Stefan Radev

We also need a function that validates if NaNs, Infs, etc are present in the approximator / simulator outputs. The latter is especially important, as it can help users immediately...

Having `batch_size` as default for all flat (non-hierarchical models) is definitely nice. I can definitely see where `batch_shape` can also come in handy for dynamic / hierarchical models.

I second that. We can also emphasize `batch_size` (int) for everyday users and also allow `batch_shape` (tuple, None as default) for power users or special cases, which overrides `batch_size`.

Yup, generally, this needs a slight re-design, which will come together with inferring the batch_size / shape. For now, you can avoid the bug by setting `momentum=None`.

@marvinschmitt Is this still open given the recent developments?

That's pretty cool!

Checking on the state of it now that the dev branch is stable enough.

Hey Erik, indeed, the default hyperparameter settings are not the same between branches, but those for the `CouplingFlow` are pretty close. For instance, flow matching will be pretty shallow when...

Hi Jerry, can you pls make sure that the linter is passing?