BruceNju/crosswoz-sft

这是一个中文的跨域任务型对话数据集，名为crosswoz-sft，专为大型模型的监督微调任务设计。数据集包括多轮对话、用户目标、对话状态等信息，非常适合训练任务型对话系统。它包含6012个对话和102000个话语，覆盖五个领域：酒店、餐厅、景点、地铁和出租车。大约60%的对话包含跨域用户目标，数据集的对话复杂度高于现有的MultiWOZ等数据集。

This is a Chinese cross-domain task-oriented dialogue dataset named crosswoz-sft, designed for supervision fine-tuning tasks of large-scale models. The dataset includes multi-turn dialogues, user goals, dialogue states, and other information, which is very suitable for training task-oriented dialogue systems. It contains 6,012 dialogues and 102,000 utterances, covering five domains: hotels, restaurants, attractions, subways, and taxis. About 60% of the dialogues contain cross-domain user goals, and the dialogue complexity of the dataset is higher than that of existing datasets like MultiWOZ.