---
title: "Shieldstral：3B参数的多模态安全分类器，性能媲美大模型"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.25857v1"
permalink: "https://feed.astralor.com/items/itm_9f1a439bae05dc26828f0064"
published: "2026-07-28T15:27:53.000Z"
score: 77
channel: "tech"
tags: ["AI安全","Shieldstral","内容安全","多模态分类"]
---

# Shieldstral：3B参数的多模态安全分类器，性能媲美大模型

## AI 点评

> 这篇论文最值得关注的点在于它用很小的模型成本实现了大模型级别的安全分类性能，对于预算有限的团队或需要低延迟部署场景非常实用。数据构建的配方也很值得借鉴——如何把不同来源的安全数据集统一起来训练。

## 速览

Shieldstral是一个仅3B参数的多模态安全分类器，通过将内容审核建模为二元问答任务，统一了多种安全数据集。它在大约5410万样本上训练，在文本安全基准上匹配或超越近7倍大小的模型，并在多模态安全分类上达到新最佳。该工作展示了小模型通过精心设计的数据配方和训练框架可实现高效策略适应。

## 关键点

- 仅3B参数，性能超越近7倍大小的模型
- 将内容审核统一为二元问答任务，合并异质安全数据集
- 使用约5410万样本训练，包含细粒度评估集
- 在多模态安全分类上达到新最佳

## 全文

We introduce Shieldstral, a 3B-parameter policy-adaptive multimodal safety classifier that matches or outperforms models nearly 7$\times$ its size on text safety benchmarks and sets a new state of the art on multimodal safety classification. Shieldstral formulates content moderation as a binary question-answering task. This simple formulation unifies diverse moderation tasks into a single yes/no problem, enabling heterogeneous safety datasets with divergent taxonomies to be consolidated under one training framework. We present the data construction recipe, covering curation and generation of approximately 54.1M samples and a fine-grained evaluation set to evaluate policy adaptability. Together, these enable a small adaptive model to match or outperform much larger models.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_9f1a439bae05dc26828f0064