human-sim
收藏资源简介:
Xuhui/human-sim 是一个用于用户模拟的已处理数据集,每条记录代表一个用户及其分组对话。数据集包含多个配置(JinaLeejnl_AlignX、Anthropic_hh-rlhf、ConvLab_dailydialog、Neph0s_CoSER、allenai_WildChat-48M、lmsys_lmsys-chat-1m、neulab_agent-data-collection),每个配置都有训练集和测试集,存储为parquet文件。数据模式中,每条记录包含用户ID(SHA-256哈希IP)、用户元数据(如国家)和该用户的所有对话。对话信息包括对话ID、来源数据集标识、消息列表(角色和内容对)以及元数据(模型、语言、轮次、时间戳、状态、是否脱敏等)。数据集适用于中期训练和用户模拟对话任务,采用'other'许可协议。各配置的用户数量在README中有详细说明。
Xuhui/human-sim is a processed dataset designed for user simulation, where each record corresponds to a single user and their grouped conversations. The dataset encompasses seven distinct configurations: JinaLeejnl_AlignX, Anthropic_hh-rlhf, ConvLab_dailydialog, Neph0s_CoSER, allenai_WildChat-48M, lmsys_lmsys-chat-1m, and neulab_agent-data-collection. Each configuration provides both training and test subsets, which are stored in Parquet file format. Per the data schema, each record contains user ID (SHA-256 hashed IP address), user metadata (e.g., country), and all conversations associated with the user. Conversation details include conversation ID, source dataset identifier, a list of message pairs (comprising role and content), and supplementary metadata including model used, language, turn count, timestamp, status, and anonymization status. This dataset is suitable for intermediate-stage training and user simulation dialogue tasks, and is released under the "other" license. The number of users for each configuration is detailed in the accompanying README file.



