Papers with RMs
M-RewardBench: Evaluating Reward Models in Multilingual Settings (2025.acl-long)
Copied to clipboard
Srishti Gureja, Lester James Validad Miranda, Shayekh Bin Islam, Rishabh Maheshwary, Drishti Sharma, Gusti Triandi Winata, Nathan Lambert, Sebastian Ruder, Sara Hooker, Marzieh Fadaee
| Challenge: | Reward models (RMs) are primarily trained and evaluated in English and their capabilities in multilingual settings remain understudied. |
| Approach: | They construct a multilingual RM evaluation benchmark that tests the chat, safety, reasoning, and translation capabilities of RMs in 23 languages. |
| Outcome: | The proposed model performs better for high-resource languages and improves with translation quality. |
A Systematic Analysis of Base Model Choice for Reward Modeling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Reinforcement learning from human feedback (RLHF) and reward modeling are key to training powerful large language models (LLMs). |
| Approach: | They propose to combine RLHF and reward modeling to boost model selection . they also demonstrate that a small set of benchmarks could be combined to boost the model selection. |
| Outcome: | The results show that the model selection can be improved by up to 14% compared to the most common (default) choice. |
Cross-lingual Transfer of Reward Models in Multilingual Alignment (2025.naacl-short)
Copied to clipboard
| Challenge: | Recent studies in reward modeling schemes are skewed towards English, limiting the applicability of RLHF in multilingual alignments. |
| Approach: | They investigate cross-lingual transfer of English RMs by representation shifts . they also analyze cross-linguistic transfer of RM through the representation shift . |
| Outcome: | The results show that English RMs can be transferred across languages by 34% . |
Structural Reward Model: Enhancing Interpretability, Efficiency, and Scalability in Reward Modeling (2025.emnlp-industry)
Copied to clipboard
Xiaoyu Liu, Di Liang, Hongyu Shan, Peiyang Liu, Yonghao Liu, Muling Wu, Yuntao Li, Xianjie Wu, Li Miao, Jiangrong Shen, Minlong Peng
| Challenge: | Generative RMs (GRMs) lack contextual and background information during inference, leading to incomplete evaluations. |
| Approach: | They propose a modular and interpretable framework that integrates side-branch models as auxiliary feature generators. |
| Outcome: | The proposed framework outperforms scalar and saline reward models in robustness and alignment with human preferences. |
EQA-RM: A Generative Embodied Reward Model with Test-time Scaling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing generic Reward Models are ill-equipped for dynamic and interactive domains. |
| Approach: | They propose a novel generative multimodal reward model specifically architected for EQA that provides interpretable, structured reward feedback. |
| Outcome: | The proposed model outperforms proprietary benchmarks, including Gemini-2.5-Flash, GPT-4o, Claude-3.5-Haiku, and open-sourced state-of-the-art models such as RoVRM and VisualPRM. |
RecStream: Graph-aware Stream Management for Concurrent Recommendation Model Online Serving (2025.coling-industry)
Copied to clipboard
| Challenge: | Existing systems that use recommendation models perform poorly under highly concurrent scenarios. |
| Approach: | They propose a system that optimizes stream configurations based on model characteristics and concurrency levels. |
| Outcome: | The proposed system outperforms existing methods under high concurrency scenarios. |
Transferring Textual Preferences to Vision-Language Understanding through Model Merging (2025.acl-short)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) perform outstandingly across multimodal tasks, but training them with preference data is computationally expensive. |
| Approach: | They propose to merge text-based reward models with LVLMs to create visionlanguage reward models (VLRMs) this approach offers an efficient method for incorporating textual preferences into LVRMs. |
| Outcome: | The proposed model improves over LVLMs’ scoring and text-based RMs, and offers an efficient method for incorporating textual preferences into LVRMs. |
RewardBench: Evaluating Reward Models for Language Modeling (2025.findings-naacl)
Copied to clipboard
Nathan Lambert, Valentina Pyatkin, Jacob Morrison, Lester James Validad Miranda, Bill Yuchen Lin, Khyathi Chandu, Nouha Dziri, Sachin Kumar, Tom Zick, Yejin Choi, Noah A. Smith, Hannaneh Hajishirzi
| Challenge: | Evaluating reward models presents an opportunity to understand the opaque technologies used for alignment of language models. |
| Approach: | They present a benchmark dataset and code-base for evaluation of reward models . they use prompt-chosen-rejected trios to benchmark how they perform on queries . |
| Outcome: | The proposed dataset compares RMs with other models on a set of questions. |
Test-Time Scaling of Reasoning Models for Machine Translation (2026.eacl-long)
Copied to clipboard
| Challenge: | Using TTS, Reasoning Models (RMs) are able to perform tasks such as math and coding with limited results. |
| Approach: | They evaluate 12 Reasoning Models across a diverse suite of MT benchmarks, examining three scenarios: direct translation, forced-reasoning extrapolation, and post-editing. |
| Outcome: | The proposed approach improves translation quality on three domains, with inconsistent results for general-purpose RMs and performance plateauing. |
HARM: Learning Hate-Aware Reward Model for Evaluating Natural Language Explanations of Offensive Content (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing reward models for explaining hate speech are optimized for broad notions of safety, but they assign lower scores to contextually rich explanations. |
| Approach: | They propose a reward model that integrates interpretable signals to better align reward scores with the needs of hate speech explanation. |
| Outcome: | The proposed model outperforms general-purpose baselines and improves pair-wise preference. |
Improving Reward Models with Synthetic Critiques (2025.findings-naacl)
Copied to clipboard
| Challenge: | a recent study shows that reward models overfit on superficial features, hindering generalization performance . prevailing approach to training preference-based reward models presents several challenges . |
| Approach: | They propose a method that uses synthetic natural language critiques to provide additional feedback to large language models. |
| Outcome: | The proposed approach improves performance and data efficiency of RMs initialized from different pretrained models, reducing the reliance on costly human annotations. |
SSR-Zero: Simple Self-Rewarding Reinforcement Learning for Machine Translation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated remarkable capabilities in machine translation, but most MT-specific LLMs rely heavily on external supervision during training. |
| Approach: | They propose a reinforcement learning framework for machine translation that is reference-free and relies solely on self-judging rewards. |
| Outcome: | The proposed framework outperforms existing LLMs and larger general LLM models on English Chinese translation benchmarks and performs competitively with leading closed-source systems. |
InternLM-XComposer2.5-Reward: A Simple Yet Effective Multi-Modal Reward Model (2025.findings-acl)
Copied to clipboard
Yuhang Zang, Xiaoyi Dong, Pan Zhang, Yuhang Cao, Ziyu Liu, Shengyuan Ding, Shenxi Wu, Yubo Ma, Haodong Duan, Wenwei Zhang, Kai Chen, Dahua Lin, Jiaqi Wang
| Challenge: | Despite the promising performance of Large Vision Language Models, they sometimes generate incorrect outputs. |
| Approach: | They propose a multi-modal reward model that aligns LVLMs with human preferences. |
| Outcome: | The proposed model achieves excellent results on the latest multi-modal reward model benchmark and shows competitive performance on text-only reward model. |
RAGferee: Building Contextual Reward Models for Retrieval-Augmented Generation (2025.emnlp-main)
Copied to clipboard
Andrei Catalin Coman, Ionut Teodor Sorodoc, Leonardo F. R. Ribeiro, Bill Byrne, James Henderson, Adrià de Gispert
| Challenge: | Existing Reward Models (RMs) struggle in Retrieval Augmented Generation settings. |
| Approach: | They propose a method that repurposes question-answering datasets into preference pairs that prioritise groundedness over stylistic features. |
| Outcome: | The proposed method surpasses existing RMs trained on larger general corpora with an absolute improvement of +15.5%. |
Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling (2026.acl-long)
Copied to clipboard
| Challenge: | Existing failure discovery methods rely on prior knowledge of preference attributes . Existing methods do not scale to new models or data. |
| Approach: | They propose a preference distribution agnostic procedure that uses the reward model itself to guide controlled decoding toward mis specified responses while preserving the underlying preference class. |
| Outcome: | The proposed procedure improves robustness without degrading reward quality across models. |
ToolRM: Towards Agentic Tool-Use Reward Modeling (2026.findings-acl)
Copied to clipboard
Renhao Li, Jianhong Tu, Yang Su, Yantao Liu, Fei Huang, Hamid Alinejad-Rokny, Derek F. Wong, Junyang Lin, Min Yang
| Challenge: | lack of reliable reward models for tool-use tasks has limited progress toward agentic AI . recent advances in agentic artificial intelligence are driven by tool-using capabilities of large language models. |
| Approach: | They propose a pipeline that constructs pairwise preference data using rule-based scoring and multidimensional sampling to build lightweight reward models. |
| Outcome: | The proposed model outperforms existing models on tool calling tasks with higher accuracy. |
Mind the (DH) Gap! A Contrast in Risky Choices Between Reasoning and Conversational LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models are increasingly used in decision support systems and workflows . traditional computational paradigms for decision-making under uncertainty choose an option that maximizes expected utility or payoff . |
| Approach: | They compare large language models as decision support systems and agentic workflows . they find that LLMs cluster into reasoning models and conversational models . |
| Outcome: | The proposed models differ in their ability to perform tasks and their ability in a human-like way. |
Synergistic Interplay between Search and Large Language Models for Information Retrieval (2024.acl-long)
Copied to clipboard
| Challenge: | Information retrieval (IR) is an indispensable technique for locating relevant resources from vast amounts of data. |
| Approach: | They propose a framework that facilitates information refinement through synergy between RMs and LLMs. |
| Outcome: | The proposed framework improves the performance of large-scale retrieval benchmarks on web searches and low-resource retrieval tasks. |
Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Reinforcement learning from human feedback (RLHF) is the primary method for aligning large language models with human preferences. |
| Approach: | They propose to train an Absolute-Rating Multi-Objective Reward Model with multi-dimensional absolute-rating data. |
| Outcome: | The proposed model outperforms the LLM-as-a-judge method on RewardBench . it achieves state-of-the-art performance on the benchmark . |
Rethinking Reward Model Evaluation Through the Lens of Reward Overoptimization (2025.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks for reward models show a weak correlation with performance of optimized policies . existing benchmarks do not accurately assess the true capabilities of reward models . |
| Approach: | They explore how reward overoptimization captures how well a reward model aligns with human preferences and the dynamics of the learning signal it provides to the policy. |
| Outcome: | The proposed benchmarks show that reward overoptimization is a weak factor . the high correlation with degree of overoptimalization leads to lower correlation with downstream performance . |
Axiomatic Preference Modeling for Longform Question Answering (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large language models have helped bridge the "alignment gap" between the responses of raw pretrained language models and responses that resonate more closely with human preferences. |
| Approach: | They propose to use a axiomatic framework to generate a rich variety of preference signals to uphold these signals. |
| Outcome: | The proposed model outperforms GPT-4 and ChatGPT in preference scoring. |
Cheems: A Practical Guidance for Building and Evaluating Chinese Reward Models from Scratch (2025.acl-long)
Copied to clipboard
Xueru Wen, Jie Lou, Zichao Li, Yaojie Lu, XingYu XingYu, Yuqiu Ji, Guohai Xu, Hongyu Lin, Ben He, Xianpei Han, Le Sun, Debing Zhang
| Challenge: | Existing Chinese resources are small in scale and limited to specific domains, making them insufficient for LLM post-training. |
| Approach: | They propose a Chinese-annotated reward model and a preference dataset to address this gap . they evaluate Chinese RMs on CheemsBench and construct an RM that captures human preferences . |
| Outcome: | The proposed RM achieves state-of-the-art performance on CheemsBench and CheeMePreference. |
Reward Model Perspectives: Whose Opinions Do Reward Models Reward? (2025.emnlp-main)
Copied to clipboard
| Challenge: | a recent study shows that reward models are poorly aligned with demographic groups and can reward harmful stereotypes. |
| Approach: | They propose a framework for measuring the alignment of opinions captured by RMs . they also investigate the extent to which RM's demonstrate sociodemographic biases a . |
| Outcome: | The proposed framework measures the alignment of opinions captured by RMs . it shows that RM models are poorly aligned with several demographic groups . the findings highlight the need for more careful consideration of RM behavior in model alignment . |
Agentic Reward Modeling: Integrating Human Preferences with Verifiable Correctness Signals for Reliable Reward Systems (2025.acl-long)
Copied to clipboard
| Challenge: | Existing reward models focus on human preferences, neglecting verifiable correctness signals. |
| Approach: | They propose a reward system that combines human preference rewards with verifiable correctness signals to provide reliable rewards. |
| Outcome: | The proposed reward agent significantly outperforms vanilla reward models on benchmarks and inference-time best-of-n searches on real-world tasks. |
WildReward: Learning Reward Models from In-the-Wild Human Interactions (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work focused on collecting preference pairs, requiring substantial annotation efforts. |
| Approach: | They propose a pipeline to extract reliable human feedback from in-the-wild interactions . they propose to use WildChat as an interaction source to train the model . |
| Outcome: | The proposed model achieves comparable or even superior performance compared to conventional models with improved calibration and cross-sample consistency. |
RAG-RewardBench: Benchmarking Reward Models in Retrieval Augmented Generation for Preference Alignment (2025.findings-acl)
Copied to clipboard
Zhuoran Jin, Hongbang Yuan, Tianyi Men, Pengfei Cao, Yubo Chen, Jiexin Xu, Huaijun Li, Xiaojian Jiang, Kang Liu, Jun Zhao
| Challenge: | Existing retrieval augmented language models often overlook effective alignment with human preferences. |
| Approach: | They propose a benchmark to evaluate RMs in retrieval augmented language models . they incorporate 18 RAG subsets, six retrievers, and 24 RALMs to increase diversity . |
| Outcome: | The proposed benchmark combines 18 RAG subsets, six retrievers, and 24 RALMs to increase diversity of data sources. |
PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modeling (2026.acl-long)
Copied to clipboard
| Challenge: | Existing reward models lack generative and reasoning capabilities, resulting in poor performance. |
| Approach: | They propose a reward-aware task-adaptive reward model that enables pointwise training using readily available pairwise data via a novel Preference-Aware Reward mechanism. |
| Outcome: | The proposed reward model achieves an average relative improvement of 8.7% over the base models on RewardBench and RMBench. |
From Outcomes to Processes: Guiding PRM Learning from ORM for Inference-Time Alignment (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to align large language models with human preferences suffer from inconsistent scoring and suboptimal alignment. |
| Approach: | They propose a dual-consistency framework that aligns partial sequences with human preferences. |
| Outcome: | The proposed framework significantly reduces granularity discrepancies and improves GPT-4 evaluation scores. |
Debiasing Reward Models via Causally Motivated Inference-Time Intervention (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches for mitigating spurious features in RMs focus on response length . Existing methods focus on RM activation, resulting in performance trade-offs . |
| Approach: | They propose a method that uses neurons to suppress spurious features in RMs at inference time. |
| Outcome: | The proposed method reduces sensitivity to spurious features without inducing performance trade-offs on RM benchmarks. |
Aligning Agents via Planning: A Benchmark for Trajectory-Level Reward Modeling (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) evolve into agentic systems capable of autonomous tool invocation and complex reasoning. |
| Approach: | They propose a trajectory-level preference benchmark to evaluate judges' ability to distinguish preferred versus distractor agent trajectories in tool-integrated environments. |
| Outcome: | The proposed benchmark evaluates how well judges distinguish preferred versus distractor agent trajectories in complex tool-using scenarios. |
Removing Prompt-template Bias in Reinforcement Learning from Human Feedback (2025.findings-acl)
Copied to clipboard
| Challenge: | Reinforcement Learning from Human Feedback (RLHF) has shown promise for enhancing pre-trained large language models to generate responses that align with human preferences and societal values. |
| Approach: | They propose a method to estimate prompt-template bias term during reward modeling and use it to calibrate reward scores. |
| Outcome: | The proposed method can be flexibly combined with existing algorithms of removing length bias, leading to a further improvement in the aspect of enhancing the quality of generated responses. |
ReflectRM: Boosting Generative Reward Models via Self-Reflection within a Unified Judgment Framework (2026.acl-long)
Copied to clipboard
Kai Qin, Liangxin Liu, Yu Liang, Longzheng Wang, null Wangyan, Zhang Yueyang, Long Xia, Zhiyuan Sun, Houde Liu, Daiting Shi
| Challenge: | Existing methods for generating reward models focus on outcome-level supervision, neglecting analytical process quality, which constrains their potential. |
| Approach: | They propose a novel reward model that leverages self-reflection to assess analytical quality and enhance preference modeling. |
| Outcome: | The proposed model improves performance on four benchmarks and significantly mitigates positional bias. |
DARM: Distribution-Aware Reward Modeling by Alleviating Biases from Low Preference-Context Dependency Data (2026.acl-long)
Copied to clipboard
Shaofan Liu, Guoqiang Zhang, Shihan Dou, Huiyuan Zheng, Yiming Zhou, Junjie Ye, Shaowen Wang, Shichun Liu, Jiazheng Zhang, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Existing methods for training reward models are vulnerable to context neglect and degraded accuracy. |
| Approach: | They propose distribution-aware reward modeling that augments the RM objective with a conditional mutual information regularizer that maximizes context and the predicted reward conditioned on the response. |
| Outcome: | The proposed model improves performance in RLHF and improves accuracy in other settings. |