optimum/onnxruntime/modeling_ort.py [1403:1425]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        }

        if self.use_io_binding:
            output_shapes, output_buffers = self._prepare_io_binding(model_inputs)

            # run inference with binding & synchronize in case of multiple CUDA streams
            if self.device.type == "cpu":
                self.session.run_with_iobinding(self._io_binding)
            else:
                self._io_binding.synchronize_inputs()
                self.session.run_with_iobinding(self._io_binding)
                self._io_binding.synchronize_outputs()

            logits = output_buffers["logits"].view(output_shapes["logits"])
        else:
            onnx_inputs = self._prepare_onnx_inputs(use_torch, model_inputs)
            onnx_outputs = self.model.run(None, onnx_inputs)
            model_outputs = self._prepare_onnx_outputs(use_torch, onnx_outputs)

            logits = model_outputs["logits"]

        if not return_dict:
            return (logits,)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



optimum/onnxruntime/modeling_ort.py [1797:1819]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        }

        if self.use_io_binding:
            output_shapes, output_buffers = self._prepare_io_binding(model_inputs)

            # run inference with binding & synchronize in case of multiple CUDA streams
            if self.device.type == "cpu":
                self.session.run_with_iobinding(self._io_binding)
            else:
                self._io_binding.synchronize_inputs()
                self.session.run_with_iobinding(self._io_binding)
                self._io_binding.synchronize_outputs()

            logits = output_buffers["logits"].view(output_shapes["logits"])
        else:
            onnx_inputs = self._prepare_onnx_inputs(use_torch, model_inputs)
            onnx_outputs = self.model.run(None, onnx_inputs)
            model_outputs = self._prepare_onnx_outputs(use_torch, onnx_outputs)

            logits = model_outputs["logits"]

        if not return_dict:
            return (logits,)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -