Train faster static embedding models with sentence transformers

Very interesting. I wonder if they tried this on classification as well. They mention but then don't talk about it in detail.