arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2407.21772cs.CLcs.LG
收藏

ShieldGemma:基于Gemma的生成式AI内容审核

ShieldGemma: Generative AI Content Moderation Based on Gemma

Google LLC(谷歌公司)
Wenjun Zeng, Yuchi Liu, Ryan Mullins, Ludovic Peran, Joe Fernandez, Hamza Harkous, Karthik Narasimhan, Drew Proud, Piyush Kumar, Bhaktipriya Radharapu, Olivia S… 展开作者

Wenjun Zeng, Yuchi Liu, Ryan Mullins, Ludovic Peran, Joe Fernandez, Hamza Harkous, Karthik Narasimhan, Drew Proud, Piyush Kumar, Bhaktipriya Radharapu, Olivia Sturman, Oscar Wahltinez

AI总结:

本文提出基于Gemma2的ShieldGemma安全内容审核模型套件,在四类伤害类型上超越Llama Guard和WildCard,并引入基于LLM的数据策管流程,主要用合成数据训练即实现强泛化,为LLM安全与内容审核提供开源资源。

摘要:

我们提出ShieldGemma,这是一套基于Gemma2构建的、全面的基于大语言模型(LLM)的安全内容审核模型。这些模型能够在用户输入和LLM生成输出中,对关键伤害类型(色情内容、危险内容、骚扰、仇恨言论)提供稳健的、最先进的安全风险预测。通过在公开和内部基准上的评估,我们展示了相较于现有模型(如Llama Guard,在公开基准上AU-PRC提升10.8%,以及WildCard提升4.3%)的优越性能。此外,我们提出了一种新颖的基于LLM的数据策管流程,可适配于各种安全相关任务及其他领域。我们展示了主要在合成数据上训练的模型具有强大的泛化性能。通过发布ShieldGemma,我们为研究社区提供了宝贵资源,推进LLM安全,并为开发者创建更有效的内容审核解决方案提供了可能。

英文摘要:

We present ShieldGemma, a comprehensive suite of LLM-based safety content moderation models built upon Gemma2. These models provide robust, state-of-the-art predictions of safety risks across key harm types (sexually explicit, dangerous content, harassment, hate speech) in both user input and LLM-generated output. By evaluating on both public and internal benchmarks, we demonstrate superior performance compared to existing models, such as Llama Guard (+10.8\% AU-PRC on public benchmarks) and WildCard (+4.3\%). Additionally, we present a novel LLM-based data curation pipeline, adaptable to a variety of safety-related tasks and beyond. We have shown strong generalization performance for model trained mainly on synthetic data. By releasing ShieldGemma, we provide a valuable resource to the research community, advancing LLM safety and enabling the creation of more effective content moderation solutions for developers.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑