Struct TensorQuantizer 

pub struct TensorQuantizer { /* private fields */ }

Calibration-based tensor quantizer.

Usage: call calibrate with representative data, then compute_params to derive quantization parameters.

Implementations

impl TensorQuantizer

pub fn new() -> Self

Create a new uncalibrated quantizer.

pub fn calibrate(&mut self, values: &[f64])

Update calibration range from observed values.

Can be called multiple times; min/max accumulate across calls.

pub fn compute_params( &self, mode: QuantMode, bits: QuantBits, ) -> Result<QuantParams, String>

Compute quantization parameters from calibration data.

Returns an error if no samples have been observed.

pub fn quantize(values: &[f64], params: &QuantParams) -> Vec<i32>

Quantize floating-point values to integers using the given parameters.

Each value is mapped as: clamp(round(value / scale) + zero_point, qmin, qmax)

pub fn dequantize(quantized: &[i32], params: &QuantParams) -> Vec<f64>

Dequantize integer values back to floating-point.

Each value is mapped as: (q - zero_point) * scale

pub fn quantization_error(original: &[f64], params: &QuantParams) -> f64

Compute mean squared error between original values and their quantize→dequantize roundtrip.

pub fn reset_calibration(&mut self)

Reset calibration state.

pub fn stats(&self) -> QuantizerStats

Return current calibration statistics.

Trait Implementations

impl Default for TensorQuantizer

fn default() -> Self

Returns the “default value” for a type.