nemotron-gym-knowledge-web-search-mcqa-qwen3.5-122b-131k-opencode-sft-serveparity
收藏资源简介:
该数据集是一个结构化对话数据集,专门设计用于支持工具调用和多轮对话任务。数据集包含2897个训练样本,总大小约100MB。每个样本由多个核心字段构成:messages字段记录对话历史,每条消息包含role(角色)、content(内容)和tool_calls(工具调用)三个子字段,其中工具调用进一步细分为type(类型)和function(函数)结构,函数包含name(名称)和arguments(参数)。此外,数据集还提供tools(工具定义)、task(任务类型)、num_turns(对话轮次数)和num_tool_calls(工具调用次数)等元数据字段。数据格式表明该数据集适用于训练和评估能够理解工具使用、执行多轮交互的对话系统。
This dataset is a structured dialogue dataset specifically designed to support tool calling and multi-turn dialogue tasks. It contains 2897 training samples with a total size of approximately 100MB. Each sample consists of multiple core fields: the messages field records dialogue history, with each message containing three sub-fields: role (role), content (content), and tool_calls (tool calls), where tool calls are further subdivided into type (type) and function (function) structures, with the function including name (name) and arguments (arguments). Additionally, the dataset provides metadata fields such as tools (tool definitions), task (task type), num_turns (number of dialogue turns), and num_tool_calls (number of tool calls). The data format indicates that this dataset is suitable for training and evaluating dialogue systems capable of understanding tool usage and performing multi-turn interactions.




