AI Paper Atlas

Research map / Large language models / Alignment and preferences

Autoregressive Direct Preference Optimization

Masanari Oi, Mahiro Ukai, Masahiro Kaneko, Naoaki Okazaki, Nakamasa Inoue

ICML 2026 · Large language models › Alignment and preferences · approach: DPO design choices · 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
replace DPO's summed log-ratio with prefix-wise ADPO loss (m=256)DPO, Llama-3-8B · GSM8Kaccuracy: 64.37 -> 68.08Improved
replace DPO loss with ADPO (m=256)DPO, Qwen-3-8B · MATH500accuracy: 53.80 -> 55.40Improved
apply ADPO to cDPO (cADPO)cDPO, Qwen-3-8B · GSM8Kaccuracy: 90.98 -> 91.74Improved
apply ADPO to cDPO (cADPO)cDPO, Llama-3-8B · MATH500accuracy: 16.80 -> 20.20Mixed
use ADPO instead of DPO on conversation tasksDPO, Llama3-8B-Instruct · AlpacaEval 2 (LC)win rate: 40.3 -> 45.8Improved
use ADPO instead of SimPOSimPO, Llama3-8B-Instruct · AlpacaEval 2 (LC)win rate: 44.7 -> 45.8Improved
use few segments (adaptive m=2) instead of full-response DPO (ablation)ADPO, Llama-3-8B · GSM8Kaccuracy: 64.37 -> 64.29No clear effect
increase adaptive segments to m>=128 (ablation)ADPO · GSM8K and MATH500accuracyImproved
finer static window (k<=4) vs DPO (ablation)ADPO static family · GSM8K and MATH500accuracyImproved
token-level static ADPO (k=1) (ablation)ADPO, DeepSeek-Math-7B · GSM8Kaccuracy: 67.78 -> 70.36Improved
adaptive vs static family (ablation)ADPO · GSM8K and MATH500accuracyImproved
increase m from 256 to 512 (ablation)ADPO adaptive, Llama-3-8B · GSM8Kaccuracy: 68.08 -> 67.32Hurt

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