when using the saved_model format for storing pretrained model, model loading causes many unnecessary warnings.

[aynesss]

I use TF2.4.1
I would like to finetune the savedModel of pretrained .I write this :

target_model_path = "gs://simclr-checkpoints-tf2/simclrv1/pretrain/1x/saved_model/"
def create_model():
    baseModel = tf.saved_model.load(target_model_path)
    baseModel= tf.saved_model.save('./saved_model.h5')

    headModel = baseModel.output
    model_output = tf.keras.layers.Dense(3, activation="softmax", name="output")(headModel)  

    # compile the model
    model = tf.keras.Model(inputs=baseModel.input, outputs=model_output)
    for layer in model.layers:
        layer.trainable = True 

    model.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=0.001),
              loss=tf.keras.losses.CategoricalCrossentropy(),  # from_logits=True
              metrics=[tf.keras.metrics.CategoricalAccuracy()])  # TopKCategoricalAccuracy(k=1)
    return model

I only add a classification layer. But model loading causes many unnecessary warnings like this :

WARNING:absl:Importing a function (__inference___call___15879) with ops with custom gradients. Will likely fail if a gradient is requested.
WARNING:absl:Importing a function (__inference___call___15879) with ops with custom gradients. Will likely fail if a gradient is requested.

Any ideas what could be the possible reason behind this? This https://colab.research.google.com/github/google-research/simclr/blob/master/tf2/colabs/finetuning.ipynb#scrollTo=6WXspghpERRG also have same problem.