Reproduce “Replacing Large Language Models with Jev Decision Models for Low-Latency Edge Service Orchestration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Delong Li, Xu Wang, Haochen Gong, Rui Lang, Guangsheng Yu
Jev interprets natural-language edge-service requests using parallel typed decisions for service, locality, quality, and urgency, then sends them through a shared validator and scheduler. In a modeled load sweep with four interpretation slots, it achieves 0.907 exact, on-time completion at 16 requests/s, versus 0.093 for DeepSeek. Its advantage is bounded: exact match falls to 0.567 for eight-field contracts.
Key insights
- For bounded contracts, interpretation speed affects system capacity: slow calls occupy admission slots and leave less time for service execution, especially under high load.
- Jev’s latency is relatively stable as inputs and catalogs grow, but its exact-match accuracy weakens as contracts widen; it is therefore better suited to concise decision schemas than broad extraction tasks.
- Caching validated decisions for repeated text narrows the gap between interpreters, making Jev most useful when requests require fresh interpretation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.