AI Paper Atlas

Research map / Large language models / Alignment and preferences

DuPO: Enabling Reliable Self-Verification via Dual Preference Optimization

Shuaijie She, Yu Bao, Yu Lu, Lu Xu, Tao Li, Wenhao Zhu, Jianbing Zhang, Shujian Huang, Shanbo Cheng, Lu Lu, Yuxuan Wang

ICLR 2026 · Large language models › Alignment and preferences · approach: DPO design choices

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
apply DuPO RL trainingSeed-X-7B-Instruct · Flores 756 directionsCOMET: 86.96 -> 89.09Improved
apply DuPO RL trainingQwen3-4B · AMC23/AIME24/AIME25 averageaccuracy: 77.2 -> 83.6Improved
apply DuPO RL trainingDeepSeek-R1-Distill-Qwen-1.5B · AMC23/AIME24/AIME25 averageaccuracy: 35.8 -> 39.7Improved
apply DuPO RL trainingOpenReasoning-Nemotron-7B · AIME24accuracy: 83.3 -> 83.3No clear effect
apply DuPO RL trainingOpenReasoning-Nemotron-7B · AIME25accuracy: 73.3 -> 90.0Improved
apply DuPO instead of SimpleRL-Zoo (oracle labels)LlaMA-3.1-8B · AMC23/MATH500 averageaccuracy: 19.0 -> 32.1Improved
apply DuPO RL trainingOctoThinker-8B-Hybrid-Base · AMC23/MATH500 averageaccuracy: 23.8 -> 62.5Improved
apply DuPO directly to base model without SFTQwen3-4B-Base · AMC23accuracy: 20% -> 70%Improved
use DuPO backward accuracy for inference-time rerankingQwen3-4B · AIME24/AIME25 averageaccuracy: 68.4 -> 77.7Improved
use DuPO backward accuracy for inference-time rerankingDeepSeek-R1-Distill-Qwen-1.5B · AIME24/AIME25 averageaccuracy: 20.0 -> 38.7Improved
add unknown component selection / dual question filtering (ablation)DuPO on DeepSeek-R1-Distill-Qwen-1.5B · math benchmarksaccuracy: 3.6 point drop without filteringImproved
add unknown component selection / dual question filtering (ablation)DuPO on Qwen3-4B · math benchmarksaccuracy: +5.4 points on average over poorer duality variantImproved

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