ValueError: Unknown metric function: top_2_accuracy. Please ensure this object is passed to the `custom_objects` argument

Question

I'm working on a CNN classification project, and I've used the top 2 Accuracy (top_k_categorical_accuracy) as the accuracy matrix for training. the function in the model notebook is:

from tensorflow.keras.metrics import top_k_categorical_accuracy

def top_2_accuracy(y_true, y_pred):
    return top_k_categorical_accuracy(y_true, y_pred, k=2)

then I used it as

model.compile(optimizer = Adam(lr = 4e-3), 
              loss='categorical_crossentropy', 
              metrics=['accuracy', top_2_accuracy])

Now I need to load the model to use it in an application I tried to use this code:

model = tf.keras.models.load_model('model.h5')

but an error occurred while loading

Unknown metric function: top_2_accuracy. Please ensure this object is passed to the `custom_objects` argument. See https://www.tensorflow.org/guide/keras/save_and_serialize#registering_the_custom_object for details.

then I've googled and tried this too:

model = tf.keras.models.load_model('model.h5', custom_objects={'top_k_categorical_accuracy(y_true, y_pred, k=2)':top_2_accuracy})

but it generated another error

model = tf.keras.models.load_model('model.h5', custom_objects={'top_k_categorical_accuracy(y_true, y_pred, k=2)':top_2_accuracy})
NameError: name 'top_2_accuracy' is not defined

How can I solve this??