tf.quantization.quantize_and_dequantize | TensorFlow v2.16.1 (original) (raw)

tf.quantization.quantize_and_dequantize

Stay organized with collections Save and categorize content based on your preferences.

Quantizes then dequantizes a tensor. (deprecated)

View aliases

Compat aliases for migration

tf.compat.v1.quantization.quantize_and_dequantize

tf.quantization.quantize_and_dequantize(
    input,
    input_min,
    input_max,
    signed_input=True,
    num_bits=8,
    range_given=False,
    round_mode='HALF_TO_EVEN',
    name=None,
    narrow_range=False,
    axis=None
)

Args
input	A Tensor to quantize and dequantize.
input_min	If range_given=True, the minimum input value, that needs to be represented in the quantized representation. If axis is specified, this should be a vector of minimum values for each slice along axis.
input_max	If range_given=True, the maximum input value that needs to be represented in the quantized representation. If axis is specified, this should be a vector of maximum values for each slice along axis.
signed_input	True if the quantization is signed or unsigned.
num_bits	The bitwidth of the quantization.
range_given	If true use input_min and input_max for the range of the input, otherwise determine min and max from the input Tensor.
round_mode	Rounding mode when rounding from float values to quantized ones. one of ['HALF_TO_EVEN', 'HALF_UP']
name	Optional name for the operation.
narrow_range	If true, then the absolute value of the quantized minimum value is the same as the quantized maximum value, instead of 1 greater. i.e. for 8 bit quantization, the minimum value is -127 instead of -128.
axis	Integer. If specified, refers to a dimension of the input tensor, such that quantization will be per slice along that dimension.

Returns
A Tensor. Each element is the result of quantizing and dequantizing the corresponding element of input.

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates. Some content is licensed under the numpy license.

Last updated 2024-04-26 UTC.