R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing

Open in new window