How can I quantize a keras model while converting it to a TensorflowJS Layers Model?

Question

I created a tensorflow model and then converted it into tensorflow.js model using below code:

import tensorflowjs as tfjs
from tensorflow.keras.models import load_model

classifier = load_model("model")
tfjs.converters.save_keras_model(classifier, "js")

It works fine and now I would like to reduce the size of the model using quantization when converting it to a TensorflowJS Layers Model.

Answer 1

Yes it is possible you have four options for quantization:

1. --quantization_bytes(depreceated)
2. --quantize_float16
3. --quantize_uint8
4. --quantize_uint16

Example conversion from keras format to tfjs_layers_model:

tensorflowjs_converter \
   --input_format keras \
   --output_format tfjs_layers_model \
   --quantize_uint16 \
   original_model/model.json
   quantized_model/

Or if you need more help with the entire process just type the following in your terminal

tensorflowjs_wizard

More info from Tensorflowjs Converter