Comment by simopa

It's crazy to see a 400B model running on an iPhone. But moving forward, as the information density and architectural efficiency of smaller models continue to increase, getting high-quality, real-time inference on mobile is going to become trivial.