RouteWeaver: Weaving Mode Selection and Execution into Unified LLM Routing

一句话要点

RouteWeaver 把单轮、多轮和智能体式路由统一进一个 LLM 策略，同时学习用哪种模式、模式内选哪个模型；配套的 COMET-GRPO 把优势信号拆到模式选择和模式内执行两层分别归因。

RouteWeaver unifies single-round, multi-round and agentic routing in one LLM policy that learns both which mode to use and which models to call; its COMET-GRPO splits the trajectory advantage into separate signals for mode selection and within-mode execution.