We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
DeepMind Blog·May 20·5 min read
Announcing Gemma 3n preview: Powerful, efficient, mobile-first AI
TL;DR
Mobile devices often lack the processing power for advanced AI applications. Gemma 3n, a new AI model, utilizes Per-Layer Embeddings to reduce memory usage significantly.
✦ Why It Matters
Engineers can leverage Gemma 3n to build efficient AI applications that run directly on mobile devices.
Key Takeaways
- Experiment with Gemma 3n in Google AI Studio for cloud-based development.
- Utilize Google AI Edge for local integration of Gemma 3n.
- Leverage advanced audio and multilingual features for your applications.
- Explore dynamic submodel creation for tailored performance and quality.
Full Summary
Gemma 3n is the latest addition to Google DeepMind's AI models, following the successful launches of Gemma 3 and Gemma 3 QAT. This model is engineered for mobile devices, enabling powerful AI experiences directly on smartphones, tablets, and laptops.
It features a new architecture developed in collaboration with hardware leaders like Qualcomm and Samsung, optimizing performance and memory usage through innovations like Per-Layer Embeddings (PLE). With a parameter count of 5B and 8B, Gemma 3n can run with a memory footprint similar to smaller models, making it suitable for on-device applications.
Key capabilities include improved audio processing, multilingual support, and the ability to dynamically create submodels for specific tasks. Developers can access Gemma 3n through Google AI Studio and Google AI Edge for both cloud-based and local development.