ALPS: An Application-Aware, Length-Predictive Scheduler for Efficient LLM Inference at Edge

ALPS schedules LLM requests on edge devices by predicting output length per application, aiming to cut queueing delay versus first-come scheduling.