If the initial benchmarks scores (and graphs used for PR) showcased a 3x reduction in size for the same performance, I think the broader public reception would have been less tepid. Just looking at this, it just seems 7% behind other existing models in the 4.x series…

→ View original post on X — @alexjc