FakeQuantize

class torch.ao.quantization.fake_quantize.FakeQuantize(observer=<class 'torch.ao.quantization.observer.MovingAverageMinMaxObserver'>, quant_min=None, quant_max=None, **observer_kwargs) [source]

Simulate the quantize and dequantize operations in training time. The output of this module is given by:

x_out = (
  clamp(round(x/scale + zero_point), quant_min, quant_max) - zero_point
) * scale

• scale defines the scale factor used for quantization.
• zero_point specifies the quantized value to which 0 in floating point maps to
• fake_quant_enabled controls the application of fake quantization on tensors, note that statistics can still be updated.
• observer_enabled controls statistics collection on tensors

• dtype specifies the quantized dtype that is being emulated with fake-quantization,

  allowable values are torch.qint8 and torch.quint8.

Parameters

• observer (module) – Module for observing statistics on input tensors and calculating scale and zero-point.
• observer_kwargs (optional) – Arguments for the observer module

Variables

activation_post_process (Module) – User provided module that collects statistics on the input tensor and provides a method to calculate scale and zero-point.