GuardReasoner-VL: Safeguarding VLMs via Reinforced Reasoning

TL;DR

This paper presents GuardReasoner-VL, a reasoning-based guard model for vision-language models that uses reinforcement learning and a large reasoning corpus to improve moderation safety and performance.

Contribution

It introduces a novel reasoning-based guard model with reinforcement learning, a large reasoning corpus, and safety-aware training techniques for VLM moderation.

Findings

Surpasses baseline by 19.27% F1 score

Uses online RL with safety-aware rewards

Constructs a large reasoning corpus with 123K samples

Abstract

To enhance the safety of VLMs, this paper introduces a novel reasoning-based VLM guard model dubbed GuardReasoner-VL. The core idea is to incentivize the guard model to deliberatively reason before making moderation decisions via online RL. First, we construct GuardReasoner-VLTrain, a reasoning corpus with 123K samples and 631K reasoning steps, spanning text, image, and text-image inputs. Then, based on it, we cold-start our model's reasoning ability via SFT. In addition, we further enhance reasoning regarding moderation through online RL. Concretely, to enhance diversity and difficulty of samples, we conduct rejection sampling followed by data augmentation via the proposed safety-aware data concatenation. Besides, we use a dynamic clipping parameter to encourage exploration in early stages and exploitation in later stages. To balance performance and token efficiency, we design a length-aware safety reward that integrates accuracy, format, and token cost. Extensive experiments demonstrate the superiority of our model. Remarkably, it surpasses the runner-up by 19.27% F1 score on average. We release data, code, and models (3B/7B) of GuardReasoner-VL at https://github.com/yueliu1999/GuardReasoner-VL/