GeneralScheduler with discriminative learning rates?

is it possible to create TrainingPhases for the GeneralScheduler and use discriminative learning rates with it? After not finding specifics in the docs, I have naively tried passing slice(lr) to the lr-tuple in the trainingphase, but that didn’t go well. :wink:
If it is is not possible, any suggestions on how to do that “manually”?
Any insights would be greatly appreciated! Thanks.

Slices are changed in numpy arrays by the learner behind the scenes. Try passing a numpy array in each one of your TrainingPhase while waiting for me to apply the same magic here :wink:

1 Like

I‘ll try that, thanks for the response!