CONSONANCE.for your information
Monday, 5 October 2026frenvi

Worth reading closely

01 — inference 10 upvotes

SeLMRoute: Probabilistic Semantic Evidence for Large Language Model Routing

QUESTION — How can we improve LLM routing decisions by leveraging independent semantic evidence rather than relying solely on query embeddings or model representations?

The paper proposes SeLMRoute, an LLM routing framework that separates the extraction of candidate-independent semantic evidence from learning model performance. A decision model first evaluates interpretable questions about a query's reasoning requirements and external knowledge use as probability distributions. A lightweight supervised router then uses this probabilistic semantic state to estimate candidate performance for performance-oriented and cost-aware decisions. On LLMRouterBench (15 datasets, 11,481 queries), SeLMRoute achieves an average accuracy of 72.08% pm 0.45, while grouped five-fold cross-evaluation reaches 72.64%, compared with 69.23% for the strongest fixed candidate.

On the LLMRouterBench (15 datasets, 20 candidate models, 11,481 queries), SeLMRoute achieves an average accuracy of 72.08% pm 0.45, while grouped five-fold out-of-fold evaluation reaches 72.64%.

Compared with 69.23% for the strongest fixed candidate.

In a separate 13-model performance-cost setting, SeLMRoute improves performance in all five grouped splits, with a mean PerfGain of 2.66%.

nikopavl · 28 Sept 2026 read the original ↗
↑