Method partial_graph_forward

ppq/executor/torch.py:654–682

This forward function allows you to execute a series operations in your graph. (only operations list in your params will be executed with this function) Which serves as a great feature for quantization aware training. Args: operations (List[Operation]):

(
        self, operations: List[Operation],
        feed_dict: Dict[str, torch.Tensor],
        output_names:List[str])

        self.forward(inputs=feed_dict, hooks=hooks)

    def partial_graph_forward(
        self, operations: List[Operation],
        feed_dict: Dict[str, torch.Tensor],
        output_names:List[str]) -> List[torch.Tensor]:
        """This forward function allows you to execute a series operations in
        your graph. (only operations list in your params will be executed with
        this function) Which serves as a great feature for quantization aware
        training.

        Args:
            operations (List[Operation]):
                operations that you want to execute,
                notice that executing will strictly follow your operation order.

            feed_dict (Dict[str, torch.Tensor]):
                an dictionary contains {variable name: value}, as an input to this execution.

            output_names (List[str]):
                output variable names.

        Returns:
            List[torch.Tensor]: [description]
        """

        return self.__forward(
            inputs=feed_dict,
            output_names=output_names,
            executing_order=operations
        )