Struct QuantizationConfig 

pub struct QuantizationConfig {

    pub quantization_type: QuantizationType,
    pub strategy: QuantizationStrategy,
    pub granularity: Granularity,
    pub calibration_method: CalibrationMethod,
    pub symmetric: bool,
    pub quantize_weights: bool,
    pub quantize_activations: bool,
    pub skip_layers: Vec<String>,
    pub qat_epochs: usize,
    pub qat_learning_rate: Float,
    pub fake_quantize: bool,
    pub observer_momentum: Float,
    pub calibration_samples: usize,
    pub random_state: Option<u64>,
}

Quantization configuration

Fields

quantization_type: QuantizationType

Type of quantization

strategy: QuantizationStrategy

Quantization strategy

granularity: Granularity

Granularity of quantization

calibration_method: CalibrationMethod

Calibration method

symmetric: bool

Whether to use symmetric quantization (zero_point = 0)

quantize_weights: bool

Whether to quantize weights

quantize_activations: bool

Whether to quantize activations

skip_layers: Vec<String>

Layers to skip quantization

qat_epochs: usize

Number of quantization-aware training epochs

qat_learning_rate: Float

Learning rate used during quantization-aware training

fake_quantize: bool

Fake quantization during training

observer_momentum: Float

Observer momentum for running statistics

calibration_samples: usize

Number of calibration samples

random_state: Option<u64>

Random seed

Implementations

impl QuantizationConfig

pub fn quantization_type(self, qtype: QuantizationType) -> Self

Set quantization type

pub fn strategy(self, strategy: QuantizationStrategy) -> Self

Set quantization strategy

pub fn granularity(self, granularity: Granularity) -> Self

Set granularity

pub fn calibration_method(self, method: CalibrationMethod) -> Self

Set calibration method

pub fn symmetric(self, symmetric: bool) -> Self

Set symmetric quantization

pub fn quantize_components(self, weights: bool, activations: bool) -> Self

Configure what to quantize

pub fn skip_layers(self, layers: Vec<String>) -> Self

Set layers to skip

pub fn qat_config(self, epochs: usize, lr: Float) -> Self

Configure quantization-aware training

Trait Implementations

impl Clone for QuantizationConfig

fn clone(&self) -> QuantizationConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizationConfig

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

Formats the value using the given formatter.

impl Default for QuantizationConfig

fn default() -> Self

Returns the “default value” for a type.