Inverse Autoregressive Flow (IAF) "symbolic Keras input/output" TypeError?

In the AutoregressiveNetwork documentation

https://github.com/tensorflow/probability/blob/fb99b12772513dc6888ea0854a4a891b4bbb4b21/tensorflow_probability/python/bijectors/masked_autoregressive.py#L652-L669

following

# Generate data -- as in Figure 1 in [Papamakarios et al. (2017)][2]).
n = 2000
x2 = np.random.randn(n).astype(dtype=np.float32) * 2.
x1 = np.random.randn(n).astype(dtype=np.float32) + (x2 * x2 / 4.)
data = np.stack([x1, x2], axis=-1)

# Density estimation with MADE.
made = tfb.AutoregressiveNetwork(params=2, hidden_units=[10, 10])

this code works

distribution = tfd.TransformedDistribution(
    distribution=tfd.Sample(tfd.Normal(loc=0., scale=1.), sample_shape=[2]),
    bijector=tfb.MaskedAutoregressiveFlow(made))

# Construct and fit model.
x_ = tfkl.Input(shape=(2,), dtype=tf.float32)
log_prob_ = distribution.log_prob(x_)
model = tfk.Model(x_, log_prob_)

but if tfb.Invert is added as is done in the MaskedAutoregressiveFlow documentation

https://github.com/tensorflow/probability/blob/fb99b12772513dc6888ea0854a4a891b4bbb4b21/tensorflow_probability/python/bijectors/masked_autoregressive.py#L164-L188

then the code fails

distribution = tfd.TransformedDistribution(
    distribution=tfd.Sample(tfd.Normal(loc=0., scale=1.), sample_shape=[2]),
    bijector=tfb.Invert(tfb.MaskedAutoregressiveFlow(made))) # Here is the only change from previously working code

# Construct and fit model.
x_ = tfkl.Input(shape=(2,), dtype=tf.float32)
log_prob_ = distribution.log_prob(x_)
model = tfk.Model(x_, log_prob_)

with this error message

---------------------------------------------------------------------------

TypeError                                 Traceback (most recent call last)

<ipython-input-11-3e5fc03eaed6> in <module>()
      9 # Construct and fit model.
     10 z1_ = tfkl.Input(shape=(2,), dtype=tf.float32)
---> 11 log_prob_ = f_z1.log_prob(z1_)
     12 model = tfk.Model(z1_, log_prob_)
     13 

20 frames

/usr/local/lib/python3.7/dist-packages/keras/engine/keras_tensor.py in __array__(self)
    243   def __array__(self):
    244     raise TypeError(
--> 245         'Cannot convert a symbolic Keras input/output to a numpy array. '
    246         'This error may indicate that you\'re trying to pass a symbolic value '
    247         'to a NumPy call, which is not supported. Or, '

TypeError: Cannot convert a symbolic Keras input/output to a numpy array. This error may indicate that you're trying to pass a symbolic value to a NumPy call, which is not supported. Or, you may be trying to pass Keras symbolic inputs/outputs to a TF API that does not register dispatching, preventing Keras from automatically converting the API call to a lambda layer in the Functional Model.

Does anyone have any experience with this issue or workarounds for implementing an inverse autoregressive flow (IAF) within the keras framework?

Start with the AutoregressiveNetwork and MaskedAutoregressiveFlow documentation examples linked in the issue, then reproduce the two Keras models using a symbolic tfkl.Input. Compare the working flow with the version wrapped in tfb.Invert and trace the TypeError; done means the inverse autoregressive flow can be used in the Keras model without that symbolic-input failure.