Resonans/tooltrace-verifier

ToolACE幻觉跨度数据集是一个基于ToolACE派生的数据集，专门用于检测工具调用对话中的跨度级幻觉。该数据集包含四个配置：contradiction（矛盾）、overgeneration（过度生成）、missing_tool（缺失工具）和all（全部），每个配置对应一个训练分割的数据文件（parquet格式）。数据字段包括Query（查询）、Context（上下文）、Output（输出）、Hallucination_labels（幻觉标签）、available_tools（可用工具）、tool_call（工具调用）和tool_output（工具输出）。数据集规模分别为：contradiction配置18382行，overgeneration配置18146行，missing_tool配置18290行，all配置23150行。该数据集旨在支持自然语言处理任务，特别是针对幻觉检测在工具交互对话中的应用。ToolACE hallucination spans is a ToolACE-derived dataset designed for span-level hallucination detection in tool-calling dialogs. It includes four configurations: contradiction, overgeneration, missing_tool, and all, each corresponding to a training split data file in parquet format. The fields in the dataset are Query, Context, Output, Hallucination_labels, available_tools, tool_call, and tool_output. The dataset sizes are: contradiction with 18382 rows, overgeneration with 18146 rows, missing_tool with 18290 rows, and all with 23150 rows. This dataset is intended to support natural language processing tasks, specifically focusing on hallucination detection in tool-interaction dialogs.