Pre-Training and Fine Tuning in NLP

The next category is training. Right now, the situation in NLP for those of us who are users and practitioners is we go to a model hub, there's a thousand different embeddings to choose from. It works well, but usually not well enough for the deploy because it was trained on general data. Then we have the fine tuning step, right? So what is the situation in terms of research and pre-training and fine tuning? Yeah.