MinMaxObserver#

class MinMaxObserver(mode=QuantMode.SYMMERTIC, eps=0.00001, dtype='qint8', **kwargs)[source]#

A Observer Module records input tensor’s running min and max values to calc scale.

Parameters:

  • mode (QuantMode) – set quantization mode.

  • eps (float) – a initial maximum value to avoid division by zero problem.

  • dtype (Union[str, QuantDtypeMeta]) – a string indicating which dtype to collect scale and zero_point of.