RUSlogitlens_220_from_all
收藏资源简介:
CValues-Comparison是一个用于大语言模型安全对齐的数据集,属于CValues系列数据集的一部分。该数据集包含中文和英文两个子集,专门设计用于训练和评估模型在价值观对齐方面的能力。数据集采用JSONL格式,每条数据包含prompt(用户输入)和response(模型响应)两个字段,其中中文子集包含约21,000条数据,英文子集包含约11,000条数据。这些数据基于人工标注的安全对齐比较,通过对比安全响应与非安全响应,为直接偏好优化(DPO)等对齐方法提供训练数据。该数据集适用于大语言模型的安全对齐、价值观对齐、有害内容检测等相关任务的研究与开发。
CValues-Comparison is a dataset for safety alignment of large language models, part of the CValues series. It includes Chinese and English subsets, specifically designed for training and evaluating models in value alignment. The dataset is in JSONL format, with each entry containing prompt (user input) and response (model response) fields. The Chinese subset contains approximately 21,000 entries, and the English subset contains approximately 11,000 entries. These data are based on human-annotated safety alignment comparisons, providing training data for alignment methods such as Direct Preference Optimization (DPO) by comparing safe and unsafe responses. The dataset is suitable for research and development in safety alignment, value alignment, harmful content detection, and related tasks for large language models.
根据您提供的数据集详情页面地址和README文件内容,以下是该数据集的关键信息概述:
数据集概述
- 数据集名称:RUSlogitlens_220_from_all
- 数据集来源:Hugging Face 数据集平台
- 许可证:MIT 许可证(开源且允许自由使用、修改和分发)
补充说明
- 该README文件内容非常简短,除许可证信息外未提供数据集的详细描述、使用说明、数据样例或其他元数据。
- 建议访问数据集页面(https://huggingface.co/datasets/mooooosha/RUSlogitlens_220_from_all)获取更完整的信息,如数据规模、字段定义、构建方法等。




