Struct Quantizer 

pub struct Quantizer<T: FloatBounds> { /* private fields */ }

Main quantization engine

Implementations

impl<T: FloatBounds> Quantizer<T>

pub fn new(config: QuantizationConfig) -> Self

Create new quantizer

pub fn calibrate( &mut self, model_data: &HashMap<String, Array2<T>>, ) -> NeuralResult<()>

Calibrate quantization parameters using sample data

pub fn quantize_tensor( &self, tensor: &Array2<T>, layer_name: &str, ) -> NeuralResult<QuantizedTensor>

Quantize a tensor

pub fn dequantize_tensor( &self, quantized: &QuantizedTensor, ) -> NeuralResult<Array2<T>>

Dequantize a quantized tensor

pub fn quantize_weights( &mut self, weights: &HashMap<String, Array2<T>>, ) -> NeuralResult<HashMap<String, QuantizedTensor>>

Quantize model weights

pub fn fake_quantize_tensor( &self, tensor: &Array2<T>, layer_name: &str, ) -> NeuralResult<Array2<T>>

Apply fake quantization (for QAT)

pub fn compute_quantization_error( &self, original: &Array2<T>, quantized: &QuantizedTensor, ) -> NeuralResult<QuantizationMetrics>

Compute quantization error

pub fn analyze_layer_sensitivity( &mut self, layers_data: &HashMap<String, Array2<T>>, ) -> NeuralResult<HashMap<String, f64>>

Analyze quantization sensitivity