Google open-sources compact Japanese Gemma AI model optimized for local use on mobile devices.

Anyone from the local llama communities experienced with Gemma models?

I've heard good things but I only use Mistral because it's proved the most versatile.