tfmot.quantization.keras.QuantizeLayoutTransform

Apply transformations to the model.

Transforms the original model to perform better while quantized and also match the layout of the target backend.

Methods

apply

View source

@abc.abstractmethod
apply(
    model, layer_quantize_map
)

Transform model to a quantization friendly model.

Args
model Keras model to be quantized.
layer_quantize_map Map containing list of layers to be quantized and associated metadata. Keys are layer names which need to be quantized, and values are dicts containing relevant metadata. For example, any custom QuantizeConfig passed with a layer is present.

Returns
New keras model based on model which has been transformed to match the layout of the target backend.