Gemma 3N Preview: On-Device AI with Enhanced Multimodal Capabilities

Introducing Gemma 3N

• 🚀 Gemma 3N is announced as the latest model focused on unlocking compelling on-device experiences and leveraging portable compute.
• 💡 The 'N' in Gemma 3N signifies its core design around efficiency and state-of-the-art on-device performance, stemming from its lineage with nanoiz models.
• 🔑 Gemma 3N is built on the same technology as Gemini Nano models, offering developers an early preview of cutting-edge architecture for next-generation on-device experiences.

Enhanced Capabilities

• ✨ Gemma 3N delivers best-ever quality for a small model, meticulously tailored for on-device use cases.
• 🧠 Revolutionary capabilities include on-device function calling and interleaved text and image understanding.
• 🗣️ For the first time in the Gemma family, it can understand audio and video inputs, allowing natural interaction with applications.
• 🎯 Demonstrations showcased include saving notes, generating poems, calculating room dimensions, and identifying figures like Einstein.

Performance and Flexibility

• ⚡ Gemma 3N features the most efficient inference profile for on-device AI, optimized for mobile processors, leading to significantly faster processing.
• ⏱️ Prefill time is much faster compared to previous Gemma 34B models, with less memory usage, ensuring fluid and responsive applications.
• 🧩 A unique 2-in-1 transformer architecture includes a smaller embedded submodel, allowing dynamic choice between peak quality or greater speed and lower resource usage on the fly.

Open and Accessible AI

• 🌐 Gemma 3N embodies Google's commitment to open, accessible AI models for everyone.
• 🤝 It incorporates advancements from Android, Chrome, and Pixel teams, ensuring robustness for real-world on-device deployment.
• 🛠️ Developers can explore the models using various tools and providers, with excitement for what Gemma 3 unlocks.