Comment by alyxya

The blog post was published a couple months ago, and it looks like there hasn't been a follow-up release with the fully trained model. I'm not sure if there's much to take away from an early checkpoint besides the unique architectural choices they made in their model for faster inference.