Reproduce “Pathwise-Budgeted LLM Routing: Online Allocation with K-Bit Tail-Cost Learning” 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
Hanzhang Qin, Sophie H. Yu, Yuewei Zhang, Zhi-Hai Zhang
ADP-HT routes tasks among language models under a hard spending limit. It estimates task-specific solve probabilities and conservative response-cost bounds from response-length categories, then uses them to select models and decide whether to retry. Because a completed call can cross the limit and stop later service, the method solved 216.2 of 400 MATH500 test problems on average at $0.025; this five-seed result is not an end-to-end guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.