ToolACE Hallucination Detection Dataset for Tool-Calling Dialogues

1,347 tool-calling dialogues annotated for hallucination detection, generated for the assignment Hallucination Detection in Tool Calling. The dataset uses a RAGTruth-style schema with character-level hallucination span labels. It was created by author HASSANI8046 and last updated on 2026-05-26.

Use Cases

• Training models for hallucination detection based on character-level hallucination span labels.
• Evaluating the factual accuracy of LLM tool-calling outputs based on the provided context.
• Benchmarking RAG systems on tool-use dialogues using the RAGTruth-style schema.
• Analyzing error patterns in tool-calling assistants using the annotated query-context-output triples.

Strengths

• Dataset contains 1,347 annotated tool-calling dialogues.
• Provides character-level hallucination span labels for precise error localization.
• Follows a structured RAGTruth-style schema with query, context, and output fields.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known only for one file; overall dataset scale may be limited.
• Freshness should be verified as the last update timestamp is in the future (2026-05-26).

Provenance

Source

Based on ToolACE tool-calling dialogues.

Collection Method

Generated for the assignment Hallucination Detection in Tool Calling.

Freshness

Last updated 2026-05-26 19:39:11