jan-hq/sailor-instruct-merge-long

这是一个对话数据集，包含对话ID、消息（包括角色和内容）、语言、分类信息等字段。数据集被划分为训练集，并提供了默认的配置文件。每个样本可能包括多个对话回合，并且包含了评估指标如perplexity（困惑度）和奖励值等。

This is a dialogue dataset containing fields such as conversation ID, messages (including roles and content), language, category information, etc. The dataset is split into a training set and provides a default configuration file. Each sample may include multiple dialogue turns and includes evaluation metrics such as perplexity and reward values.