aihwkit.simulator.digital_low_precision.range_estimators module

class aihwkit.simulator.digital_low_precision.range_estimators.AllMinMaxEstimator(*args, **kwargs)

Bases: RangeEstimatorBase

forward(x)

Accepts an input tensor, updates the current estimates of x_min and x_max and retruns them. :param x: :type x: Input tensor

Returns:

• self.current_xmin (tensor)

• self.current_xmax (tensor)

class aihwkit.simulator.digital_low_precision.range_estimators.CrossEntropyEstimator(*args, **kwargs)

Bases: MSE_Estimator

loss_fx(data, neg_thr, pos_thr, per_channel_loss=False)

class aihwkit.simulator.digital_low_precision.range_estimators.CurrentMinMaxEstimator(percentile=None, *args, **kwargs)

Bases: RangeEstimatorBase

forward(x)

Accepts an input tensor, updates the current estimates of x_min and x_max and retruns them. :param x: :type x: Input tensor

Returns:

• self.current_xmin (tensor)

• self.current_xmax (tensor)

class aihwkit.simulator.digital_low_precision.range_estimators.MSE_Estimator(num_candidates=100, opt_method=OptMethod.grid, range_margin=0.5, *args, **kwargs)

Bases: RangeEstimatorBase

forward(data)

Accepts an input tensor, updates the current estimates of x_min and x_max and retruns them. :param x: :type x: Input tensor

Returns:

• self.current_xmin (tensor)

• self.current_xmax (tensor)

golden_asym_range_loss(range, data)

Outer Loss function (range) passed to the golden section optimizer from scipy in case of

asymmetric quantization

golden_asym_shift_loss(shift, range, data)

Inner Loss function (shift) passed to the golden section optimizer from scipy in case of asymmetric quantization

golden_sym_loss(range, data)

Loss function passed to the golden section optimizer from scipy in case of symmetric quantization

loss_fx(data, neg_thr, pos_thr, per_channel_loss=False)

property optimization_method

quantize(x_float, x_min=None, x_max=None)

reset()

Reset the range estimator.

property step_size

exception aihwkit.simulator.digital_low_precision.range_estimators.NoDataPassedError

Bases: Exception

Raised data has been passed inot the Range Estimator.

class aihwkit.simulator.digital_low_precision.range_estimators.OptMethod(*values)

Bases: Enum

golden_section = 2

grid = 1

classmethod list()

class aihwkit.simulator.digital_low_precision.range_estimators.RangeEstimatorBase(per_channel=False, quantizer=None, axis=None, n_groups=None, *args, **kwargs)

Bases: Module

forward(x)

Accepts an input tensor, updates the current estimates of x_min and x_max and retruns them. :param x: :type x: Input tensor

Returns:

• self.current_xmin (tensor)

• self.current_xmax (tensor)

reset()

Reset the range estimator.

class aihwkit.simulator.digital_low_precision.range_estimators.RangeEstimatorMap(value, cls)

Bases: tuple

cls

Alias for field number 1

value

Alias for field number 0

class aihwkit.simulator.digital_low_precision.range_estimators.RangeEstimators(*values)

Bases: Enum

MSE = (3, <class 'aihwkit.simulator.digital_low_precision.range_estimators.MSE_Estimator'>)

allminmax = (1, <class 'aihwkit.simulator.digital_low_precision.range_estimators.AllMinMaxEstimator'>)

property cls

cross_entropy = (4, <class 'aihwkit.simulator.digital_low_precision.range_estimators.CrossEntropyEstimator'>)

current_minmax = (0, <class 'aihwkit.simulator.digital_low_precision.range_estimators.CurrentMinMaxEstimator'>)

classmethod list()

running_minmax = (2, <class 'aihwkit.simulator.digital_low_precision.range_estimators.RunningMinMaxEstimator'>)

class aihwkit.simulator.digital_low_precision.range_estimators.RunningMinMaxEstimator(momentum=0.9, *args, **kwargs)

Bases: RangeEstimatorBase

forward(x)

Accepts an input tensor, updates the current estimates of x_min and x_max and retruns them. :param x: :type x: Input tensor

Returns:

• self.current_xmin (tensor)

• self.current_xmax (tensor)