Sure, I am completely available to discuss the detail.
I am not using the NUTS sampler, which requires gradients, but just a plain Metropolis sampler. Therefore, not having the gradients should not be a problem.
The error is raised when doing model.initial_point() inside the initialization of the Metropolis Sampler. I don’t know what might be causing it, but it seems that the size argument is not correctly passed to some function, causing some shape-inconsistency problems.
What would you like to check in more detail? Do you want me to provide you with a minimal working example where it fails?
Thanks!