I'm so excited for when their 70B version comes out!

I'm wondering how much of a factor the larger token dictionary plays in this progress. Seems like greater token lexicon translates to better learning. What do you think?