This dataset was generated for span-level hallucination detection in tool-calling dialogues.
It is derived from Team-ACE/ToolACE.
Main file:
generated_data/toolace_ragtruth_all_with_negatives_and_splits.jsonl
Each example contains a query, tool context, final answer, hallucination labels,
hallucination type, split, and tool metadata. Labels are character-level spans
in the answer.
