Struct QuantizationParams 

pub struct QuantizationParams {
    pub min: Vec<f32>,
    pub max: Vec<f32>,
    pub dim: usize,
}

Per-dimension quantization calibration: lower and upper bounds.

Fields

min: Vec<f32>

max: Vec<f32>

dim: usize

Implementations

impl QuantizationParams

pub fn fit(vectors: &[Vec<f32>]) -> Result<Self>

Fit calibration bounds from a sample of vectors. If a dimension is constant (min == max) a tiny epsilon range is used so the value quantizes to a stable code rather than dividing by zero.

pub fn quantize(&self, vector: &[f32]) -> Vec<i8>

Quantize a single f32 vector into Int8 codes.

pub fn dequantize(&self, codes: &[i8]) -> Vec<f32>

Dequantize an Int8 code back to an approximate f32 vector.

Trait Implementations

impl Clone for QuantizationParams

fn clone(&self) -> QuantizationParams

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizationParams

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

Formats the value using the given formatter.