module testing.test_utils.quantized_tensor

Short summary

module mlprodict.testing.test_utils.quantized_tensor

Initializes a quantized tensor from float values.

source on GitHub

Classes

class	truncated documentation
QuantizedBiasTensor	Instantiates a quantized tensor (uint8) with bias from a float tensor.
QuantizedTensor	Instantiates a quantized tensor (uint8) from a float tensor.

Functions

function	truncated documentation
test_qlinear_conv	Checks a runtime for operator QLinearConv.

Methods

method	truncated documentation
__init__
__init__	constructor
_init	Initialization when dtype is float32.

Documentation

Initializes a quantized tensor from float values.

source on GitHub

class mlprodict.testing.test_utils.quantized_tensor.QuantizedBiasTensor(data, X_or_scale, W: QuantizedTensor | None = None)

Bases: object

Instantiates a quantized tensor (uint8) with bias from a float tensor.

Parameters:

• data – array

• X_or_scale – a QuantizedTensor or a float

• zero_point – a QuantizedTensor or or None

source on GitHub

__init__(data, X_or_scale, W: QuantizedTensor | None = None)

class mlprodict.testing.test_utils.quantized_tensor.QuantizedTensor(data, scale=None, zero_point=None)

Bases: object

Instantiates a quantized tensor (uint8) from a float tensor.

Parameters:

• data – array

• scale – scale if data.dtype is float32, None otherwise

• zero_point – zero_point if data.dtype is float32, None otherwise

source on GitHub

constructor

__init__(data, scale=None, zero_point=None)

constructor

_init(data)

Initialization when dtype is float32.

mlprodict.testing.test_utils.quantized_tensor.test_qlinear_conv(x: QuantizedTensor, x_shape, w: QuantizedTensor, w_shape, b: QuantizedBiasTensor, y: QuantizedTensor, y_shape, opset=None, runtime='python', pads=None, strides=None, group=None)

Checks a runtime for operator QLinearConv.

Parameters:

• x – QuantizedTensor

• x_shape – shape of X

• w – QuantizedTensor

• w_shape – shape of W

• b – QuantizedBiasTensor or None

• y – expected output, QuantizedTensor or None

• y_shape – shape of Y

• opset – desired onnx opset

• runtime – runtime for OnnxInference

• pads – optional parameter for operator QLinearConv

• strides – optional parameter for operator QLinearConv

• group – optional paramerer for operator QLinearConv

source on GitHub