ToolACE RAGTruth-Style Hallucination Detection Dataset

A dataset for detecting hallucinations in tool-calling systems, built from ToolACE tool-use dialogues. It is formatted for RAGTruth-style evaluation tasks, focusing on classifying whether an assistant's answer is grounded in a tool response. The dataset was created by Ali-Bhai and last updated on Hugging Face in May 2026.

Use Cases

• Benchmarking hallucination detection models based on the described query-context-output structure.
• Training classifiers to identify 'tool_output_conflict' and 'overgeneration' error types mentioned in the description.
• Evaluating the grounding of language model responses in provided tool outputs.
• Studying failure modes in tool-use dialogue systems.

Strengths

• Dataset is specifically designed for hallucination detection in tool-calling, a defined NLP task.
• Built from ToolACE dialogues, suggesting a foundation in real tool-use interactions.
• Last updated timestamp is provided: 2026-05-23 06:53:25.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face, user Ali-Bhai.

Collection Method

Built from ToolACE tool-use dialogues and converted into a RAGTruth-style format.

Freshness

Last updated 2026-05-23 06:53:25.