This project aims to implement and evaluate quantization techniques on machine learning models deployed to Jetson Nano edge devices.
Install dependencies: pip install -r requirements.txt
Create quantized model from scratch: make
- Extract unquantized parameters:
make setup - Calculate quantized parameters:
make quantize_params - Create calibration prep model:
make prep_model - Calculate quantized activations:
make quantize_activations - Calculate quantized biases:
make quantize_biases - Fully quantize model:
make quantize_model
Validate quantized parameters: make validate
Clear activations/, biases/, params/, and models/ directories: make clean