Tensorflow 2.0.0 / tf.keras.layers.TimeDistributed layer can't be save to saved Model

[luvwinnie]

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution : Colaboratory (GPU Runtime)
• tensorflow version: 2.0.0
• python version: 3.6.8

Describe the current behavior
The model defined which has tf.keras.layers.timeDistributed layer cannot be save by model.save() function.

It shows the error below

ValueError                                Traceback (most recent call last)
/usr/lib/python3.6/inspect.py in getfullargspec(func)
   1125                                        skip_bound_arg=False,
-> 1126                                        sigcls=Signature)
   1127     except Exception as ex:

41 frames
ValueError: no signature found for builtin <tensorflow.python.keras.saving.saved_model.save_impl.LayerCall object at 0x7f74467284a8>

The above exception was the direct cause of the following exception:

TypeError                                 Traceback (most recent call last)
/usr/lib/python3.6/inspect.py in getfullargspec(func)
   1130         # else. So to be fully backwards compatible, we catch all
   1131         # possible exceptions here, and reraise a TypeError.
-> 1132         raise TypeError('unsupported callable') from ex
   1133 
   1134     args = []

TypeError: unsupported callable

Describe the expected behavior
In tensorflow 2.0.0 supposed the model.save model default to be saved in SavedModel format.

Code to reproduce the issue

def get_data():
  datasets, ds_info = tfds.load(name='mnist', with_info=True, as_supervised=True)
  mnist_train, mnist_test = datasets['train'], datasets['test']

  BUFFER_SIZE = 10000

  BATCH_SIZE_PER_REPLICA = 64
  BATCH_SIZE = BATCH_SIZE_PER_REPLICA * mirrored_strategy.num_replicas_in_sync

  def scale(image, label):
    image = tf.cast(image, tf.float32)
    image /= 255

    return image, label

  train_dataset = mnist_train.map(scale).cache().shuffle(BUFFER_SIZE).batch(BATCH_SIZE)
  eval_dataset = mnist_test.map(scale).batch(BATCH_SIZE)

  return train_dataset, eval_dataset

def get_model():
  with mirrored_strategy.scope():
    model = tf.keras.Sequential([
        tf.keras.layers.Conv2D(32, 3, activation='relu', input_shape=(28, 28, 1)),
        tf.keras.layers.MaxPooling2D(),
        # tf.keras.layers.Flatten(),
        tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(64, activation='softmax')),
        # tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(64, activation='softmax')),
        # tf.keras.layers.Dense(10, activation='softmax')
    ])

    model.compile(loss='sparse_categorical_crossentropy',
                  optimizer=tf.keras.optimizers.Adam(),
                  metrics=['accuracy'])
    return model

model = get_model()
model.save("test_save")

Other info / logs
The not only can be reproduce in Colaboratory, but also in normal Ubuntu machien which installed with tensorflow-gpu 2.0.0

[rmothukuru]

Could reproduce this issue with TF Version 2.0. Here is the Gist.

[aldenks]

Also running into this. Here's a concise test case:

model = tf.keras.Sequential([tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(1), input_shape=(None, 1))])
model.save('test_save')

[luvwinnie]

does this issues got any progress?

[SavanK]

Any update?

[k-w-w]

This appears to be an error with Python 3. Submitting a fix to resolve this.

[tensorflow-bot]

Are you satisfied with the resolution of your issue?
Yes
No

[luvwinnie]

so is this been fixed to tensorflow-gpu 2.0 stable build? or i have to install another build?

[jvishnuvardhan]

@luvwinnie It was fixed in tf-nightly which is development version of TF2.1 (will be released in
the near future). Thanks!