operators

Additional or modified QDQ operators on top of ORT quantized operators.

Classes

QDQConvTranspose

QDQ for ConvTranspose operator.

QDQCustomOp

By default, ORT does not quantize custom ops.

QDQNormalization

By default, ORT does not quantize Normalization ops.

class QDQConvTranspose

Bases: QDQOperatorBase

QDQ for ConvTranspose operator.

__init__(onnx_quantizer, onnx_node)

ConvTranspose quantizer init.

quantize()

Main function to quantize the ConvTranspose ops.

class QDQCustomOp

Bases: QDQOperatorBase

By default, ORT does not quantize custom ops. This module is intended to help with that.

Note. QDQOperatorBase is not sufficient for dynamic input and output only quantization.

__init__(onnx_quantizer, onnx_node)

Normalization quantizer init.

quantize()

Main function to quantize the custom ops.

class QDQNormalization

Bases: QDQOperatorBase

By default, ORT does not quantize Normalization ops. This module is intended to help with that.

Note. QDQOperatorBase is not sufficient for dynamic input only quantization.

__init__(onnx_quantizer, onnx_node)

Normalization quantizer init.

quantize()

Main function to quantize the Normalization ops.