def inputs()

in optimum/exporters/onnx/model_configs.py [0:0]

• 18 lines of code
• 6 McCabe index (conditional complexity)

    def inputs(self) -> Dict[str, Dict[int, str]]:
        common_inputs = {}

        # Batched inference is not supported in Transformers.
        if self._behavior is ConfigBehavior.ENCODER:
            common_inputs["input_ids"] = {1: "encoder_sequence_length"}
        elif self._behavior is ConfigBehavior.DECODER:
            # NOTE: even when past is used, the decoder takes the full sequence as input as the prenet seem to require it:
            # https://github.com/huggingface/transformers/blob/v4.33.2/src/transformers/models/speecht5/modeling_speecht5.py#L2573
            common_inputs["output_sequence"] = {1: "decoder_sequence_length"}
            common_inputs["speaker_embeddings"] = {}  # No dynamic shape here.
            common_inputs["encoder_outputs"] = {1: "encoder_sequence_length"}
            common_inputs["encoder_attention_mask"] = {1: "encoder_sequence_length"}

            if self.variant == "with-past" and self.use_past_in_inputs:
                self.add_past_key_values(common_inputs, direction="inputs")
        elif self.is_postnet_and_vocoder:
            common_inputs["spectrogram"] = {0: "n_spectrums x reduction_factor"}
        else:
            raise ValueError(
                "self._behavior is neither encoder or decoder, and is_postnet_and_vocoder=False. This should not happen."
            )

        return common_inputs