Enum CalibrationMethod 

pub enum CalibrationMethod {
    MinMax,
    Percentile(f32),
    Entropy,
    MSE,
    Adaptive,
    Custom(Arc<dyn CalibrationFunction>),
}

Calibration methods for quantization parameter optimization

Different calibration methods offer trade-offs between computational cost, robustness to outliers, and final quantization accuracy.

Variants

MinMax

Simple min-max calibration

Uses the minimum and maximum values in the data to set quantization range. Fast but sensitive to outliers.

Percentile(f32)

Percentile-based calibration

Uses a specified percentile to clip outliers before determining range. More robust than min-max but requires tuning the percentile parameter.

Entropy

Entropy-based calibration (KL divergence minimization)

Minimizes the KL divergence between original and quantized distributions. Provides good accuracy but is computationally expensive.

MSE

Mean squared error minimization

Finds parameters that minimize MSE between original and quantized values. Good balance between accuracy and computational cost.

Adaptive

Adaptive method selection

Automatically selects the best method based on data characteristics. Uses multiple methods and picks the one with best validation score.

Custom(Arc<dyn CalibrationFunction>)

Custom calibration with user-defined function

Allows users to provide their own calibration logic for specialized use cases or domain-specific optimization.

Trait Implementations

impl Clone for CalibrationMethod

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for CalibrationMethod

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.