arXiv：cs.CL· Xiaohan Wang, Haozhen Zhang, Qingyuan Liu, Tao Feng, Wenya Wang·· 7 小时前
RouteWeaver：将模式选择与执行织入统一 LLM 路由
RouteWeaver: Weaving Mode Selection and Execution into Unified LLM Routing
AI 导读
RouteWeaver 是一个统一路由框架，用单一 LLM 策略整合单轮、多轮、智能体等路由模式，联合学习模式选择与模式内模型选择。配套的 COMET-GRPO 将轨迹级优势解耦为模式选择与模式内执行两个信号，配合渐进式探索让路由先学会执行再学会自主选模式。在数学、编程、知识、逻辑推理与事实召回五类任务上平均准确率 61.4%，相对最强基线提升 11.03%，且五个领域均超越全部对比基线。
正文
Abstract:LLM routing dynamically selects and coordinates heterogeneous language models across diverse tasks. However, existing approaches are largely built around a predefined execution paradigm, such as single-round, multi-round, or agentic routing, which fixes how model calls are organized and limits adaptation to different query needs. We propose RouteWeaver, a unified routing framework that integrates these routing modes within a single LLM policy, jointly learning which mode to use and how to select models within that mode. To jointly optimize mode selection and within-mode execution, we develop COMET-GRPO (COordinated Mode-Execution Training), which decouples the trajectory-level advantage into separate signals for mode selection and within-mode execution, enabling more targeted credit assignment to each decision level. We further introduce progressive exploration, which gradually shifts training from externally assigned modes to autonomous selection, so that the router first learns how to execute each mode before deciding when to use it. We evaluate RouteWeaver across mathematics, coding, knowledge, logical reasoning, and factual recall. RouteWeaver achieves an average accuracy of 61.4%, an 11.03% relative improvement over the strongest baseline, and outperforms all compared baselines in each of the five domains. Ablations show that both COMET-GRPO and progressive exploration are important for adaptive routing; without progressive exploration, the routing policy collapses entirely to single-round execution.
| Comments: | Submitted to ICLR 2027 |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.19705 [cs.CL] |
| (or arXiv:2610.19705v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19705 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xiaohan Wang [view email]
[v1]
Wed, 14 Oct 2026 07:33:05 UTC (2,987 KB)
来源：arXiv：cs.CL · arxiv.org