大模型内容安全文本数据集

“大模型内容安全文本数据集”，包含各种可能违规、不当或有害内容，如恶意攻击、人身攻击、歧视言论、色情、暴力等语料，可用于大模型的内容安全训练学习和评测。1）作为训练集，可提升大模型对各类负面指令的识别和回答能力，更加安全可靠、最大程度地防止模型产生有害输出；2）作为测试集，用于大模型的内容安全测试，评估大模型在针对各类负面指令时是否具有拒识或按照正确的价值观来回答问题的能力，确保大模型输出的内容符合社会价值观。

Large Model Content Safety Text Dataset contains various corpus of violating, inappropriate or harmful content, such as malicious attacks, personal attacks, discriminatory remarks, pornography, violence and other similar materials. It can be used for content safety training, learning and evaluation of large language models (LLMs). 1) As a training set: it can enhance the ability of LLMs to recognize and handle various negative instructions, making the models more secure and reliable, and maximizing the prevention of harmful model outputs; 2) As a test set: it is used for content safety testing of LLMs, to evaluate whether the models have the ability to reject harmful instructions or respond in accordance with correct social values, so as to ensure that the outputs of LLMs conform to social values.