AI Paper Atlas

Research map / Large language models / Alignment and preferences

Normalized Rewards for Preference Optimization

Shawn Im, Federico Danieli, Skyler Seto, Barry-John Theobald, Katherine Metcalf

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
add probability-mass regularization and length normalization to DPO (N-DPO)DPO Llama-3.1-8B-Instruct · AlpacaEval 2 (LC)win rate: 21.72 -> 24.97Improved
add N-DPO regularizationDPO Mistral-7B-Instruct · AlpacaEval 2 (LC)win rate: 23.66 -> 25.48Improved
add N-DPO regularizationDPO OLMo-7B-SFT · AlpacaEval 2 (LC)win rate: 8.96 -> 9.88Improved
add N-SimPO regularizationSimPO OLMo-7B-SFT · AlpacaEval 2 (WR)win rate: 3.66 -> 10.02Improved
add N-SimPO regularizationSimPO Llama-3.1-8B-Instruct · AlpacaEval 2 (LC)win rate: 26.74 -> 25.54Hurt
add N-SimPO regularizationSimPO Mistral-7B-Instruct · AlpacaEval 2 (LC)win rate: 26.34 -> 26.05No clear effect
add regularization lambda>0 to length-normalized DPO (ablation)N-DPO Mistral-7B-Instruct · AlpacaEval 2 (LC)win rate: 20.21 -> 22.10Improved
add regularization lambda>0 to length-normalized DPO (ablation)N-DPO Llama-3.1-8B-Instruct · AlpacaEval 2 (LC)win rate: 28.50 -> 29.41Improved
add regularization lambda>0 to length-normalized DPO (ablation)N-DPO OLMo-7B-SFT · AlpacaEval 2 (LC)win rate: 7.34 -> 8.02Improved
add N-DPO regularizationDPO Llama-3.1-8B-Instruct · Common sense and reasoning avgscore change vs reference: -3.25 -> +3.04Improved
add N-DPO regularizationDPO Mistral-7B-Instruct · Common sense and reasoning avgscore change vs reference: -2.80 -> -0.98Improved
add N-SimPO regularizationSimPO OLMo-7B-SFT · Common sense and reasoning avgscore change vs reference: -0.73 -> -4.81Hurt

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