Large Language Models in the Cloud Network
摘要
This chapter examines how large language models are developed and operated within cloud environments, focusing on the data pipelines, computational workflows, and system mechanisms that support large-scale training, fine-tuning, and inference. It analyzes the communication, storage, and scheduling requirements that arise when models interact with distributed accelerator clusters and retrieval systems and explains how these requirements shape practical design decisions in modern LLM services. By establishing the cloud-side foundations of LLM operation, the chapter provides the necessary background for later chapters on cloud-edge collaboration, on-device intelligence, and network-assisted model optimization.