coreai_opt.quantization.spec.RangeCalculatorBase

class coreai_opt.quantization.spec.RangeCalculatorBase(granularity, **kwargs)

Bases: ClassRegistryMixin, Module

Base class and registry for classes used to compute the range of a given tensor.

Parameters:

granularity (QuantizationGranularity)

__init__(granularity, **kwargs)

Parameters:

granularity (QuantizationGranularity)

Methods

forward(x)	Compute range statistics on an input and return the min/max bounds.
get_class(key)
list_registry_keys()
list_registry_values()
register(key)	Register a virtual subclass of an ABC.

forward(x)

Compute range statistics on an input and return the min/max bounds.

Calls _generate_min_max to compute range statistics and validates that the returned min/max shapes match the original tensor number of dimensions.

Parameters:

x (torch.Tensor) – Tensor to compute range statistics upon.

Return type:

tuple[Tensor, Tensor]