Kkii

Prompt

Measures factual recall reliability, specifically tracking a model's knowledge accuracy while penalizing hallucinations and rewarding it when it correctly abstains from answering things it doesn't know. What does it mean?