tomaarsen/zelo-pairs-10kx100-quantile-anchor

该数据集是一个用于对话系统或信息检索任务的数据集，包含查询、文档对和对话消息。每个样本包括一个查询标识符（query_id）、查询领域（domain）、查询文本（query）、两个相关文档（doc_a和doc_b）及其索引（doc_a_idx和doc_b_idx），以及一个布尔值（swapped）指示文档是否交换。此外，数据集提供了教师标签（teacher_label_a和teacher_label_b），可能用于监督学习或偏好建模。对话消息（messages）以列表形式存储，包含角色（如用户或助手）和内容，适用于训练对话模型。数据集包含训练集，有1,001,000个示例，总大小约2.68 GB。This dataset is designed for dialogue systems or information retrieval tasks, containing queries, document pairs, and dialogue messages. Each sample includes a query identifier (query_id), query domain (domain), query text (query), two relevant documents (doc_a and doc_b) with their indices (doc_a_idx and doc_b_idx), and a boolean value (swapped) indicating whether the documents are swapped. Additionally, the dataset provides teacher labels (teacher_label_a and teacher_label_b), likely used for supervised learning or preference modeling. The dialogue messages (messages) are stored as a list with roles (e.g., user or assistant) and content, suitable for training dialogue models. The dataset includes a training split with 1,001,000 examples and a total size of approximately 2.68 GB.