fig2

From: Resource-efficient federated large language models: challenges and mechanisms

Figure 2. Federated LLM framework and resource-accounting view. Heterogeneous clients adapt LLM-specific objects through communication and aggregation, while total cost is accounted across client-side execution, network exchange, server-side processing, orchestration, and trust/lifecycle operations at a comparable utility target. LLM: Large language model; PEFT: parameter-efficient fine-tuning.