nuriasane/llm-hallucination-detector

LLM Hallucination Detector

A logistic regression classifier trained on sentence embeddings to detect hallucinations in LLM-generated responses.

Classes

• factual - the answer is correct
• partial - the answer contains some errors
• hallucination - the answer is fabricated or incorrect

Usage

The model takes a (question, answer) pair, embeds both with sentence-transformers/all-MiniLM-L6-v2, concatenates the vectors, and predicts the class.

Demo

Try it on Hugging Face Spaces

Dataset

nuriasane/llm-hallucination-detection