Thanks for the helpful response! You’re right that for my current 3-parameter model, the as_op approach works fine with Metropolis sampling. I appreciate the variable rescaling tip as well.
The current code is a very primary version. Since I do plan to expand this to more parameters eventually, I’m interested in exploring gradient-based options. I tried sunode following LLM suggestions but ran into issues getting it to work properly.
Regarding diffrax - the link you shared appears to be invalid on my end. Could you point me to other tutorials or examples for using diffrax with PyMC? I’d love to explore this approach for future model extensions.
Thanks again for the guidance!