Reproduce “RouteWeaver: Weaving Mode Selection and Execution into Unified LLM Routing” 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
Xiaohan Wang, Haozhen Zhang, Qingyuan Liu, Tao Feng, Wenya Wang
RouteWeaver lets one policy choose whether to use single-round, multi-round, or agentic model calls, as well as which frozen Worker models to call. Its COMET-GRPO training separates signals for mode selection and execution. Across nine benchmarks in five domains, it averages 61.4% accuracy versus Router-R1’s 55.3% at α = 0, leading in all domains; ablations show lower scores.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.