pytorch_quantization.calib.histogram.HistogramCalibrator Class Reference

Inheritance diagram for pytorch_quantization.calib.histogram.HistogramCalibrator:

Collaboration diagram for pytorch_quantization.calib.histogram.HistogramCalibrator:

Public Member Functions
def	__init__ (self, num_bits, axis, unsigned, num_bins=2048, grow_method=None, skip_zeros=False)
def	collect (self, x)
def	reset (self)
def	compute_amax (self, str method, *int stride=1, int start_bin=128, float percentile=99.99)
def	__str__ (self)
def	__repr__ (self)
def	compute_amax (self, *args, **kwargs)

Private Member Functions
def	_compute_amax_entropy (self, stride, start_bin)
def	_compute_amax_mse (self, stride, start_bin)
def	_compute_amax_percentile (self, percentile)

Private Attributes
	_num_bins
	_skip_zeros
	_calib_bin_edges
	_calib_hist
	_num_bits
	_axis
	_unsigned

Detailed Description

Unified histogram calibrator

Histogram will be only collected once. compute_amax() performs entropy, percentile, or mse
    calibration based on arguments

Args:
    num_bits: An integer. Number of bits of quantization.
    axis: A tuple. see QuantDescriptor.
    unsigned: A boolean. using unsigned quantization.
    num_bins: An integer. Number of histograms bins. Default 2048.
    grow_method: A string. DEPRECATED. default None.
    skip_zeros: A boolean. If True, skips zeros when collecting data for histogram. Default False.

Constructor & Destructor Documentation

__init__()

def pytorch_quantization.calib.histogram.HistogramCalibrator.__init__	(		self,
			num_bits,
			axis,
			unsigned,
			num_bins = 2048,
			grow_method = None,
			skip_zeros = False
	)

Member Function Documentation

collect()

def pytorch_quantization.calib.histogram.HistogramCalibrator.collect	(		self,
			x
	)

Collect histogram

Reimplemented from pytorch_quantization.calib.calibrator._Calibrator.

reset()

def pytorch_quantization.calib.histogram.HistogramCalibrator.reset

(

self

)

Reset the collected histogram

Reimplemented from pytorch_quantization.calib.calibrator._Calibrator.

_compute_amax_entropy()

def pytorch_quantization.calib.histogram.HistogramCalibrator._compute_amax_entropy (   self,   stride,   start_bin  )

private

Returns amax that minimizes KL-Divergence of the collected histogram

Here is the caller graph for this function:

_compute_amax_mse()

def pytorch_quantization.calib.histogram.HistogramCalibrator._compute_amax_mse (   self,   stride,   start_bin  )

private

Returns amax that minimizes MSE of the collected histogram

Here is the caller graph for this function:

_compute_amax_percentile()

def pytorch_quantization.calib.histogram.HistogramCalibrator._compute_amax_percentile (   self,   percentile  )

private

Returns amax that clips the percentile fraction of collected data

Here is the caller graph for this function:

compute_amax() [1/2]

def pytorch_quantization.calib.histogram.HistogramCalibrator.compute_amax	(		self,
		str	method,
		*int	stride = 1,
		int	start_bin = 128,
		float	percentile = 99.99
	)

Compute the amax from the collected histogram

Args:
    method: A string. One of ['entropy', 'mse', 'percentile']

Keyword Arguments:
    stride: An integer. Default 1
    start_bin: An integer. Default 128
    percentils: A float number between [0, 100]. Default 99.99.

Returns:
    amax: a tensor

Here is the call graph for this function:

__str__()

def pytorch_quantization.calib.histogram.HistogramCalibrator.__str__

(

self

)

__repr__()

def pytorch_quantization.calib.histogram.HistogramCalibrator.__repr__

(

self

)

Reimplemented from pytorch_quantization.calib.calibrator._Calibrator.

compute_amax() [2/2]

def pytorch_quantization.calib.calibrator._Calibrator.compute_amax (   self, *  args, **  kwargs  )

inherited

Abstract method: compute the amax from the collected data

Returns:
    amax: a tensor

Member Data Documentation

_num_bins

pytorch_quantization.calib.histogram.HistogramCalibrator._num_bins

private

_skip_zeros

pytorch_quantization.calib.histogram.HistogramCalibrator._skip_zeros

private

_calib_bin_edges

pytorch_quantization.calib.histogram.HistogramCalibrator._calib_bin_edges

private

_calib_hist

pytorch_quantization.calib.histogram.HistogramCalibrator._calib_hist

private

_num_bits

pytorch_quantization.calib.calibrator._Calibrator._num_bits

privateinherited

_axis

pytorch_quantization.calib.calibrator._Calibrator._axis

privateinherited

_unsigned

pytorch_quantization.calib.calibrator._Calibrator._unsigned

privateinherited

---

The documentation for this class was generated from the following file:

• histogram.py