CS Paper Radar

Poolwise: Evaluating Reward Models for Selection across Prompts

Mateusz Modrzejewski, Tomasz Radzikowski

cs.CL cs.AI ·submitted 2026-10-22 11:00 UTC · v2

Abstract

Reward models are used to select training data, but common benchmarks compare responses only within the same prompt. Selecting data across prompts also depends on how scores compare between prompts. We introduce Poolwise, a protocol that selects a fixed fraction of a response pool by score, measures the selected set's quality using external labels, and reports within-prompt ranking separately. In matched experiments on two rating sets and two model sizes, training each prompt's mean score to match its average response rating improves selection over permuted or omitted mean targets in every seed at 30 percent retention. In grouped HelpSteer experiments, the model with the highest within-prompt accuracy selects a lower-rated set than the correct-mean model. Subtracting prompt means estimated from separate reference responses preserves local rankings but lowers selected helpfulness for six open reward models and two closed judges on HelpSteer. Poolwise provides a direct way to compare reward models and score transformations for data selection.

Comment
38 pages, 9 figures, 47 tables. v2: reorganized around Poolwise, a protocol for evaluating reward models on pooled selection across prompts; adds matched training on HelpSteer's training split at two model sizes, two closed judges, a third reference generator and external panels (RoSE, IFEval); revised abstract, introduction and conclusion

Closest papers

Nearest neighbours by meaning across the whole index.

RankBuffer: Efficient Ranking-Based Rewards for Open-Ended Generation

cs.AI ·2026-09-29 ·Zixuan Yang, Yiqun Chen, Qi Liu, Wei Yang, Erhan Zhang +4 more
similarity 0.69
Abstract

Open-ended generation lacks canonical answers, making pointwise rewards difficult to calibrate for group-based reinforcement learning. Directly ranking same-query rollouts provides a more suitable relative reward signal, but existing ranking-based reward methods can incur substantial judging cost. We introduce RankBuffer, which maintains an ordered, query-specific buffer of previously judged responses as a reusable quality scale. Each rollout is first inserted into an anchor interval through an independent coarse judgment, after which only rollouts assigned to the same interval undergo local fine ranking. The resulting complete order is converted into bounded rank rewards, while boundary expansion, local refinement, and inactive-anchor pruning adapt the buffer as the policy evolves. Across four open-ended benchmarks, RankBuffer consistently outperforms all pointwise baselines. It also achieves nearly on-par performance with the strongest ranking-based reward baseline while substantially reducing judging cost. Ablations demonstrate the importance of both local fine ranking and anchor response content, while buffer analyses show that rollout-derived anchors progressively extend and refine the covered quality scale. These results establish response reuse as an effective approach to efficient relative reward construction.

0.20

Small Language Models as Judges for Rubric-Based Reinforcement Learning

cs.CL ·2026-08-30 ·Fengyu Xie, Yilun Zhao, Bingsen Chen, Arman Cohan, Chen Zhao
top venue similarity 0.66
Abstract · 9 pages, 1 figure; EMNLP 2026 Findings

Rubric-based reinforcement learning extends RL beyond tasks with exact answers or rule-based verifiers by scoring responses against instance-specific criteria. However, this makes reward computation expensive: training requires repeated rubric judging, often with proprietary APIs or local generative LLM judges with 7B parameters or more. We study whether smaller language models can serve as efficient and reliable rubric-based judges. To make this question measurable, we construct PointRubric and RaR-Science-Static, two pointwise rubric-based evaluation datasets with instance-specific criteria and itemwise satisfaction labels. We compare three ways of extracting criterion-level judgments from small models: Generative verdicts, Yes/No Logprob margins, and Probe judges. Across both datasets, the Qwen3-1.7B Probe judge achieves the strongest criterion-level agreement among these methods, outperforming Generative and Logprob judges. Used as a GRPO reward model, it trains a policy from 0.232 to 0.643 on RaR-Science rubric score, compared with 0.594 for an 8B Generative judge baseline, while the baseline requires 10.7$\times$ more reward-judge time. Task and domain transfer experiments further suggest that Probe judges preserve criterion-level reward structure across settings.

0.64

PerSPL: Personalized Semi-supervised Preference Learning

cs.LG ·2026-10-21 ·Xiaoan Xu, Tingzhou Wei, Juncheng Dong, Korel Gundem, Ethan X. Fang +1 more
similarity 0.66
Abstract · A coauthor has explicitly requested withdrawal due to unresolved concerns about the manuscript. In particular, Section 4 does not adequately…

Personalized reward models require user-specific preference comparisons, but collecting such feedback is costly and provides only limited supervision for each user. At the same time, large pools of responses can be generated without user identities or preference labels, yet existing personalized reward models leave this data unused. Directly learning from these responses is difficult because pairwise objectives constrain relative scores within a prompt, whereas pseudo-labeling individual responses requires scores that are comparable across prompts for the same user. We propose \textbf{PerSPL} (Personalized Semi-supervised Preference Learning), a backbone-agnostic framework that uses pair-mean score centering to stabilize this cross-prompt score coordinate and iteratively turns unlabeled responses into user-specific training signals. Against seven representative baselines across four benchmarks---PersonalLLM, HelpSteer2, Reddit TLDR, and Chatbot Arena---PerSPL achieves the highest reported mean accuracy among the evaluated methods for both seen and unseen users. The two latter benchmarks provide real-user preference comparisons in summarization and open-domain chatbot interaction, respectively. These results indicate that unlabeled responses can provide additional supervision for personalized reward modeling when per-user preference labels are scarce.

0.13

SESSE: Sketch, Expand, Sort, Summarize, Evaluate -- LLM-as-Judge Evaluation via Structured Decomposition

cs.AI cs.LG ·2026-08-18 ·Dae Lee, Mihai Delgeanu, Adel Youssef
similarity 0.64
Abstract

LLM-as-judge evaluation reduces response quality assessment to a single holistic A/B preference choice, providing no mechanism to isolate which quality dimensions drove the preference or distinguish model errors from genuine label ambiguity. We propose SESSE (Sketch, Expand, Sort, Summarize, Evaluate), a training-free framework that decomposes holistic judgment into structured sub-questions mined directly from the judge's own error cases; requiring no oracle responses, task-specific rubrics, or fine-tuning. On RewardBench (n=1,000), SESSE achieves near-parity with the chain-of-thought baseline and is competitive with RISE-Judge-32B (92.7%), a fine-tuned specialist, while remaining fully training-free. Per-criterion vote evidence provides an interpretable audit trail for diagnosing label ambiguity and judge failure modes unavailable from a single holistic output token.

0.18

DeepResearch RewardBench: Evaluating Reward Models for Deep Research Report

cs.AI cs.CL ·2026-11-06 ·Shudong Liu, Yuting Zhong, Guanhua Chen, Xinyi Yang, Junnan Liu +3 more
top venuebenchmark similarity 0.64
Abstract · Accepted by EMNLP 2026 Findings

Empowered by Large Language Models (LLMs), deep research systems integrate online information into research reports through multi-step web searching and information synthesis. Evaluating these reports presents a significant challenge due to their long context, high information density, and domain specificity, necessitating the use of AI evaluation as a substitute for prohibitively expensive human assessment. However, few studies have examined the effectiveness of existing reward models or LLMs in understanding and evaluating such research reports. To bridge this gap, we introduce DeepResearch RewardBench, the first benchmark dedicated to evaluating reward modeling for research reports. We collect diverse report candidates from multiple agents or tool-augmented LLMs. These candidates are scored using query-specific golden rubrics, and we ensure data quality and difficulty through a combination of multi-round filtering and human verification. Our benchmark comprises preference datasets formatted as pairs, triplets, and quadruplets to test ranking capabilities across varying levels of difficulty. Extensive experiments demonstrate that while current reward models handle general coherence effectively, their performance significantly degrades when evaluating research reports. Generative LLMs acting as judges demonstrate comprehensive superiority in this domain, with comparative evaluation proving more effective than direct scoring methods. We anticipate that this benchmark will serve as a vital testbed for the evaluation and advancement of research agents.

0.24

Human Evaluation with Automatic Scores: Separating Sampling and Estimation Gains

cs.CL ·2026-10-28 ·Yang Liu, Ran Zou, Zhengtao Gui, Chao Xue
similarity 0.64
Abstract

Automatic judges offer inexpensive information about model outputs, but an evaluation team must decide how to turn that information into fewer human annotations. Scores can guide sampling, correct an estimate, or support adaptive acquisition, and these uses may capture overlapping gains. We compare their incremental value on fixed evaluation pools from three preference datasets, and 19 attributed-question-answering systems. Matched comparisons use the same scores, account for pilot labels and budget rules, and separate acquisition from estimation. Score stratification is a strong reference: after ten-bin stratification, the best fixed local linear correction offers only 1.00--5.27% further MSE reduction in the primary configurations. Unequal allocation offers larger gains in some pools, but can increase error when confident scores conceal variable human outcomes. Comparison on AttributedQA, shrinking learned quotas toward score-only quotas lowers pooled MSE for 17 of 19 systems relative to the score-only procedure, but ordinary pooled 95% intervals cover only 88.09--93.54%. The same shrinkage improves on the score-only reference in only one of eight preference configurations. Ranking experiments further show that lower individual MSE need not improve system selection. These findings support a simple starting procedure, proportional sampling within score strata followed by a stratified human mean, and identify what additional evidence is needed to justify correction or adaptation for the intended evaluation output.

0.26

Equal Ranking Quality, Different Decisions: Measuring and Reducing Order Dependence in LLM Scorers

cs.CL cs.IRcs.LG ·2026-08-27 ·Markus Frohmann, Mahdiyar Ali Akbar Alavi, Elizabeth Lingg, Navid Rekabsaz
code/datahot direction similarity 0.64
Abstract · 9 pages main text, 45 pages total

In passage reranking, response ranking and multi-document question answering, LLMs can score several candidate documents or responses together in one prompt, each still receiving its own score. Such scorers are selected on ranking quality, but their scores determine a decision: what a score threshold retains, a reader answers, or which chosen/rejected pair enters preference training. Because the candidates share that prompt, reordering them changes their scores. The same query over the same candidates should still yield the same decision. However, equal ranking quality does not imply equal decisions: on passage reranking, five trained scorers within 0.010 nDCG@10 retain sets that overlap by only 0.66-0.84 when reordered. No prompt-time change we test resolves that dependence: the only one that improves ranking quality does not measurably improve decision stability. We introduce order-consistency SFT (OC-SFT), which attenuates it in the weights by penalizing disagreement between a candidate's scores across orderings. It holds ranking quality and leads every decision-stability measure among trained scorers on all three tasks. It is also more stable on 12 base models than order-averaged distillation, which trains on labels averaged across permutations. One OC-SFT permutation retains sets that overlap more than ten averaged off-the-shelf permutations. A comparison of such scorers should therefore report what a threshold retains and a reader answers, not ranking quality alone. Code is available at https://github.com/thomsonreuters/presentation-dependence.

0.68

Scaling Laws for Generative Reward Models

cs.AI ·2026-10-22 ·Angel Raychev, Nathan Lile, Rafael Rafailov
code/data similarity 0.63
Abstract · 28 pages. Code and research artifacts: https://github.com/RaychevAngel/generative-reward-models

We study generative reward models (GenRMs) as online preference providers for policy optimization in creative writing. Across Qwen3 reward-model sizes from 0.6B to 14B, Thinking judges trained with GRPO have higher static Gold-label accuracy than Answer-Only judges trained with SFT. This ranking often reverses when the judges train policies through online DPO: Answer-Only rewarders yield higher peak and top-five-average policy Elo in seven of eight matched settings under a fixed, learned 32B Gold evaluator. Intermediate Thinking-judge checkpoints can also outperform later, more statically accurate checkpoints as rewarders. Sweeps over rewarder size, policy size, training duration and DPO regularization show substantial benefits from reward-model capacity, including judges larger than the policy. We analyze trajectories with evaluator-specific Elo arenas, conditional match-resampling intervals and descriptive scaling fits. These results demonstrate an evaluator-rewarder gap in the studied pipelines: fixed-distribution judging accuracy does not determine downstream optimization quality. The comparison changes judge-training objective as well as response format, and the main ranking is conditional on its learned reference rather than a universal human-quality result. A limited follow-up using a different learned judge is inconclusive overall.

0.52