我們從 TF1 部署了很多模型，通過圖形凍結來保存它們：tf.train.write_graph(self.session.graph_def, some_path)# get graph definitions with weightsoutput_graph_def = tf.graph_util.convert_variables_to_constants(        self.session,  # The session is used to retrieve the weights        self.session.graph.as_graph_def(),  # The graph_def is used to retrieve the nodes        output_nodes,  # The output node names are used to select the usefull nodes)# optimize graphif optimize:    output_graph_def = optimize_for_inference_lib.optimize_for_inference(            output_graph_def, input_nodes, output_nodes, tf.float32.as_datatype_enum    )with open(path, "wb") as f:    f.write(output_graph_def.SerializeToString())然后通過以下方式加載它們：with tf.Graph().as_default() as graph:    with graph.device("/" + args[name].processing_unit):        tf.import_graph_def(graph_def, name="")            for key, value in inputs.items():                self.input[key] = graph.get_tensor_by_name(value + ":0")我們想以類似的方式保存 TF2 模型。一個包含圖形和權重的 protobuf 文件。我怎樣才能做到這一點？我知道有一些保存方法：keras.experimental.export_saved_model(model, 'path_to_saved_model')這是實驗性的并創建多個文件:(。model.save('path_to_my_model.h5')保存 h5 格式:(。tf.saved_model.save(self.model, "test_x_model")這又保存了多個文件:(。