Problem on creating and saving tflite model

TensorFlow General Discussion
,
1

Hello

I am trying to make a tflite model using a tf.module with modified model inside of it for inference, however I encounter an error that the model doesn’t have call argument. but when I saved the original model as tflite, it works just fine.

here the notebook in kaggle looks like notebook

any help would be appreciated thank you

3

Hi @Gibran_Alfil, Could you please try with the below given code and let us know if the issue gets resolved or not.

def saves () :
    # Create Model Converter
    keras_model_converter = tf.lite.TFLiteConverter.from_keras_model(tflite_keras_model)
   
    #Added this code line
    keras_model_converter.converter.experimental_new_converter = True


# Convert Model
    tflite_model = keras_model_converter.convert()  
    with open('/kaggle/working/model2.tflite', 'wb') as f:
        f.write(tflite_model)

Thank You.

4

Update :
i fixed it by naming all the layer inside each part (decoder, embedding, etc). How I found it is by trying to convert each part into tflite. the only working part is the classifier. And its not just conversion either, saving it normally using saved_model.save throw bad argument type for built in operation. Should i report this as a bug?