| title | RUBRIC-ARROW: Alternating Pointwise Rubric Reward Modeling for LLM Post-training in Non-verifiable Domains |
|---|---|
| authors | Haoxiang Jiang, Zihan Dong, Tianci Liu, Wanying Wang, Ran Xu, Tony Yu, Linjun Zhang, Haoyu Wang |
| date | 2026-05-27 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2605.29156 |
| affiliation | OpenRubrics |
| upvotes | 9 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | Pointwise reward modeling offers critical signals for LLM post-training, yet struggles with absolute scoring in subjective, non-verifiable settings. Rubric-based methods address this by decomposing evaluation into explicit criteria, but existing approaches typically depend on frontier LLMs and suffer from ties caused by hard Boolean aggregation. We present RUBRIC-ARROW, an alternating framework that jointly trains a rubric generator and a rubric-conditioned judge, with its RL stage using only pairwise preference data. Our method couples a probability-based scoring rule that reduces ties with phase-specific preference-based rewards and an alternating GRPO scheme that together train the pointwise evaluator. Extensive experiments show that RUBRIC-ARROW achieves competitive reward-modeling accuracy and yields consistent gains for downstream policy post-training. |
cite
@misc{jiang2026rubric,
title={RUBRIC-ARROW: Alternating Pointwise Rubric Reward Modeling for LLM Post-training in Non-verifiable Domains},
author={Haoxiang Jiang and Zihan Dong and Tianci Liu and Wanying Wang and Ran Xu and Tony Yu and Linjun Zhang and Haoyu Wang},
year={2026},
eprint={2605.29156},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2605.29156}
}