Mozilla/history-search-tool-calling

该数据集名为历史搜索-评估数据集，包含用于历史搜索工具使用的仅评估数据集。它包括三个配置：intent、intent_turns和temporal。intent数据集评估在多轮对话中模型是否应调用历史搜索，每行代表一个对话并保留完整上下文。intent_turns数据集是intent的扁平化视图，每行代表一个对话轮次，仅用于HF数据集查看器和分析。temporal数据集评估调用历史搜索时的时间解析正确性，包含真实开始/结束时间戳和搜索词提取的健全性检查。数据集为英文，可使用提供的代码片段通过pandas加载。

This repository contains evaluation-only datasets for history search tool usage. It includes three configurations: intent, intent_turns, and temporal. The intent dataset evaluates whether the model should invoke history search in a multi-turn conversation, with one row per conversation and preserving full context. The intent_turns dataset is a flattened, one row per turn view of intent, used for HF Dataset Viewer and analysis only. The temporal dataset evaluates temporal parsing correctness when invoking history search, with ground-truth start/end timestamps and sanity check for search term extraction. The datasets are in English and can be loaded with pandas using the provided code snippets.