AI Paper Atlas

Research map / Large language models / Alignment and preferences

AMaPO: Adaptive Margin-attached Preference Optimization for Language Model Alignment

Ruibo Deng, Duanyu Feng, Wenqiang Lei

AAAI 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 1 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 SimPO static margin with AMaPO adaptive marginSimPO · AlpacaEval 2, Llama3-8B-BaseLC win rate (%): 22.0 -> 26.4Improved
replace SimPO static margin with AMaPO adaptive marginSimPO · AlpacaEval 2, Mistral-7B-InstructLC win rate (%): 32.1 -> 34.5Improved
replace SimPO static margin with AMaPO adaptive marginSimPO · MT-Bench, Llama3-8B-BaseGPT-4 Turbo score: 6.6 -> 6.4Hurt
replace SimPO static margin with AMaPO adaptive marginSimPO · RM-Bench Hard, Llama3-8B-Baseranking accuracy (%): 23.3 -> 25.4Mixed
replace SimPO static margin with AMaPO adaptive marginSimPO · UltraFeedback Mutual-OOD, Mistral-7B-Baseranking accuracy (%): 75.94 -> 82.97Improved
replace SimPO static margin with AMaPO adaptive marginSimPO · UltraFeedback ID, Llama3-8B-Baseranking accuracy (%): 78.1 -> 77.33No clear effect
remove Z-normalization (ablation)AMaPO · AlpacaEval 2, Llama3-8B-BaseLC win rate (%): 26.3 -> 24.8 when removedImproved
add exponential scaling (ablation)AMaPO · AlpacaEval 2 WR, Llama3-8B-Basewin rate (%): 21.4 -> 17.6 when removedImproved
add adaptive margin (Z-norm plus exp scaling) (ablation)AMaPO · AlpacaEval 2, Llama3-8B-BaseLC win rate (%): 26.3 -> 20.7 when removedImproved
add adaptive margin (Z-norm plus exp scaling) (ablation)AMaPO · AlpacaEval 2, Llama3-8B-InstructLC win rate (%): 46.1 -> 47.2 when removedHurt
add zero margin for correctly ranked samples (ablation)AMaPO · MT-Bench, Llama3-8B-InstructGPT-4 Turbo score: 7.2 -> 6.7 when removedImproved
add zero margin for correctly ranked samples (ablation)AMaPO · AlpacaEval 2 LC, Llama3-8B-InstructLC win rate (%): 46.1 -> 48.6 when removedHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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