SyNAP Model Import and Quantization

• Introduction
  • Online Model Conversion
  • Offline Model Conversion
• Import Workflow Example
  • Use case
  • Preliminary Evaluation
  • Use an Appropriate Input Size
  • Quantize Model
  • Remove Un-necessary Layers
  • Improve Quantization Dataset
  • Per-Channel and KL_divergence Quantization
  • 16-bits Quantization
  • Mixed Quantization
  • Remove Un-Needed Outputs
  • Perform Input Preprocessing with the NPU
  • I Still Can’t Meet My Requirements
• Conclusions