如何在Tensorflow中保存不可序列化的模型

import tensorflow as tf

class CustomModel(tf.keras.Model):
    # 模型定义
    def __init__(self):
        super(CustomModel, self).__init__()
        # 模型的层定义

    def call(self, inputs):
        # 模型的前向计算逻辑

    def get_config(self):
        # 模型的配置信息

    def from_config(cls, config):
        # 根据配置信息创建模型实例

    def serialize(self):
        # 自定义序列化函数

    @staticmethod
    def deserialize(serialized):
        # 自定义反序列化函数

model = CustomModel()
# 训练模型
model.save('path/to/save/model', save_format='tf')

model = tf.keras.models.load_model('path/to/save/model', custom_objects={'CustomModel': CustomModel})

import tensorflow as tf

class CustomModel(tf.keras.Model):
    # 模型定义
    def __init__(self):
        super(CustomModel, self).__init__()
        # 模型的层定义

    def call(self, inputs):
        # 模型的前向计算逻辑

model = CustomModel()
# 模型编译和训练
checkpoint_callback = tf.keras.callbacks.ModelCheckpoint('path/to/save/checkpoint', save_weights_only=True)
model.fit(x_train, y_train, epochs=10, callbacks=[checkpoint_callback])

model = CustomModel()
model.load_weights('path/to/save/checkpoint')