Hi, I'm new to active learning and modAL. I want to fine-tune a RoBERTa model on a (yet) unannotated dataset with active learning. Has anyone tried applying modAL to huggingface models like RoBERTa or DeBERTa? I wonder if the two modules are ever compatible.