skysight-inc/synthetic-humans-1m

这个数据集包含了100万合成的美国人，基于实际的美国人口统计抽样。它主要用于为大型语言模型提供多样化的响应种子，也可以用于分析目的。数据集中的qualitative_descriptions列包含了大约24亿个token，由Qwen/QwQ-32B模型生成，并带有完整的推理轨迹。数据集的结构包括唯一标识符、人口统计信息（如年龄、性别、地点、职业类别和年薪）、由LLM生成的描述以及从描述中提取的结构化信息。

This dataset contains 1 million synthetic humans based on actual US demographics. It is primarily intended to seed diverse responses for large language models and can also be used for analytical purposes. The qualitative_descriptions column contains approximately 2.4 billion tokens, generated by the Qwen/QwQ-32B model with full reasoning traces. The dataset structure includes a unique identifier, demographic information (such as age, gender, location, occupation category, and median annual wage), LLM-generated descriptions, and structured extractions from those descriptions.