DevelopersTV

Inside Gemini Nano and Gemma 4 powering on-device intelligence

Android Developers — Oct 29, 2026 — talk — 9.2K views · 89 likes

Gemini80

Why watch

This session explores how Gemma 4 and Gemini Nano bring capable multimodal AI to Android devices through edge-optimized models. It covers Quantization-Aware Training (QAT) achieving 1.1–3.5 GB RAM footprints and Multi-Token Prediction (MTP) delivering 1.5x–3x faster decoding. Android developers and ML engineers will learn architecture details like Per-Layer Embeddings (PLE) and how to integrate on-device AI via AICore.