们对现有合并方法进行了广泛评估，并进一步提出了赢家通吃方法，这是一种简单且无超参数的方法，只保留专家模型中最具主导性的参数。通过保留控制模型行为的关键参数并丢弃较弱的参数，我们的方法避免了基于平均或对齐策略中固有的信息稀释。尽管简单，赢家通吃方法始终优于现有方法，为LoRA合并提供了新视角，并为未来研究提供了强大的实用基线。
英文摘要
Large vision-language models (LVLMs) can be adapted to specialized medical imaging tasks via parameter-efficient fine-tuning approaches such as low-rank adaptation (LoRA), leading to a growing ecosystem of expert models tailored to specific imaging modalities and clinical scenarios. However, deploying multiple expert LVLMs in practice incurs substantial computational and operational overhead. Model merging provides a promising solution by consolidating multiple experts into a single model without retraining, yet it remains largely unexplored in the medical domain. In this work, we present the first systematic study of model merging for medical LVLMs. We introduce MergeMedBench, a comprehensive benchmark spanning eight imaging modalities and diverse clinical task types, comprising 16 LoRA fine-tuned models built upon two mainstream architectures. We conduct an extensive evaluation of existing merging methods and further propose winner-take-all, a simple and hyperparameter-free approach that retains only the most dominant parameters across expert models. By preserving the critical parameters that govern model behavior and discarding weaker ones, our method avoids the information dilution inherent in averaging- or alignment-based strategies. Despite its simplicity, winner-take-all consistently outperforms existing approaches, offering both a new perspective on LoRA merging and a strong practical baseline for future research.
CommentsProject Page: https://github.com/MedAI-T/MergeMedBench
