Reproduce “Mind the Delay: Latency-aware Query Routing for Heterogeneous LLM pools” 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

Shivam Patel, Akaash R. Parthasarathy, Akash Dhasade, Ankur Mallick, Gauri Joshi

BEAR estimates quality, cost, and time-to-first-token (TTFT) for candidate model instances. Using workload snapshots and predicted output lengths, it replays serving batches and selects the highest-utility instance predicted to meet the query’s TTFT target. With vLLM and three Qwen3 instances, it achieved 30.5% higher OnTimeUtility than the best baseline at each tested arrival rate, within the evaluated deployment.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.