AI Paper Atlas

Research map / Large language models / Alignment and preferences

RewardEval: Advancing Reward Model Evaluation

Saumya Malik, Valentina Pyatkin, Sander Land, Jacob Morrison, Noah A. Smith, Hannaneh Hajishirzi, Nathan Lambert

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models

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
train reward model on combined Tulu and Skywork preference data instead of either alone (ablation)Tulu 8B SFT reward model · RewardBench 2average accuracy: 63.5 (Tulu) and 66.7 (Skywork) -> 68.2 (both)Improved
initialize reward model from Llama 8B Instruct instead of Llama 8B Base (ablation)Llama 8B reward model · RewardBench 2average accuracy: 64.9 -> 72.8Improved
initialize reward model from Qwen 7B Instruct instead of Qwen 7B Base (ablation)Qwen 7B reward model · RewardBench 2average accuracy: 68.2 -> 73.3Improved
initialize from Tulu DPO or RL rather than Tulu SFT (ablation)Tulu 8B reward model · RewardBench 2average accuracy: 68.2 -> 68.7; Safety 89.8 -> 86.0Mixed
add Skywork preference data to Tulu data (ablation)Tulu 8B SFT reward model · RewardBench 2 Focusaccuracy: 71.3 -> 93.7 Focus; 81.1 -> 88.0 SafetyImproved
use Qwen 7B Instruct as reward model base for Math (ablation)reward model · RewardBench 2 Mathaccuracy: Qwen 7B Instruct 71.6 vs Tulu 70B SFT 67.8Improved
train reward models for two epochs rather than one (ablation)reward model training · RewardBench 2average accuracy: eight among the eighteen best models trained for two epochsImproved
scale reward model base from 8B to 70B (ablation)Tulu SFT reward model, both data · RewardBench 2average accuracy: 68.2 -> 72.2Improved
use benchmark average to predict best-of-N downstream performanceRewardBench 2 · BoN over 113 RMsPearson correlation: 0.87Improved
use reward model from different lineage than the policy in PPO (ablation)PPO with Tulu 3 8B SFT policy · Tulu 3 eval suite averageaverage score: 72.9 RB2 -> 54.5 PPO versus 49.8 RB2 -> 60.2 PPO for on-policy RMHurt
train PPO with the best on-policy reward model instead of the SFT policy or DPO modelTulu 3 8B SFT · Tulu 3 eval suite averageaverage score: 54.1 -> 60.3Improved
increase reward model RB2 score among decent on-policy RMs for PPOPPO · Tulu 3 eval suite averageaverage score: RB2 49.8 to 68.5 all give similar PPO scoreNo clear effect

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