google/mobile-actions

该数据集包含对话轨迹，旨在训练轻量级模型（如FunctionGemma 270M）将自然语言指令转换为Android OS系统工具的可执行函数调用。数据集以JSONL格式提供，每行代表一个数据样本，预分为训练集和评估集。每个JSON对象包含元数据（用于区分训练或评估）、可用工具列表（包括函数名称、描述和参数）以及消息列表（包含用户输入和预期的函数调用）。

The dataset contains conversational traces designed to train lightweight models (such as FunctionGemma 270M) to translate natural language instructions into executable function calls for Android OS system tools. The dataset is provided in JSONL format, with each line representing a data sample pre-split into training and evaluation sets. Each JSON object contains metadata (for splitting into train or eval), a list of available tools (including function names, descriptions, and parameters), and a list of messages (containing user input and the expected function calls).