AI has entered an exciting new phase. It’s no longer just about describing the world—it’s about interacting with it. This shift is driven by tool-use, where AI predicts and executes function calls, such as opening apps or adjusting settings. Google’s latest innovation, showcased in the Google AI Edge Gallery, demonstrates the incredible potential of on-device function calling.
Takeaways 🚀
✅AI enables real-time interactions offline.
✅Crawler impact notes on products.
✅Mobile Actions demo showcases efficiency.
✅Tiny Garden highlights interactive gaming.
✅Available now on Android, iOS.
Why On-Device AI Matters
To begin with, the magic of AI lies in its ability to act in real-time. By moving tool-use directly onto devices, Google enables instant interactions without relying on internet connectivity. As a result, tasks like creating calendar events, navigating maps, or turning on a flashlight now happen seamlessly, even offline.
However, achieving this wasn’t easy. Traditional function-calling models are large and resource-intensive, making them unsuitable for mobile devices. To address this, Google compressed these models to fit mobile hardware while maintaining both accuracy and battery efficiency.
Meet FunctionGemma: The Engine Behind It All
At the heart of this breakthrough is FunctionGemma, a compact AI model with 270 million parameters. Specifically, it translates natural language commands into actions. For instance, it can handle requests like “Show me the San Francisco airport on the map” or “Plant sunflowers in the top row and water them,” and execute them directly on the device.
Interactive Demos: Mobile Actions and Tiny Garden
To showcase FunctionGemma’s capabilities, the Google AI Edge Gallery app includes two standout demos:
- Mobile Actions: First, this demo lets you perform tasks like creating calendar events or navigating maps using natural language commands. Most importantly, it works entirely offline, making it both fast and reliable.
- Tiny Garden: Next, this mini-game allows you to manage a virtual garden with voice commands. For example, you can say, “Plant sunflowers” or “Water the top row,” and the model will execute these actions in the game. Consequently, it’s a fun and interactive way to see how FunctionGemma adapts to custom app logic.
Now Available on Android and iOS
In addition to its features, Google has expanded the AI Edge Gallery to both Android and iOS platforms. This ensures that developers and users across ecosystems can experience the benefits of on-device AI. Furthermore, the app includes features like multi-turn AI chat, local transcription, and performance benchmarking.
Test Model Performance on Your Device
For those curious about performance, the app also lets you benchmark FunctionGemma on your device. For example, on a Pixel 7 Pro, the model processes 1916 tokens per second during prefill and 142 tokens per second during decoding. To test this yourself, follow these steps:
- Open the app menu.
- Select the Models section.
- Run the benchmark tool to see how FunctionGemma performs on your hardware.
Build Your Own AI Solutions
Moreover, developers can fine-tune FunctionGemma to fit their specific app needs. The app provides tools and guides to help you customize AI solutions for your projects. Whether you’re building a game, a productivity tool, or something entirely new, FunctionGemma makes it possible.
The Future of Interactive AI
In conclusion, Google’s on-device function calling is a game-changer. It makes AI faster, more interactive, and accessible to everyone. By enabling real-time, offline interactions, this technology enhances user experiences and opens new doors for developers.
So, are you ready to explore? Download the Google AI Edge Gallery app on Android or iOS. Whether you’re a developer or a curious user, this is your chance to see the future of AI in action.
