Module quantize 

INT8 Quantization Module (ADR-091)

This module provides comprehensive INT8 quantization support for CNN models:

• Phase 1 (params, tensor): Core quantization infrastructure (NEW)
• Phase 2 (calibration): Histogram-based range estimation
• Phase 3 (graph_rewrite): BatchNorm fusion, zero-point optimization, Q/DQ insertion
• Phase 4: Kernel Dispatch - Runtime selection of optimized INT8 kernels

ADR-091 Phase 1 Components (New)

• params: Quantization parameters (scale, zero_point, qmin, qmax)
• tensor: Quantized tensor types with metadata
• Enhanced calibration: CalibrationCollector with MinMax, Percentile, MSE, Entropy methods

Re-exports

pub use params::QuantizationParams as QuantParams;

pub use params::QuantizationScheme;

pub use params::QuantizationMode;

pub use tensor::QuantizedTensor;

pub use tensor::QuantizationMetadata;

pub use calibration::CalibrationHistogram;

pub use calibration::QuantizationParams;

pub use calibration::Quantizer;

pub use graph_rewrite::ComputationGraph;

pub use graph_rewrite::GraphNode;

pub use graph_rewrite::NodeParams;

pub use graph_rewrite::NodeType;

pub use graph_rewrite::fuse_batchnorm_to_conv;

pub use graph_rewrite::fuse_relu;

pub use graph_rewrite::fuse_hardswish;

pub use graph_rewrite::fuse_zp_to_bias;

pub use graph_rewrite::generate_hardswish_lut;

pub use graph_rewrite::insert_qdq_nodes;

Modules

calibration

Calibration and Quantization Parameters (ADR-091 Phase 2)

graph_rewrite

Graph Rewrite Passes for INT8 Quantization (ADR-091 Phase 3)

params

Quantization parameters for INT8 quantization.

tensor

Quantized tensor types with metadata.