AI Paper Atlas

Research map / Large language models / Alignment and preferences

GRAM-R²: Self-Training Generative Foundation Reward Models for Reward Reasoning

Chenglong Wang, Yongyu Mu, Hang Zhou, Yifu Huo, Ziming Zhu, Jiali Zeng, Murun Yang, Bei Li, Xiaoyang Hao, Chunliang Zhang, Fandong Meng, Jingbo Zhu and 1 more

AAAI 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models · 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 reward reasoning with self-training on unlabeled data (vs Generative RM)LLaMA-3.1-8B-Instruct reward model · RM-Bench Overallaccuracy: 79.2 -> 85.7Improved
add reward reasoning with self-training (vs Generative RM)LLaMA-3.1-8B-Instruct reward model · JudgeBench Overallaccuracy: 76.9 -> 81.0Improved
add reward reasoning with self-training (vs Generative RM)LLaMA-3.2-3B-Instruct reward model · RM-Bench Overallaccuracy: 77.1 -> 83.8Improved
add voting@16 at test timeGRAM-R2 8B · RM-Bench Overallaccuracy: 85.7 -> 86.1Improved
add reward reasoning (preference-proving rationales) (ablation)GRAM-R2 · RM-Bench Overallaccuracy: 80.4 -> 85.7Improved
add preference-proving model during self-training on unlabeled data (ablation)GRAM-R2 · RM-Bench Overallaccuracy: 84.1 -> 85.7Improved
add rationale-free labeled data pre-training (ablation)GRAM-R2 · RM-Bench Overallaccuracy: 82.6 -> 85.7Improved
use trained preference-proving model Qwen3-14B instead of prompting GPT-4o (ablation)GRAM-R2-100k · RM-Benchaccuracy: 64.7 -> 74.3Improved
scale preference-proving model from Qwen3-14B to Qwen3-32B (ablation)GRAM-R2-100k · RM-Benchaccuracy: 74.3 -> 74.8No clear effect
use GRAM-R2 as PPO reward model instead of generative baselineLlama-3.1-8B policy · AlpacaEval2 WinRatewin rate: 11.62 -> 15.62Improved
use GRAM-R2 as PPO reward model instead of Skywork-Reward-8BLlama-3.1-8B policy · AlpacaEval2 LC-WinRatelength-controlled win rate: 8.03 -> 13.80Improved
fine-tune GRAM-R2 instead of vanilla generative RM on 1K STEM samplesGRAM-R2 · HelpSteer3 STEMaccuracy: +17.1 pointsImproved

Later papers that compared with this method

1 later accepted paper reports GRAM-r2 as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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