Compare different caching strategies for frequently requested model outputs in a GenAI system at Google DeepMind.