Gemma 3 270M: The compact model for hyper-efficient AI