Struct QuantizationParams 

pub struct QuantizationParams {
    pub scale: f32,
    pub zero_point: i8,
    pub dims: usize,
}

Quantization parameters for int8 scalar quantization

Implements symmetric quantization: Q(x) = round(x / scale) Following Jacob et al. (2018) and Wu et al. (2020)

Fields

scale: f32

Scale factor: absmax / 127.0 for symmetric quantization

zero_point: i8

Zero point (0 for symmetric quantization)

dims: usize

Original embedding dimensions

Implementations

impl QuantizationParams

pub fn new(scale: f32, dims: usize) -> Result<Self, QuantizationError>

Create new quantization parameters

Errors

Returns error if scale is invalid (zero, negative, or non-finite)

pub fn from_absmax(absmax: f32, dims: usize) -> Result<Self, QuantizationError>

Create from absmax value (symmetric quantization)

pub fn quantize_value(&self, value: f32) -> i8

Quantize a single f32 value to i8

pub fn dequantize_value(&self, value: i8) -> f32

Dequantize a single i8 value to f32

pub fn max_error_bound(&self) -> f32

Maximum quantization error bound: scale / 2

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.

impl PartialEq for QuantizationParams

fn eq(&self, other: &QuantizationParams) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for QuantizationParams