AI Paper Atlas

Research map / Large language models / Alignment and preferences

Token-Importance Guided Direct Preference Optimization

Ning Yang, Hai Lin, Yibo Liu, Baoliang Tian, Guoqing Liu, Haijun Zhang

ICLR 2026 · Large language models › Alignment and preferences · approach: DPO design choices · 26 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
Replace DPO with TI-DPO (token weights + triplet loss)DPO · six-benchmark averageavg score: 57.7 -> 62.3Improved
TI-DPO vs GRPO on averageTI-DPO · six-benchmark averageavg score: 62.3 vs 62.1No clear effect
Add triplet loss (ablation)TI-DPO · Llama-3.2-3B-Instruct ablationMath: 79.0 -> 80.7Improved
Add triplet loss (ablation)TI-DPO · Llama-3.2-3B-Instruct ablationReliability: 83.0 -> 86.8Improved
Use gradient-attribution importance weights instead of uniform weights (ablation)TI-DPO · Llama-3.2-3B-Instruct ablationReasoning: 30.5 -> 34.6Improved
Use gradient-attribution weights instead of random weights (ablation)TI-DPO · Llama-3.2-3B-Instruct ablationInstr-Follow: 55.0 -> 63.5Improved
Add Gaussian prior to weights (ablation)TI-DPO · Llama-3.2-3B-Instruct ablationReliability: 82.5 -> 86.8Improved
Use Gaussian prior instead of softmax prior (ablation)TI-DPO · Llama-3.2-3B-Instruct ablationReliability: 81.0 -> 86.8Improved
Use TI-DPO instead of DPO under 40% label noiseDPO · label-noise robustnessaccuracy: 60.1 -> 68.3Improved
Use TI-DPO instead of DPODPO · generation diversityS-BLEU: 34.2% -> 30.1%Improved
Use TI-DPO instead of GRPO on reasoning-heavy tasksGRPO · GSM8Kaccuracy: 75.7 -> 73.0Hurt

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