Smaller, Safer, More Transparent: Advancing Responsible AI with Gemma- Google Developers Blog

Introduces Gemma 2 2B, a 2B parameter model optimized for on-device and edge deployment using NVIDIA TensorRT-LLM and NIM. Also introduces ShieldGemma, a suite of safety content classifier models built on Gemma 2, and Gemma Scope, a model interpretability tool using sparse autoencoders (SAEs) for Gemma 2 2B and 9B models. Highlights on-device capabilities and responsible AI focus.