nl2bash-tasks-cleaned-oracle-qwen3.5-122b-131k-opencode-sft-serveparity
收藏资源简介:
该数据集是一个用于训练和评估涉及工具调用(Tool-Calling)能力的对话模型的数据集。其核心结构围绕多轮对话展开,每条样本包含一个完整的对话序列(messages),其中每条消息标注了发言者角色(如用户、助手)、文本内容以及可能触发的工具调用信息。工具调用详细记录了调用类型、函数名称及其参数。此外,每条样本还提供了对话所使用的工具定义列表(tools)、所属的任务类别(task)、对话总轮次(num_turns)以及工具调用总次数(num_tool_calls)等元数据。数据集共包含 633 个训练样本,适用于研究或开发能够理解用户指令、规划并执行外部工具/API 调用的智能对话系统。
This dataset is designed for training and evaluating dialogue models with Tool-Calling capabilities. Its core structure is centered around multi-turn dialogues, where each sample contains a complete dialogue sequence (messages). Each message in the sequence is annotated with the speaker role (such as user or assistant), text content, and possibly triggered tool-call information. Tool-call details record the call type, function name, and its parameters. Furthermore, each sample also provides a list of tool definitions used in the dialogue (tools), the corresponding task category (task), the total number of dialogue turns (num_turns), and the total number of tool calls (num_tool_calls) as metadata. The dataset contains a total of 633 training samples, and is suitable for researching or developing intelligent dialogue systems that can understand user instructions, plan and execute external tool or API calls.
- 数据集名称:nl2bash-tasks-cleaned-oracle-qwen3.5-122b-131k-opencode-sft-serveparity
- 数据集来源:由 LAION 组织提供,托管于 Hugging Face 平台,地址为 https://huggingface.co/datasets/laion/nl2bash-tasks-cleaned-oracle-qwen3.5-122b-131k-opencode-sft-serveparity
- 数据集规模:
- 训练集(train):633 条样本,占用 22,835,976 字节
- 总下载大小:9,438,384 字节
- 总数据集大小:22,835,976 字节
- 数据集特征:
messages:消息列表,每条消息包含:role(字符串):角色content(字符串):内容tool_calls(列表):工具调用列表,每条工具调用包含:type(字符串):类型function(结构体):函数信息,包含:name(字符串):函数名称arguments(字符串):函数参数
tools(字符串):工具定义task(字符串):任务描述num_turns(整数):对话轮次数量num_tool_calls(整数):工具调用次数
- 数据集配置:
- 配置名称:default
- 数据文件:训练集数据路径为
data/train-*




