AI Paper Atlas

Research map / Large language models / Alignment and preferences

Keep the Best, Forget the Rest: Reliable Alignment with Order-Aware Preference Optimization

Jiahui Zhu, Yuanjie Shi, Xiyue Peng, Xin Liu, Yan Yan, Honghao Wei

ICLR 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO

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
filter out the 1 highest-loss sample per batch (RAPPO-1) vs DPODPO · IMDb sentiment (GPT2-Large)reward score: 1.5513 -> 1.6625Improved
filter out 4 highest-loss samples per batch (RAPPO-4) vs DPODPO · IMDb sentiment (GPT2-Large)reward score: 1.5513 -> 1.6811Improved
add order-aware filtering (RAPPO-4) vs DPODPO · Toxicity (GPT-Neo-2.7B)toxicity (%): 6.30 -> 2.28Improved
RAPPO-4 vs SimPOSimPO · IMDb sentimentreward score: 1.5537 -> 1.6811Improved
raise filtered q from 1 to 2 (ablation)RAPPO · IMDb batch 32 tau=1.0reward score: 1.6625 -> 1.6790Improved
raise filtered q from 4 to 8 (over-filtering) (ablation)RAPPO · IMDb batch 32 tau=1.0reward score: 1.6811 -> 1.6432Hurt
vary q across 1,2,4,8 against SimPO (ablation)RAPPO · IMDb batch 16 and 32reward score: 1.6600 -> 1.7481No clear effect
vary threshold tau from 0.8 to 1.2 at q=4 (ablation)RAPPO · IMDb batch 32reward score: 1.6765 -> 1.6828No clear effect
use RAPPO instead of DPODPO · PKU-SafeRLHF (Mistral-7B)safety rate: 55.89% -> 57.26%Improved
use RAPPO instead of DPODPO · PKU-SafeRLHF (Mistral-7B)Beaver-Help helpfulness: 0.51 -> 0.69Improved
use RAPPO instead of DPODPO · PKU-SafeRLHF (Mistral-7B)Beaver-Harm: 0.45 -> 0.36Improved
RAPPO vs SimPO under four LLM judgesSimPO · Summarization (Llama-3.1-8B)win rate: 58.30% to 72.00%Improved

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