ResearchBenchmarksLLM Serving on Real-world trace (ShareGPT and BurstGPT)Follow91.13SLO AttainmentDiLaServe59.77467.914576.05584.1955Jun 27, 2026Evaluation ResultsMethodMethodLinksSLO AttainmentOverall ScoreAverage Latency (Overall)Average Latency (Low-load)Average Latency (High-load)DiLaServeConfiguration=Dynamic...Configuration=Dynamic reconfiguration, Hardware=16-GPU cluster2026.0691.136.85.752.266.33INFaaSConfiguration=Fixed 16...Configuration=Fixed 16 TP-1 workers, Hardware=16-GPU cluster2026.0670.036.898.82.719.34LlumnixConfiguration=Fixed 16...Configuration=Fixed 16 TP-1 workers, Hardware=16-GPU cluster2026.0660.986.8910.643.2311.3