Module quantize 

Model quantization utilities

Provides INT8 quantization for model weights and activations to reduce memory usage and improve inference speed.

Structs

DynamicQuantizer

Dynamic quantization - quantize at runtime

PerChannelQuant

Per-channel quantization for conv/linear layers

QuantParams

Quantization parameters

QuantizedTensor

Quantized tensor representation

Functions

dequantize

Dequantize i8 to f32

dequantize_value

Dequantize single value

quantization_error

Calculate quantization error (MSE)

quantize_value

Quantize single value

quantize_weights

Quantize f32 weights to i8

quantize_with_params

Quantize with given parameters

sqnr

Calculate signal-to-quantization-noise ratio (SQNR) in dB