Gemma 2

Gemma 2 is a family of lightweight, high-performance open-weights language models developed by Google DeepMind, built using research and technology from the Gemini series. Available in sizes like 2B, 9B, and 27B parameters, it introduces architectural innovations such as grouped-query attention, logit soft-capping, and alternating local-global attention layers to deliver class-leading efficiency, stability, and text generation capabilities on consumer hardware.