utils/scripts/gen_onnx_gru_model.py [296:351]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    )

    # Error node definition
    err_node_def = onnx.helper.make_node(
        "Sub", name="error", inputs=["Y", "Y_ref"], outputs=["Y_err"]
    )

    # --------------------------------------------- GRAPH DEFINITION  --------------------------------------------------
    graph_input = list()
    graph_init = list()
    graph_output = list()

    # GRU inputs
    graph_input.append(helper.make_tensor_value_info("X", TensorProto.FLOAT, X_shape))
    graph_input.append(helper.make_tensor_value_info("W", TensorProto.FLOAT, W_shape))
    graph_input.append(helper.make_tensor_value_info("R", TensorProto.FLOAT, R_shape))
    if has_bias:
        graph_input.append(
            helper.make_tensor_value_info("B", TensorProto.FLOAT, B_shape)
        )
    if has_sequence_lens:
        graph_input.append(
            helper.make_tensor_value_info(
                "sequence_lens", TensorProto.INT32, sequence_lens_shape
            )
        )
    if has_initial_h:
        graph_input.append(
            helper.make_tensor_value_info(
                "initial_h", TensorProto.FLOAT, initial_h_shape
            )
        )

    # Reference input
    graph_input.append(
        helper.make_tensor_value_info("Y_ref", TensorProto.FLOAT, Y_shape)
    )

    # GRU initializers
    graph_init.append(make_init("X", TensorProto.FLOAT, X))
    graph_init.append(make_init("W", TensorProto.FLOAT, W))
    graph_init.append(make_init("R", TensorProto.FLOAT, R))
    if has_bias:
        graph_init.append(make_init("B", TensorProto.FLOAT, B))
    if has_sequence_lens:
        graph_init.append(make_init("sequence_lens", TensorProto.INT32, sequence_lens))
    if has_initial_h:
        graph_init.append(make_init("initial_h", TensorProto.FLOAT, initial_h))

    # Reference initializer
    graph_init.append(make_init("Y_ref", TensorProto.FLOAT, Y_ref))

    # Graph outputs
    graph_output.append(
        helper.make_tensor_value_info("Y_err", TensorProto.FLOAT, Y_shape)
    )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



utils/scripts/gen_onnx_rnn_model.py [266:321]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    )

    # Error node definition
    err_node_def = onnx.helper.make_node(
        "Sub", name="error", inputs=["Y", "Y_ref"], outputs=["Y_err"]
    )

    # --------------------------------------------- GRAPH DEFINITION  --------------------------------------------------
    graph_input = list()
    graph_init = list()
    graph_output = list()

    # RNN inputs
    graph_input.append(helper.make_tensor_value_info("X", TensorProto.FLOAT, X_shape))
    graph_input.append(helper.make_tensor_value_info("W", TensorProto.FLOAT, W_shape))
    graph_input.append(helper.make_tensor_value_info("R", TensorProto.FLOAT, R_shape))
    if has_bias:
        graph_input.append(
            helper.make_tensor_value_info("B", TensorProto.FLOAT, B_shape)
        )
    if has_sequence_lens:
        graph_input.append(
            helper.make_tensor_value_info(
                "sequence_lens", TensorProto.INT32, sequence_lens_shape
            )
        )
    if has_initial_h:
        graph_input.append(
            helper.make_tensor_value_info(
                "initial_h", TensorProto.FLOAT, initial_h_shape
            )
        )

    # Reference input
    graph_input.append(
        helper.make_tensor_value_info("Y_ref", TensorProto.FLOAT, Y_shape)
    )

    # RNN initializers
    graph_init.append(make_init("X", TensorProto.FLOAT, X))
    graph_init.append(make_init("W", TensorProto.FLOAT, W))
    graph_init.append(make_init("R", TensorProto.FLOAT, R))
    if has_bias:
        graph_init.append(make_init("B", TensorProto.FLOAT, B))
    if has_sequence_lens:
        graph_init.append(make_init("sequence_lens", TensorProto.INT32, sequence_lens))
    if has_initial_h:
        graph_init.append(make_init("initial_h", TensorProto.FLOAT, initial_h))

    # Reference initializer
    graph_init.append(make_init("Y_ref", TensorProto.FLOAT, Y_ref))

    # Graph outputs
    graph_output.append(
        helper.make_tensor_value_info("Y_err", TensorProto.FLOAT, Y_shape)
    )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -