Characterization of Request and Token Energy Costs for LLM Inference Workloads on GPU Platforms

arXiv:2608.28044 · cs.PF, cs.DC, cs.LG · Submitted 2026-08-28 · Read on arXiv