Comment by armeet

When you step back, it's kind of absurd that a 1B model can achieve this eval score. Makes me very excited about the future of on-device inference.