SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language Model

TL;DR

This paper introduces SPA-VL, a large-scale, diverse dataset designed to improve safety alignment in Vision Language Models by providing extensive harmfulness coverage and automated data collection from multiple models.

Contribution

The paper presents SPA-VL, a comprehensive safety preference dataset for VLMs, addressing the lack of large, high-quality datasets for safety alignment in multimodal models.

Findings

Models trained on SPA-VL show improved harmlessness.

Enhanced helpfulness in VLMs after training on SPA-VL.

SPA-VL covers extensive harmfulness categories.

Abstract

The emergence of Vision Language Models (VLMs) has brought unprecedented advances in understanding multimodal information. The combination of textual and visual semantics in VLMs is highly complex and diverse, making the safety alignment of these models challenging. Furthermore, due to the limited study on the safety alignment of VLMs, there is a lack of large-scale, high-quality datasets. To address these limitations, we propose a Safety Preference Alignment dataset for Vision Language Models named SPA-VL. In terms of breadth, SPA-VL covers 6 harmfulness domains, 13 categories, and 53 subcategories, and contains 100,788 samples of the quadruple (question, image, chosen response, rejected response). In terms of depth, the responses are collected from 12 open-source (e.g., QwenVL) and closed-source (e.g., Gemini) VLMs to ensure diversity. The construction of preference data is fully automated, and the experimental results indicate that models trained with alignment techniques on the SPA-VL dataset exhibit substantial improvements in harmlessness and helpfulness while maintaining core capabilities. SPA-VL, as a large-scale, high-quality, and diverse dataset, represents a significant milestone in ensuring that VLMs achieve both harmlessness and helpfulness.