NotImplementedError: Layer ArcFace was created by passing non-serializable argument values in __init__()

[HoaiAn0906]

NotImplementedError:
Layer ArcFace was created by passing
non-serializable argument values in __init__(),
and therefore the layer must override get_config() in
order to be serializable. Please implement get_config().

Example:

class CustomLayer(keras.layers.Layer):
def init(self, arg1, arg2, **kwargs):
super().init(**kwargs)
self.arg1 = arg1
self.arg2 = arg2

def get_config(self):
    config = super().get_config()
    config.update({
        "arg1": self.arg1,
        "arg2": self.arg2,
    })
    return config

This is my code:
batch_size=32
train_datagen_names = keras.preprocessing.image.ImageDataGenerator(preprocessing_function=preprocess_input)
train_generator_names = train_datagen_names.flow_from_directory('/content/drive/MyDrive/train_dir',target_size=(224,224),color_mode='rgb',batch_size=32,class_mode='categorical',shuffle=True)
NO_CLASSES = len(train_generator_names.class_indices.values())

generate data

class train_Generator_xandy(object):
def init(self):
datagen = ImageDataGenerator(preprocessing_function=preprocess_input)

    train_generator = datagen.flow_from_directory(
        '/content/drive/MyDrive/train_dir',
        target_size=(224, 224),
        color_mode='rgb',
        batch_size=32,
        class_mode='categorical',
        shuffle=True
    )

    self.gene = train_generator

def __iter__(self):
    return self

def __next__(self):
    X, Y = self.gene.next()
    return [X, Y], Y

class val_Generator_xandy(object):
def init(self):
datagen = ImageDataGenerator(preprocessing_function=preprocess_input)

    validation_generator = datagen.flow_from_directory(
        '/content/drive/MyDrive/validation_dir',
        target_size=(224, 224),
        color_mode='rgb',
        batch_size=32,
        class_mode='categorical',
        shuffle=True
    )

    self.gene = validation_generator

def __iter__(self):
    return self

def __next__(self):
    X, Y = self.gene.next()
    return [X, Y], Y

base_model = VGG16(include_top=False, weights='imagenet', input_shape=(224, 224, 3))
base_model.summary()

x = base_model.output
y = Input(shape=(NO_CLASSES, ))
x = BatchNormalization()(x)
x = Dropout(0.5)(x)
x = Flatten()(x)
x = Dense(128, kernel_initializer='he_normal',kernel_regularizer=keras.regularizers.l2(1e-4))(x)
#x = Dense(128, activation='relu')(x)
#predictions = Dense(num_classes, activation='softmax')(x)
predictions = ArcFace(NO_CLASSES,regularizer=keras.regularizers.l2(1e-4))([x, y])

model = Model(inputs=[base_model.input, y], outputs=predictions)

for layer in base_model.layers:
layer.trainable = False

model.summary()

opt = keras.optimizers.legacy.SGD(learning_rate=0.1, momentum=0.9, decay=5e-4, nesterov=True)

model.compile(loss='categorical_crossentropy',optimizer=opt,metrics=['accuracy'])

train_generator = train_Generator_xandy()
validation_generator = val_Generator_xandy()

model.fit(train_generator,batch_size = 1,verbose = 1,epochs = 20)

model.fit(
train_generator,
steps_per_epoch=batch_size,
epochs=10,
validation_data=validation_generator,
validation_steps=batch_size
)

creates a HDF5 file

model.save('transfer_learning_trained' + '_face_cnn_model.h5')