search

bayesflow-org / bayesflow

A Python library for amortized Bayesian workflows using generative neural networks.
https://bayesflow.org/
MIT License
361 stars 50 forks source link

Flow matching with normal base distribution returns nan loss #208

Closed maltelueken closed 1 week ago

maltelueken commented 2 weeks ago

When using a continuous approximator with flow matching plus set transformer (or deepset) and base_distribution="normal", the loss becomes nan after two steps.

Is there anything that I missed to add MMD to the summary loss?

marvinschmitt commented 2 weeks ago

Thanks for reporting the issue. Could you please provide a minimal reproducible example, at least of the exact definitions of your networks?

LarsKue commented 1 week ago

Closing this as not reproducible.