Performance of Large Language Models Across Edge and Cloud Platforms in Smart Spaces

Smart spaces integrate sensing and computing technologies with artificial intelligence (AI) to enhance adaptability and user convenience. Incorporating Large Language Models (LLMs) into these environments can further enhance their capabilities, transforming them into interactive spaces that provide user-centric and personalized services. In this paper, we present a hybrid edge-cloud approach for a generic smart space that integrates open-source LLMs with commonly available edge devices (laptops and Raspberry Pis), a commercial cloud platform, historical environmental data, real-time sensor data streams, and LLM prompts. This integration enables various types of analytics including Descriptive, Diagnostic, Predictive, and Prescriptive, ranging from basic to advanced analyses. Our proposed approach enables dynamic selection between edge and cloud resources to evaluate the performance of LLMs in terms of response latency and accuracy, leveraging these data analytics. Our evaluation results demonstrate the potential of LLMs to transform smart spaces into more intelligent and interactive environments that deliver enhanced user-centric services.