Google DeepMind speakers Gus Martins and Ian Ballantyne describe the Gemma model family and the reasons an organization might run or customize an open model on its own infrastructureA local model runs on computing infrastructure controlled directly by the user rather than only through a remote model service.. They discuss model sizes, licensing and hardware requirementsAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models..
Ian Ballantyne demonstrates on-device and local workflows and discusses tradeoffs involving latency, energy, maintenance and task-specific evaluationEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. Performance and product claims are attributed to the presenters.
Watch on YouTube
