View source on GitHub
|
Module containing Quantization abstraction and quantizers.
Classes
class AllValuesQuantizer: Quantize tensor based on min/max of tensor values across all batches.
class FixedQuantizer: Quantize tensor based on min/max of tensor values with the fixed range.
class LastValueQuantizer: Quantize tensor based on range the last batch of values.
class MovingAverageQuantizer: Quantize tensor based on a moving average of values across batches.
class Quantizer: ABC interface which encapsulates the logic of how to quantize tensors.