How calad0i/hgq works
Main Components
a) Core Components:
• Quantization Engine (src/HGQ/quantizer)
• Binary Operations Module (src/HGQ/bops)
• Neural Network Layers (src/HGQ/layers)
• Proxy System (src/HGQ/proxy)
• Utility Functions (src/HGQ/utils)
b) External Interfaces:
• TensorFlow Integration
• hls4ml Integration
• QKeras Integration
Top Level (User Interface)
• Training Pipeline
• Quantization-Aware Training
• Bitwidth Optimization
• Core Components
• Quantizer Module
• Layer Implementations
• Binary Operations
• Proxy System
• External Integration Layer
• TensorFlow Backend
• hls4ml Export
• QKeras Plugin
Key Elements to Highlight
a) Data Flow:
• Blue: Core quantization components
• Green: Neural network layers
• Orange: External interfaces
• Purple: Resource estimation components
• Gray: Utility and support functions
Component Details
Quantizer Module
• Fixed Point Quantizer
• Precision Derivation
• Unary LUT
Layer Implementation
• Base Layers
• Conv Layers
• Dense Layers
• Experimental Layers (Attention)
Proxy System
• Conversion Logic
• QKeras Plugin
• Precision Derivation
Key Interactions to Emphasize
• Training flow from input to quantized model
• Resource estimation feedback loop
• Conversion path to hls4ml
• Integration points with existing frameworks