Thank you very much for your advice. I still have some questions. It looks like I need to use PyTensor Op, but I’m not sure if I have to define my own log-likelihood function. Can I just input the forward function and observations without defining log-likelihood function?