AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

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

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

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train the reasoning module with GRPO and add its logits to a frozen backbone, instead of GRPO full fine-tuningQwen2.5-3B · Math average (T=0.6)average pass@1: 40.3 -> 42.7Improved
same, on Llama backboneLlama3.2-3B · Math average (T=0.6)average pass@1: 25.4 -> 33.3Improved
add reward-trained reasoning module vs naive logit ensemble of pretrained module (ablation)Qwen2.5-3B · Math average (T=0.6)average pass@1: 35.2 -> 42.7Improved
use UniR instead of GRPO Full for translationLlama3.2-3B · IWSLT2017 EN-DEBLEU: 25.26 -> 27.03Improved
raise sampling temperature from 0 to 0.6 (ablation)GRPO Full (Llama3.2-3B) · MATH-500pass@1: 42.4 -> 36.1Hurt
raise sampling temperature from 0 to 0.6 (ablation)UniR (Llama3.2-3B) · MATH-500pass@1: 49.2 -> 48.8No clear effect
transfer 3B-trained reasoning module to a larger frozen backboneQwen2.5-14B · Math averageaverage pass@1: 51.1 -> 53.6Improved
add translation module to math task with German inputsfrozen backbone on MGSM German · German math problems to Englishmath accuracyImproved
increase weight alpha of math module versus translation module (ablation)composed UniR modules · German-to-English math taskmath accuracy vs translation qualityMixed
text-only trained math module guiding a VLMQwen2.5-VL-3B · MathVerse vision intensive and Geometry3kaccuracyImproved
compose readmission and LOS modules (ablation)UniR on MIMIC-IV · readmission accuracyaccuracy: 52.91 -> 44.88 accuracy; F1 26.77 -> 28.79Mixed
further train pretrained DeepSeek 1.5B module with UniR (ablation)Qwen2.5-3B backbone · Math averageaverage pass@1: 43.7 -> 46.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy