Scorer Definitions

This section provides formal mathematical definitions for all uncertainty quantification scorers available in UQLM. Each scorer returns a confidence score between 0 and 1, where higher scores indicate a lower likelihood of errors or hallucinations.

For detailed API documentation and usage examples, see the API Reference and Example Notebooks.

Available Scorers

• Black-Box Scorers
  • Normalized Semantic Negentropy
  • Semantic Sets Confidence
  • Non-Contradiction Probability
  • Entailment Probability
  • Exact Match Rate
  • BERTScore
  • Normalized Cosine Similarity
• White-Box Scorers
  • Single-Generation Scorers
  • Multi-Generation Scorers
• LLM-as-a-Judge Scorers
  • Ternary Judge (True/False/Uncertain)
  • Binary Judge (True/False)
  • Continuous Judge
  • Likert Scale Judge
  • Panel of Judges
• Ensemble Scorers
  • BS Detector
  • Generalized Ensemble
• Long-Text Scorers
  • Long-Text Scoring Methods
• Code-Generation Scorers
  • Code-Generation Scoring Methods