面向安全对齐多模态大模型的持续视觉指令微调中的和谐参数适配
Harmonious Parameter Adaptation in Continual Visual Instruction Tuning for Safety-Aligned MLLMs
对做多模态大模型持续指令微调的人，这篇给出了一个后训练框架：按参数对安全/任务的关注度做分区，再从平衡视角选择要保留的参数子集，并对更新施加正交约束，可直接迁移到视觉指令微调的增量学习场景以同时抑制任务遗忘与安全退化。
原文依据 we propose Harmonious Parameter Adaptation (HPA), a post-training framework composed of focusing-based parameter partition
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论尽管持续视觉指令微调（CVIT）在适配多模态大语言模型（MLLMs）方面已展现出潜力，但现有研究主要关注未经安全对齐的模型。这一关键疏漏忽视了现实世界中的 MLLMs 本身就要求此类机制以缓解潜在风险。在本工作中，我们将关注点转向面向安全对齐 MLLMs 的 CVIT，并观察到在持续适配过程中，模型不仅遭受任务遗忘，其安全性也出现退化。在安全性与任务性能之间取得和谐平衡仍是一项关键挑战。为此，我们提出和谐参数适配（HPA），这是一个由基于关注度的参数划分、和谐平衡的参数选择以及正交参数调整组成的后训练框架。具体而言，HPA 依据参数对安全性或任务性能的关注将其划分为两类，并从平衡的视角选择需保留的关注参数。此外，HPA 对参数更新施加正交约束，以进一步缓解灾难性遗忘。在 CVIT 基准与安全评测数据集上的大量实验表明，相较于现有基线，HPA 能更好地保持高安全性并缓解遗忘。代码见 https://github.com/Minato-Zackie/HPA。
英文原文摘要 1371 字符
While continual visual instruction tuning (CVIT) has shown promise in adapting multimodal large language models (MLLMs), existing studies predominantly focus on models without safety alignment. This critical oversight ignores the fact that real-world MLLMs inherently require such mechanisms to mitigate potential risks. In this work, we shift our focus to CVIT for safety-aligned MLLMs and observe that during continual adaptation, the model not only suffers from task forgetting but also exhibits degradation in its safety. Achieving a harmonious balance between safety and task performance remains a crucial challenge. To address this, we propose Harmonious Parameter Adaptation (HPA), a post-training framework composed of focusing-based parameter partition, harmoniously balanced parameter selection, and orthogonal parameter adjustment. Specifically, HPA partitions parameters into two types based on their focus on safety or task performance, and selects the focused ones to preserve from a balanced perspective. In addition, HPA imposes orthogonality constraints on parameter updates to further alleviate catastrophic forgetting. Extensive experiments on the CVIT benchmark and safety evaluation datasets demonstrate that HPA better maintains high safety and mitigates forgetting than existing baselines. Code is available at https://github.com/Minato-Zackie/HPA.