RouteWeaver: NTU Unifies Mode Selection and Execution in LLM Routing

NanyangTechnologicalUniversity · hf · 2026-10-23

Researchers from NTU, HUST and UIUC propose RouteWeaver, a unified LLM routing framework that integrates single-round, multi-round and agentic routing into one policy. COMET-GRPO decouples trajectory-level advantages for mode selection vs within-mode execution, plus progressive exploration, improving routing accuracy across reasoning tasks.

Original post →

More from coding & agent

coding & agent channel →