bayesian_cnn_real_data example return same values for every forward pass

[ahiguera-mx]

Hi,

First of all, thanks a lot for putting together these nice examples, I've tried to run bayesian_cnn_real_data, and every forward pass returns the same value, specifically when calling the function

def analyse_model_prediction() the values for
extracted_probabilities[i] = model_output_distribution.mean().numpy().flatten()
They are always the same. Any thoughts?
I'm using tf 2.14.0 and tfp 0.22.0