Positive embeddings contraint

fastai
1

I’m just wondering if it is possible to put a constraint to the embeddings of a tabular_learner model so that all numbers are bigger than zero.