Calling `Model.predict` in graph mode is not supported when the `Model` instance was constructed with eager mode enabled

The answer is simple, just load your model inside the graph just like this:

with graph.as_default():        json_file = open('models/model.json','r')        loaded_model_json = json_file.read()        json_file.close()        loaded_model = model_from_json(loaded_model_json)        #load weights into new model        loaded_model.load_weights("models/model.h5")        print("Loaded Model from disk")        #compile and evaluate loaded model        loaded_model.compile(loss='sparse_categorical_crossentropy',optimizer='adam',metrics=['accuracy'])        # perform the prediction        out = loaded_model.predict(img)        print(out)        print(class_names[np.argmax(out)])        # convert the response to a string        response = class_names[np.argmax(out)]        return str(response)

@Ilham: Try to wrap the call method in a tf.function, right after defining your network. Something like this:

model = Sequential()model.call = tf.function(model.call)

I had an issue similar to yours. I solved it just by adding that second line of code.See the following link for more details: https://www.tensorflow.org/guide/intro_to_graphs