Granite Guardian
收藏资源简介:
Granite Guardian数据集是由IBM研究院创建的一个用于训练风险检测模型的数据集,旨在检测大型语言模型(LLM)中的多种风险,包括社会偏见、暴力、性内容等。该数据集结合了来自不同来源的人类标注和合成数据,确保了标注的高质量。数据集包含7000条独特的(提示,响应)对,用于训练和评估模型。该数据集的应用领域主要集中在确保LLM的安全和负责任使用,解决模型在实际部署中可能遇到的各种风险问题。
The Granite Guardian dataset, developed by IBM Research, is a resource for training risk detection models aimed at detecting various risks within large language models (LLMs), including social bias, violent content, sexually explicit material, and more. This dataset combines human-annotated data and synthetic data from diverse sources to ensure high-quality labeling. It contains 7,000 unique (prompt, response) pairs for model training and evaluation. Its primary application domains focus on ensuring the safe and responsible use of LLMs, and addressing various risk issues that models may encounter during real-world deployment.

- 1Granite GuardianIBM研究院 · 2024年



