SlantedTriangular initialization failure

Hi,
Im trying to upgrade from 1.0-prerelease to v1.1.0.
When I retrained some of my models (sanity check), I get warnings for not-existing parameters for SlantedTriangular learning rate scheduler, but these parameters do exist in v1.1.0.
any idea?

INFO - allennlp.common.params - trainer.learning_rate_scheduler.type = slanted_triangular
WARNING - allennlp.common.from_params - Parameter num_epochs for class SlantedTriangular was found in both **extras and in params. Using the specification found in params, but you probably put a key in a config file that you didn't need, and if it is different from what we get from **extras, you might get unexpected behavior.
INFO - allennlp.common.params - trainer.learning_rate_scheduler.num_epochs = 2
WARNING - allennlp.common.from_params - Parameter num_steps_per_epoch for class SlantedTriangular was found in both **extras and in params. Using the specification found in params, but you probably put a key in a config file that you didn't need, and if it is different from what we get from **extras, you might get unexpected behavior.

Moreover, I noticed some of my models (with no warnings :wink:) performances dropped a bit, even though the code hasn’t been changed, nor the seeds. I suspect its the validation metrics computation (for early stopping; Im using F1Measure, FBetaMeasure), since the number of training epochs has changed. To check that Im training with no early stopping…
Any other idea? How can I check what went wrong?

Thanks!