My question seems stupid but I searched for an hour a light-weight code to fine tune RoBERTa like those used with Tensorflow hub (A large r
