How quic/ai-hub-apps works
Main Components
a. Platform Layer
• Android Apps Container
• Windows Apps Container
• Tutorials Container
b. Runtime Layer
• TensorFlow Lite Runtime
• ONNX Runtime
• Genie SDK (with AI Engine Direct SDK)
c. Hardware Acceleration Layer
• CPU Support
• GPU Support
• NPU Support (Hexagon HTP)
d. Application Categories
• Chat/LLM Applications
• Computer Vision Applications
• Speech Processing Applications
┌─ Applications Layer ─────────────────────────────────────┐
• Android Apps Windows Apps Tutorials │
• ChatApp - ChatApp - LLM Deploy │
• ImageClass - Classification - ONNX Deploy │
• SegmentationApp - ObjectDetection │
• SuperResolution - SuperResolution │
• WhisperKit - Whisper │
• ┬──────────────────────────────────┘
• ↓
┌─ Runtime Layer ──────────────────────────────────────────┐
• TensorFlow Lite ←→ ONNX Runtime ←→ Genie SDK │
• ┬──────────────────────────────────┘
• ↓
┌─ Hardware Acceleration Layer ────────────────────────────┐
• CPU ←→ GPU ←→ NPU (HTP) │
• ┘
Important Relationships
• Dependencies between shared components (like tflite_helpers)
• Hardware acceleration paths for each runtime
• Model configurations to their respective applications
Additional Elements
• Version/compatibility information for supported Snapdragon chipsets
• Data flow paths for each application type
• Supported API versions and OS requirements
Implementation Details
• Use standardized UML notation for components
• Clear labels for each major system
• Notes for specific hardware requirements
• Configuration paths for different deployment scenarios
Optimization Paths
• NPU acceleration paths for supported models
• Fallback paths for non-Snapdragon devices
• Different quantization options (FP16, INT8)
This system design should help developers understand how the different components interact and how to best utilize the framework for their specific needs.