Module: tfmot.quantization.keras.default_8bit.default_8bit_transforms

Module containing 8bit default transforms.

Classes

class ConcatTransform: Transform for Concatenate. Quantize only after concatenation.

class ConcatTransform3Inputs: Transform for Concatenate. Quantize only after concatenation.

class ConcatTransform4Inputs: Transform for Concatenate. Quantize only after concatenation.

class ConcatTransform5Inputs: Transform for Concatenate. Quantize only after concatenation.

class ConcatTransform6Inputs: Transform for Concatenate. Quantize only after concatenation.

class Conv2DBatchNormActivationQuantize: Ensure FQ does not get placed between Conv, BatchNorm and ReLU.

class Conv2DBatchNormQuantize: Ensure FQ does not get placed between Conv and BatchNorm.

class Conv2DBatchNormReLUQuantize: Ensure FQ does not get placed between Conv, BatchNorm and ReLU.

class Conv2DReshapeBatchNormActivationQuantize: Ensure FQ does not get placed between Conv, BatchNorm and ReLU.

class Conv2DReshapeBatchNormQuantize: Ensure FQ does not get placed between Conv, Reshape and BatchNorm.

class Conv2DReshapeBatchNormReLUQuantize: Ensure FQ does not get placed between Conv, BatchNorm and ReLU.

class InputLayerQuantize: Quantizes InputLayer, by adding QuantizeLayer after it.

class LayerReLUQuantize: Ensure FQ does not get placed between Add and ReLU.

class LayerReluActivationQuantize: Ensure FQ does not get placed between Add and ReLU.

class SeparableConv1DQuantize: Add QAT support for Keras SeparableConv1D layer.

class SeparableConvQuantize: Break SeparableConv into a DepthwiseConv and Conv layer.