Method forward

ppq/executor/torch.py:366–410

Forward function of this executor. Notice this forward function will never store and compute gradients. Args: inputs (Union[dict, list, torch.Tensor]): [input tensor or somewhat] output_names (List[str], optional): onnx output node names, wh

(
        self,
        inputs: Union[dict, list, torch.Tensor],
        output_names:List[str] = None,
        hooks: Dict[str, RuntimeHook] = None
    )

    @ torch.no_grad()
    def forward(
        self,
        inputs: Union[dict, list, torch.Tensor],
        output_names:List[str] = None,
        hooks: Dict[str, RuntimeHook] = None
    ) -> List[torch.Tensor]:
        """Forward function of this executor.

        Notice this forward function will never store and compute gradients.

        Args:
            inputs (Union[dict, list, torch.Tensor]): [input tensor or somewhat]

            output_names (List[str], optional):
                onnx output node names, which used to confirm a output order.

                Defaults to None.

            hooks (Dict[str, RuntimeHook], optional):
                A hook table for customizing operation behaviour and collate data during executing.
                All hooks should inherit from class RuntimeHook, with all necessary methods implemented.
                    See also: ppq.executor.base.RuntimeHook

                Executor calls hook.pre_forward_hook(operation, input_data) before dispatching operation,
                by using this feature, you can dynamically dispatch operation during executing,
                or processing input data as you want.(remember to return processed input data)

                Executor calls hook.post_forward_hook(operation, output_data) after the execution,
                you are supposed to  gather all necessary data from execution via this feature.

                For Quantable Operation, a much more powerful class:
                    ppq.executor.base.QuantOpRuntimeHook is provided.
                see also: ppq.executor.base.QuantOpRuntimeHook

                Defaults to None.

        Returns:
            List[torch.Tensor]: [executing result, list of tensor objects.]
        """
        return self.__forward(
            inputs=inputs,
            output_names=output_names,
            executing_order=self._executing_order,
            hooks=hooks
        )

    def forward_with_gradient(
        self,