Normalization

This section describes the normalization methods and configurations available in LM-Polygraph for converting raw uncertainty scores into interpretable confidence values.

Contents:

• LM-Polygraph Normalization Methods
  • MinMax Normalization (MinMaxNormalizer in minmax.py)
  • Quantile Normalization (QuantileNormalizer in quantile.py)
  • Performance-Calibrated Confidence (PCC) Methods
  • Common Interface: BaseUENormalizer
  • Key Benefits of PCC Methods
  • Highlight: Isotonic PCC
• Core Normalization Configuration
  • Overview
  • Base Configuration Location
  • Available Normalization Methods
  • Common Parameters
  • Usage Examples
  • Best Practices
  • Integration with Other Configs
• Dataset-Specific Normalization Configurations
  • Overview
  • Configuration Structure
  • Usage Examples
  • Key Considerations
  • Best Practices
• Instruction-Tuned Model Normalization Configurations
  • Overview
  • Configuration Structure
  • Model-Specific Configurations
  • Integration Features
  • Best Practices
  • Example Configurations
  • Common Issues and Solutions
• LM-Polygraph Normalization: Impact Areas and Default Behaviors
  • Normalization Impact Areas
  • Default Behaviors
  • Usage Guidelines