Объект `_UserObject` не имеет атрибута` call_and return_conditional_losses`

Question

Я пытаюсь использовать многоступенчатое обучение. Предсказание первой нейронной сети используется в качестве входных данных для второй нейронной сети. Поэтому мне нужно загрузить первую нейронную сеть и прогнозировать вызов при использовании второй нейронной сети. Мне нужен график / функция первой нейронной сети и использую его во второй нейронной сети.

#neural network class
class Linear(Model):
    def __init__(self, n_layers, activate = "tanh", dtype = "float32"):

        super(Linear, self).__init__()
        self.activation = activate
        self.title=title


        self.model = Sequential()
        self.n_layers=n_layers
        #self.model._set_inputs(inputs)
        num_layers = len(n_layers) 
        for l in range(0, num_layers-1):

            name = "layer_" + str(l)
            m = Dense(n_layers[l+1], input_shape=(n_layers[l],), activation = self.activation, name = name, dtype=dtype)
            self.model.add(m)

        name = "layer_" + str(num_layers-1)
        m = Dense(n_layers[-1], input_shape=(n_layers[-2],), name = name, dtype=dtype)
        self.model.add(m)

    @tf.function
    def __call__(self, X):

        Y = self.model(X)
        #Pred=self.model.predict(X)
        return Y

#Saving and loading
  def save(self,name):

       tf.saved_model.save(self.model_NN,'saved_model/')


    def load(self,name):
        restored_saved_model=keras.models.load_model('saved_model/')
        return restored_saved_model
 #I am training models one after another and using pretection of first model as input of second. So I need to load and save.
   def train(self, epoch,multistepping):

        for l in range(0, self.number_of_NN):
            print("making neural network object",l)
            self.NN_list.append(self.net)

        for l in range(0,self.number_of_NN):
            model = self.pde(self.NN_list[l],self.D,self.dt,self.q, self.cond_i,self.cond_b, self.lr, self.lr_schedule, self.dtype)
            print(" training model number ",l) 
            model.train(epoch[l])
            model.save(l)
            model.title=l
            #model_old=copy.copy(model)

            #Bring it outside

            def function(xy):
                #load weights from previous model
                if l>0:
                    temp_model=model.load(l-1)
                    x_1 = xy[:, 0][:, None]
                    y_1 = xy[:, 1][:, None]
                    U = temp_model.predict(x_1, y_1)
                    U=U[:,-1]
                return np.asarray(U)
            if multistepping==1:
                self.cond_i.u_func = function

#The error message I am prompted
~/Desktop/V3/v3/v2/Ishrak/pde_d_Poisson_2D_v3.py in load(self, name)
    131 
    132     def load(self,name):
--> 133         restored_saved_model=keras.models.load_model('saved_model/')
    134         return restored_saved_model
    135         #Have to check

~/anaconda3/envs/tf2.1/lib/python3.7/site-packages/tensorflow_core/python/keras/saving/save.py in load_model(filepath, custom_objects, compile)
    148   if isinstance(filepath, six.string_types):
    149     loader_impl.parse_saved_model(filepath)
--> 150     return saved_model_load.load(filepath, compile)
    151 
    152   raise IOError(

~/anaconda3/envs/tf2.1/lib/python3.7/site-packages/tensorflow_core/python/keras/saving/saved_model/load.py in load(path, compile)
     87   # TODO(kathywu): Add saving/loading of optimizer, compiled losses and metrics.
     88   # TODO(kathywu): Add code to load from objects that contain all endpoints
---> 89   model = tf_load.load_internal(path, loader_cls=KerasObjectLoader)
     90 
     91   # pylint: disable=protected-access

~/anaconda3/envs/tf2.1/lib/python3.7/site-packages/tensorflow_core/python/saved_model/load.py in load_internal(export_dir, tags, loader_cls)
    550       loader = loader_cls(object_graph_proto,
    551                           saved_model_proto,
--> 552                           export_dir)
    553       root = loader.get(0)
    554     root.tensorflow_version = meta_graph_def.meta_info_def.tensorflow_version

~/anaconda3/envs/tf2.1/lib/python3.7/site-packages/tensorflow_core/python/keras/saving/saved_model/load.py in __init__(self, *args, **kwargs)
    117   def __init__(self, *args, **kwargs):
    118     super(KerasObjectLoader, self).__init__(*args, **kwargs)
--> 119     self._finalize()
    120 
    121   def _finalize(self):

~/anaconda3/envs/tf2.1/lib/python3.7/site-packages/tensorflow_core/python/keras/saving/saved_model/load.py in _finalize(self)
    137     for node in self._nodes:
    138       if isinstance(node, RevivedNetwork):
--> 139         call_fn = node.keras_api.call_and_return_conditional_losses
    140         if call_fn.input_signature is None:
    141           inputs = infer_inputs_from_restored_call_function(call_fn)

AttributeError: '_UserObject' object has no attribute 'call_and_return_conditional_losses'

Как сохранить и загрузить модель TensorFlow в этом сценарии?