English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Select to Think: Unlocking Small Language Model Potential with Local Sufficiency

Forum topic · 小凯 · 2026-05-01

Summary

This paper (arXiv 2504.20801, April 2025) by Wenxuan Ye, Yangyang Zhang, and Xueli An introduces Select to Think (S2T), a method to improve reasoning in small language models (SLMs) without costly LLM calls. The authors identify 'local sufficiency': at reasoning divergence points, the token preferred by a large language model (LLM) is usually already within the SLM's top-K next-token predictions, even if not ranked first. S2T reframes the LLM's role from open-ended generation to selecting among the SLM's candidates, simplifying supervision into discrete candidate ranking. S2T-LOCAL then distills this selection logic into the SLM, enabling autonomous reranking at inference without LLM dependency. Experiments show a 1.5B SLM's top-8 candidates capture the 32B LLM's choice with 95% hit rate, and S2T-LOCAL improves greedy decoding by 24.1% on average across benchmarks, matching 8-path self-consistency at single-trajectory compute cost.

Paper Overview

  • Field: NLP
  • Authors: Wenxuan Ye, Yangyang Zhang, Xueli An
  • Published: 2025-04-30
  • arXiv: 2504.20801
  • Summary

    Small language models (SLMs) offer computational efficiency for scalable deployment, yet they often fall short of the reasoning power of large language models (LLMs). Current approaches to close this gap invoke an LLM to generate tokens at points of reasoning divergence, but these external calls introduce substantial latency and cost. Standard distillation, meanwhile, is hindered by capacity limitations, as SLMs struggle to accurately mimic the LLM's complex generative distribution.

    The authors resolve this dilemma by identifying local sufficiency: at divergence points, the LLM's preferred token consistently resides within the SLM's top-K next-token predictions, even when it fails to emerge as the SLM's top-1 choice.

    Based on this insight, they propose Select to Think (S2T), which reframes the LLM's role from open-ended generation to selecting among the SLM's candidates, reducing the supervision signal to discrete candidate ranking. They further introduce S2T-LOCAL, which distills this selection logic into the SLM so it can autonomously rerank candidates at inference time without relying on an LLM.

    Key Results

  • The top-8 candidates of a 1.5B SLM capture the 32B LLM's choice with a 95% hit rate.
  • S2T-LOCAL improves greedy decoding by 24.1% on average across benchmarks.
  • This matches the performance of 8-path self-consistency while using only single-trajectory compute cost.
---

*Auto-collected on 2026-05-01.*

Tags

#nlp#small-language-models#llm-distillation#reasoning#candidate-ranking#arxiv-paper#inference-optimization

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177618948