Comment by naasking

It's a two year old base model that's only 3B parameters, trained on only 100B tokens. It's still a research project at this point.