tfmot.quantization.keras.quantize_scope

Stay organized with collections Save and categorize content based on your preferences.

Scope which can be used to deserialize quantized Keras models and layers.

Used in the notebooks

Used in the guide

Under quantize_scope, Keras methods such as tf.keras.load_model and tf.keras.models.model_from_config will be able to deserialize Keras models and layers which contain quantization classes such as QuantizeConfig and Quantizer.

Example:

tf.keras.models.save_model(quantized_model, keras_file)

with quantize_scope():
  loaded_model = tf.keras.models.load_model(keras_file)

# If your quantized model uses custom objects such as a specific `Quantizer`,
# you can pass them to quantize_scope to deserialize your model.
with quantize_scope({'FixedRangeQuantizer', FixedRangeQuantizer}
  loaded_model = tf.keras.models.load_model(keras_file)

For further understanding, see tf.keras.utils.custom_object_scope.

*args Variable length list of dictionaries of {name, class} pairs to add to the scope created by this method.

Object of type CustomObjectScope with quantization objects included.