Module quantization 

Model Quantization Support

This module provides comprehensive quantization support for ML models, enabling efficient deployment on edge devices and reducing model size and inference latency.

Supported Quantization Schemes

• INT8 Quantization: 8-bit integer quantization with configurable ranges
• INT4 Quantization: 4-bit integer quantization for extreme compression
• Per-Tensor Quantization: Single scale/zero-point for entire tensor
• Per-Channel Quantization: Independent scale/zero-point per channel
• Symmetric Quantization: Zero-point = 0 (centered around zero)
• Asymmetric Quantization: Arbitrary zero-point for full range coverage
• Dynamic Quantization: Runtime quantization of activations

Examples

use ipfrs_tensorlogic::{QuantizedTensor, QuantizationScheme, QuantizationConfig};

// Per-tensor INT8 symmetric quantization
let weights = vec![0.5, -0.3, 0.8, -0.1];
let config = QuantizationConfig::int8_symmetric();
let quantized = QuantizedTensor::quantize_per_tensor(&weights, vec![4], config).unwrap();

// Dequantize back to f32
let dequantized = quantized.dequantize();
assert_eq!(dequantized.len(), 4);

// Per-channel quantization for Conv2D weights
let weights = vec![0.5, 0.3, -0.2, -0.4, 0.1, 0.6, -0.5, 0.2]; // 2 channels, 4 elements each
let config = QuantizationConfig::int8_per_channel(2);
let quantized = QuantizedTensor::quantize_per_channel(&weights, vec![2, 4], config).unwrap();

Structs

DynamicQuantizer

Dynamic quantization configuration for runtime quantization

QuantizationConfig

Quantization configuration

QuantizationParams

Quantization parameters for a single channel/tensor

QuantizedTensor

Quantized tensor representation

Enums

CalibrationMethod

Calibration method for determining quantization parameters

QuantizationError

Errors that can occur during quantization operations

QuantizationGranularity

Quantization granularity

QuantizationScheme

Quantization scheme