qconfig
horizon_plugin_pytorch.quantization.get_qconfig (observer:
~typing.Type[~horizon_plugin_pytorch.quantization.observer_v2.ObserverBase] =
<class 'horizon_plugin_pytorch.quantization.observer_v2.MinMaxObserver'>,
in_dtype: ~torch.dtype | ~horizon_plugin_pytorch.dtype.QuantDType | None = None,
weight_dtype: ~torch.dtype | ~horizon_plugin_pytorch.dtype.QuantDType | None = 'qint8',
out_dtype: ~torch.dtype | ~horizon_plugin_pytorch.dtype.QuantDType | None = 'qint8',
fix_scale: bool = False)
Get qconfig.
Parameters:
observer (Type[ObserverBase]) – observer type for input and output. Support MinMaxObserver and MSEObserver.
in_dtype (Union[dtype, QuantDType, None]) – input dtype.
weight_dtype (Union[dtype, QuantDType, None]) – weight dtype.
out_dtype (Union[dtype, QuantDType, None]) – output dtype.
fix_scale (bool) – Whether fix input/output scale.
