LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?
By: Jingyuan Wang , Yankai Chen , Zhonghang Li and more
Potential Business Impact:
Smaller AI teaches bigger AI to think better.
Large language models (LLMs) have demonstrated remarkable progress in reasoning, often through supervised fine-tuning (SFT). However, SFT is resource-intensive, relying on large curated datasets, rejection-sampled demonstrations, and uniform optimization across all tokens, even though only a fraction carry meaningful learning value. In this work, we explore a counterintuitive idea: can smaller language models (SLMs) teach larger language models (LLMs) by revealing high-value reasoning moments that reflect the latter's unique strength? We propose LightReasoner, a novel framework that leverages the behavioral divergence between a stronger expert model (LLM) and a weaker amateur model (SLM). LightReasoner operates in two stages: (1) a sampling stage that pinpoints critical reasoning moments and constructs supervision examples capturing the expert's advantage through expert-amateur contrast, and (2) a fine-tuning stage that aligns the expert model with these distilled examples, amplifying its reasoning strengths. Across seven mathematical benchmarks, LightReasoner improves accuracy by up to 28.1%, while reducing time consumption by 90%, sampled problems by 80%, and tuned token usage by 99%, all without relying on ground-truth labels. By turning weaker SLMs into effective teaching signals, LightReasoner offers a scalable and resource-efficient approach for advancing LLM reasoning. Code is available at: https://github.com/HKUDS/LightReasoner
Similar Papers
Towards Reasoning Ability of Small Language Models
Computation and Language
Small AI models can learn to think well.
Incentivizing Strong Reasoning from Weak Supervision
Computation and Language
Teaches smart computers to think better cheaply.
Towards Reasoning Ability of Small Language Models
Computation and Language
Small AI can think as well as big AI.