Universal Reasoner: A Single, Composable Plug-and-Play Reasoner for Frozen LLMs

Jaemin Kim, Hangeol Chang, Hyunmin Hwang, Choonghan Kim, Jong Chul Ye

Keywords: reinforcement learning, large language model

Abstract: Large Language Models (LLMs) have demonstrated remarkable general capabilities, but enhancing skills such as reasoning often demands substantial computational resources and may compromise their generalization. While Parameter-Efficient Fine-Tuning (PEFT) methods offer a more resource-conscious alternative, they typically require retraining for each LLM backbone due to architectural dependencies. To address these challenges, here we propose Universal Reasoner (UniR) - a single, lightweight, composable, and plug-and-play reasoning module that can be used with larger frozen LLMs to endow it with specialized reasoning capabilities. Specifically, UniR decomposes the reward into a standalone reasoning module that is trained in a decoupled manner using predefined rewards, effectively translating trajectory-level signals into token-level guidance. Once trained, UniR can be combined with frozen LLMs at inference time by simply adding its output logits to those of the LLM backbones. This additive structure naturally enables modular composition: multiple UniR modules trained for different tasks can be jointly applied by summing their logits, enabling complex reasoning via composition. Experiments on mathematical reasoning and machine translation show that UniR surpasses existing baseline fine-tuning methods. Furthermore, UniR demonstrates weak-to-strong generalization: reasoning modules trained on smaller models effectively guide much larger LLMs. Beyond this, UniR generalizes across modalities such as in vision language models and domains such as medical reasoning. This makes UniR a cost-efficient, adaptable, and robust solution for enhancing reasoning in LLMs without compromising their core capabilities.

Primary Area: foundation or frontier models, including LLMs

Submission Number: 5828