How to save and load trained trainable_variables in tensorflow probability distribution?

I want to build a flow model which I can save the trained model.

flow = tfd.TransformedDistribution(
    ...
    distribution=tfd.Normal(loc=0.0, scale=1.0),
    bijector=my_bijectors
)

for e in epochs:
    ...
    with tf.GradientTape() as tape:
         log_prob_loss = loss() # some loss func here
    grads = tape.gradient(log_prob_loss, flow.trainable_variables)
    optimizer.apply_gradients(zip(grads, flow.trainable_variables))
    ...

How can I save and load the flow here?

I was thinking about using pickle to save the trainable_variables, but I don't know how to apply the loaded trained trainable_variables to the orignal one. Is there a way to update it?

Start by reviewing the flow construction and the trainable_variables update shown in the issue, including the proposed pickle approach. Determine the supported save and load path for the trained flow and how restored values are applied; done means the guidance explains both operations clearly.