Struct ActivationQuantizer 

pub struct ActivationQuantizer { /* private fields */ }

Activation quantization for dynamic runtime quantization

Quantizes activations on-the-fly during inference for additional memory savings

Implementations

impl ActivationQuantizer

pub fn new_symmetric() -> Self

Create a new activation quantizer with symmetric per-tensor quantization

pub fn new_asymmetric() -> Self

Create a new activation quantizer with asymmetric per-tensor quantization

pub fn calibrate(&mut self, stats: &CalibrationStats) -> ModelResult<()>

Set calibration parameters from statistics

pub fn quantize_activation_1d( &self, activation: &Array1<f32>, ) -> ModelResult<Vec<i8>>

Quantize activation (1D)

pub fn dequantize_activation_1d( &self, quantized: &[i8], original_len: usize, ) -> ModelResult<Array1<f32>>

Dequantize activation (1D)

pub fn simulate_quantization( &self, activation: &Array1<f32>, ) -> ModelResult<Array1<f32>>

Quantize and immediately dequantize (simulates quantization error)

pub fn memory_savings(&self) -> f32

Get memory savings estimate (percentage)

Trait Implementations

impl Clone for ActivationQuantizer

fn clone(&self) -> ActivationQuantizer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ActivationQuantizer

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

Formats the value using the given formatter.

impl Default for ActivationQuantizer

fn default() -> Self

Returns the “default value” for a type.