keras_saved_model fails becase model is not json serializable

tf 1.13
tfp 0.6
ubuntu 18.04

Trying to save a Bayesian neural net in a way that can be run in production using either the java or c++ api. I've tried

save_keras_model(model, saved_model_path='model_path', serving_only=True, custom_objects={'DenseFlipout': tfp.layers.DenseFlipout, 'Conv2DFlipout': tfp.layers.Convolution2DFlipout})

and just:
save_keras_model(model, saved_model_path='model_path', serving_only=True)

Both fail because model.to_json() fails

This minimal example produces the below error:

`import tensorflow as tf
import tensorflow_probability as tfp
from tensorflow.contrib.saved_model import save_keras_model

tf.enable_eager_execution()

neural_net = tf.keras.Sequential([
tfp.layers.Convolution2DFlipout(6,
input_shape=(3, 255, 255),
kernel_size=5,
padding="SAME",
activation=tf.nn.relu),
tf.keras.layers.MaxPooling2D(pool_size=[2, 2],
strides=[2, 2],
padding="SAME"),
tfp.layers.Convolution2DFlipout(16,
kernel_size=5,
padding="SAME",
activation=tf.nn.relu),
tf.keras.layers.MaxPooling2D(pool_size=[2, 2],
strides=[2, 2],
padding="SAME"),
tfp.layers.Convolution2DFlipout(120,
kernel_size=5,
padding="SAME",
activation=tf.nn.relu),
tf.keras.layers.Flatten(),
tfp.layers.DenseFlipout(84, activation=tf.nn.relu),
tfp.layers.DenseFlipout(10)
])

save_keras_model(neural_net, saved_model_path='model_path', serving_only=True)
`

File "/home/tony/.conda/envs/tf/lib/python3.6/site-packages/tensorflow/contrib/saved_model/python/saved_model/keras_saved_model.py", line 141, in save_keras_model
_export_model_json(model, export_dir)
File "/home/tony/.conda/envs/tf/lib/python3.6/site-packages/tensorflow/contrib/saved_model/python/saved_model/keras_saved_model.py", line 151, in _export_model_json
model_json = model.to_json()
File "/home/tony/.conda/envs/tf/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py", line 1563, in to_json
return json.dumps(model_config, default=get_json_type, **kwargs)
File "/home/tony/.conda/envs/tf/lib/python3.6/json/init.py", line 238, in dumps
**kw).encode(obj)
File "/home/tony/.conda/envs/tf/lib/python3.6/json/encoder.py", line 199, in encode
chunks = self.iterencode(o, _one_shot=True)
File "/home/tony/.conda/envs/tf/lib/python3.6/json/encoder.py", line 257, in iterencode
return _iterencode(o, 0)
File "/home/tony/.conda/envs/tf/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py", line 1560, in get_json_type
raise TypeError('Not JSON Serializable:', obj)
TypeError: ('Not JSON Serializable:', <function default_loc_scale_fn.._fn at 0x7f8a640f2378>)

Are there any tested ways of taking tensorflow_probability to production other than running pure python?