Hallucination Evaluation
Hallucination evaluation measures the rate at which a model generates factually incorrect or unsupported statements, critical for RAG and knowledge-intensive tasks.
What is Hallucination Evaluation?
Hallucination evaluation measures the rate at which a model generates factually incorrect or unsupported statements, critical for RAG and knowledge-intensive tasks.
Hallucination evaluation measures the rate at which a model generates factually incorrect or unsupported statements, critical for RAG and knowledge-intensive tasks.
Where is it used?
TruthfulQA, HaluEval, and RAGAS faithfulness metrics measure hallucination; production RAG systems use self-check or LLM-judge pipelines to flag ungrounded claims.
How to build it
Use `ragas.metrics.Faithfulness` with `(response, context)` to score groundedness, or run `lm_eval --tasks truthfulqa` and inspect the MC1/MC2 scores for hallucination tendency.