Hallucination Detector (DeBERTa)

Tools

• hallucination_detector_deberta_detect Classify an LLM answer as Factual or Hallucinated given a knowledge context. Runs the fine-tuned DeBERTa-v3 detector once for the headline score, then runs Monte-Carlo Dropout passes for an uncertainty estimate (std dev across stochastic forward passes), exactly as in the paper. Returns: (verdict label dict, details markdown).
• hallucination_detector_deberta_generate_and_check Closed-loop pipeline from the paper: the DPO-tuned Qwen2.5-0.5B generator answers a grounded question, and the DeBERTa detector scores its own output. Returns: (generated answer, detector verdict label dict, details markdown).

No tool matches.

As the server's own listing gives them. We never call a tool.