fine- tuning universal sentence encoder large 5

[starcsofnorth]

i am unable to fine it
module_url = "https://tfhub.dev/google/universal-sentence-encoder-large/5"
sent1_input = tf.keras.layers.Input(shape=(32,128), name='sent1',dtype=tf.string)
use_model = hub.KerasLayer(module_url, trainable= True)
output = use_model([sent1_input])

error:-
AssertionError: in converted code:

/usr/local/lib/python3.6/dist-packages/tensorflow_hub/keras_layer.py:209 call  *
    result = f()
/tensorflow-2.1.0/python3.6/tensorflow_core/python/saved_model/load.py:438 _call_attribute
    return instance.__call__(*args, **kwargs)
/tensorflow-2.1.0/python3.6/tensorflow_core/python/eager/def_function.py:568 __call__
    result = self._call(*args, **kwds)
/tensorflow-2.1.0/python3.6/tensorflow_core/python/eager/def_function.py:606 _call
    results = self._stateful_fn(*args, **kwds)
/tensorflow-2.1.0/python3.6/tensorflow_core/python/eager/function.py:2363 __call__
    return graph_function._filtered_call(args, kwargs)  # pylint: disable=protected-access
/tensorflow-2.1.0/python3.6/tensorflow_core/python/eager/function.py:1611 _filtered_call
    self.captured_inputs)
/tensorflow-2.1.0/python3.6/tensorflow_core/python/eager/function.py:1705 _call_flat
    "StatefulPartitionedCall": self._get_gradient_function()}):
/usr/lib/python3.6/contextlib.py:81 __enter__
    return next(self.gen)
/tensorflow-2.1.0/python3.6/tensorflow_core/python/framework/ops.py:4727 _override_gradient_function
    assert not self._gradient_function_map

[gowthamkpr]

Was able to reproduce the error. Heres the gist.Thanks!

[andresusanopinto]

The error message is indded not useful. However this works if you pass the right expected input shape (shape=[]) and then call without creating an extra array (use_module(sent1_input)).

module_url = "https://tfhub.dev/google/universal-sentence-encoder-large/5"
sent1_input = tf.keras.layers.Input(shape=[], name='sent1',dtype=tf.string)
use_model = hub.KerasLayer(module_url, trainable= True)
output = use_model(sent1_input)

Note: While debugging this I found that if first you run the bad code, then using the right code on the same process fails, if however you run the correct code right away, it seems to work fine.