Can't find quantize.py for Tensorflow model quantization in develop-tf branch

analogdevicesinc / ai8x-synthesis

Quantization and Synthesis (Device Specific Code Generation) for ADI's MAX78000 and MAX78002 Edge AI Devices

Apache License 2.0

55 stars 49 forks source link

Can't find quantize.py for Tensorflow model quantization in develop-tf branch #139

Closed ramkumarkoppu closed 3 years ago

ramkumarkoppu commented 3 years ago

quantize.py contains code to quantize PyTorch model but couldn't find equivalent script to quantize Tensorflow model as per model training in develop-tf branch of ai8x-training repository. What is the script to quantize Tensorflow model for MAX78000 device?

aniktash commented 3 years ago

In develop-tf branch, the onnx file generated by training is directly quantized by ai8xize synthesis tool with --scale specifying the quantization scale. Using --generate-dequantized-onnx-file, another onnx file that is dequantized is generated to be used for evaluation. the gen-tf-demos-max78000.sh script I nai8x-synthesis develop-tf branch includes all required flags. Please check Post-Training Model Quantization and MAX78000 sections in README in ai8x-Training/TensorFlow for more info.