Fritz Obermeyer

> Would you suggest Whatever works for you, I was only looking for a minimal reproducible example and would have expected something like ```python torch.save(model, path_to_model) torch.save(guide, path_to_guide) pyro.get_param_store().save(path_to_parameters) pyro.clear_param_store()...

Hmm, I'm not sure what else to try. Is there any way you could make a reproducible example we could look at?

What are the differences between your original model and saved-then-loaded model? Do you save any randomly-generated tensors in the model? Have you considered using `torch.save()` and `torch.load()` for the whole...

Hi @Eric-Bradford this looks like a bug to me, let me take a look at what might be causing it...

OK, this is not a bug, merely a historical design choice (pyro.infer.inspect originated as an inference tool that was concerned with only latent variables, hence it ignored deterministic dependencies). I've...

I also had an import error after `pip install`ing, but `conda install`ing worked.

Agreed, this does seem like a sharp edge. One option would be to emit a warning if the param store is not empty during the first call of `SVI.step()` for...

Thanks for pointing this out, I'll look into it.

It looks like we'll need to get `inspect.signature(Normal)` to print the more informative string, since that's [what help() uses](https://github.com/python/cpython/blob/4173320920706b49a004bdddd8d7108e8984e3fc/Lib/pydoc.py#L1321). Here's a minimal working example: ```sh python -c 'import pyro.distributions as...

It looks like this is failing again as of Pyro 1.8.0. @fehiepsi have you noticed anything else that's changed in that you've had to fix in NumPyro? EDIT specifically, I...