ObservationType¶
- class torch.ao.quantization.backend_config.ObservationType(value)[source][source]¶
An enum that represents different ways of how an operator/operator pattern should be observed
- INPUT_OUTPUT_NOT_OBSERVED = 2¶
this means the input and output are never observed example: x.shape, x.size
- OUTPUT_SHARE_OBSERVER_WITH_INPUT = 1¶
this means the output will use the same observer instance as input, based on qconfig.activation example: torch.cat, maxpool
- OUTPUT_USE_DIFFERENT_OBSERVER_AS_INPUT = 0¶
this means input and output are observed with different observers, based on qconfig.activation example: conv, linear, softmax