AI Paper Atlas

Research map / Large language models / Alignment and preferences

Pre-DPO: Improving Data Utilization in Direct Preference Optimization Using a Guiding Reference Model

Junshu Pan, Wei Shen, Shulin Huang, Qiji Zhou, Yue Zhang

AAAI 2026 · Large language models › Alignment and preferences · approach: DPO design choices · 5 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
re-optimize DPO with a guiding reference model from a first DPO round (Pre-DPO)DPO · AlpacaEval 2 (Llama3.2-3B-Base)LC win rate (%): 10.5 -> 12.5Improved
use SimPO-optimized model as guiding reference for DPOSimPO · AlpacaEval 2 (Llama3.2-3B-Base)LC win rate (%): 13.1 -> 18.1Improved
add guiding reference model from DPODPO on Qwen2.5-7B-Instruct · Arena-Hard v0.1win rate (%): 62.9 -> 68.8Improved
add guiding reference model from DPODPO on Qwen2.5-7B-Base · Arena-Hard v0.1win rate (%): 33.1 -> 32.6Hurt
train DPO for 2 epochs to match Pre-DPO compute (ablation)DPO on Llama3.2-3B-Base · AlpacaEval 2LC win rate (%): 10.5 -> 11.0No clear effect
train DPO for 2 epochs to match Pre-DPO compute (ablation)DPO on Llama3.2-3B-Instruct · AlpacaEval 2LC win rate (%): 36.3 -> 35.2Hurt
use guiding reference from DPO after 1 epoch instead of 2nd DPO epoch (ablation)DPO on Llama3.2-3B-Instruct · AlpacaEval 2LC win rate (%): 35.2 -> 39.3Improved
use ORPO-optimized model as guiding referenceORPO · AlpacaEval 2 (Llama3.2-3B-Base)LC win rate (%): 10.2 -> 12.3Improved
use sDPO-optimized model as guiding referencesDPO · AlpacaEval 2 (Llama3.2-3B-Base)LC win rate (%): 12.0 -> 12.9Improved
use TR-DPO-optimized model as guiding referenceTR-DPO · AlpacaEval 2 (Llama3.2-3B-Base)LC win rate (%): 11.7 -> 12.8Improved
use SimPO guiding referenceSimPO on Qwen2.5-7B-Base · AlpacaEval 2 (response length)avg response length: 1836 -> 1758Improved
hard reference update every 32 steps (TR-DPO) vs guiding reference (ablation)DPO on Llama3.2-3B-Base · lambda distributionweight spreadHurt

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