On-device small language models with multimodality, RAG, and Function Calling- Google Developers Blog

Introduces Gemma 3n, a multimodal on-device SLM supporting text, image, video, and audio inputs. Expands support to over a dozen models on the LiteRT Hugging Face community. Introduces new on-device RAG and Function Calling libraries for Android. Details int4 post-training quantization for reduced model size and latency. Highlights Gemma 3 1B's performance on mobile GPUs. Mentions LiteRT APIs and the AI Edge Portal for benchmarking and evaluations.