How to turn on allow_unused=True through Pyro

1

I have a function that uses PyTorch library and sometimes it’ll throw this error

loss = csis.step()

“RuntimeError: One of the differentiated Tensors appears to not have been used in the graph. Set allowed_unused=True” if this is desired bahavior"

Is there a way to set this to true in Pyro?

2

Hi @DolanTheMFWizard, does this fix your issue? Allow unused params in CSIS inference by fritzo · Pull Request #2388 · pyro-ppl/pyro · GitHub

3

Hopefully. I just wrote code around it instead. But once this is merged I’ll try it out. Thanks!