View source on GitHub
|
Module containing quantization code built on Keras abstractions.
Modules
default_8bit module: Module containing 8bit default quantization scheme.
graph_transformations module: Module containing code for graph transformation.
quantizers module: Module containing Quantization abstraction and quantizers.
Classes
class QuantizeConfig: ABC interface for Keras layers to express how they should be quantized.
class QuantizeLayoutTransform: Apply transformations to the model.
class QuantizeRegistry: ABC interface which specifies how layers should be quantized.
class QuantizeScheme: ABC interface which specifies transformer and quantization registry.
class QuantizeWrapper: Quantizes the weights and activations of the keras layer it wraps.
Functions
quantize_annotate_layer(...): Annotate a tf.keras layer to be quantized.
quantize_annotate_model(...): Annotate a tf.keras model to be quantized.
quantize_model(...): Quantize a tf.keras model with the default quantization implementation.
quantize_scope(...): Scope which can be used to deserialize quantized Keras models and layers.