LLM Red Teaming datasets

LLM Red Teaming数据集由Innodata创建，包含14个新颖的数据集，用于评估大型语言模型在企业任务中的安全性。这些数据集涵盖了多种任务和安全向量，旨在通过严格的测试评估模型的准确性和安全性。数据集包括半合成数据和人工制作的数据，用于测试模型在事实性、毒性、偏见和幻觉倾向等方面的表现。该数据集的应用领域广泛，旨在解决企业环境中使用语言模型时可能遇到的安全问题。

The LLM Red Teaming Dataset, created by Innodata, consists of 14 novel datasets intended for evaluating the safety of large language models (LLMs) in enterprise tasks. These datasets cover a broad range of tasks and safety dimensions, aiming to rigorously assess the accuracy and security of LLMs through stringent testing protocols. The datasets encompass both semi-synthetic data and manually curated data, which are designed to evaluate model performance across key metrics including factuality, toxicity, bias, and hallucination tendencies. This dataset suite has wide-ranging application scenarios, with the primary goal of addressing potential safety issues that may emerge when deploying language models in enterprise environments.