Papers with judge models
OffsetBias: Leveraging Debiased Data for Tuning Evaluators (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies have identified that judge models are vulnerable to certain biases, such as favoring longer answers regardless of content. |
| Approach: | They propose to use open-source LLMs to fine-tune judge models and propose a meta-evaluation collection of test cases for each bias type. |
| Outcome: | The proposed meta-evaluation dataset shows that fine-tuning improves the robustness of judge models against biases and improves performance across most evaluation scenarios. |
Direct Judgement Preference Optimization (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing judge models are largely trained with supervised finetuning on small data scales to perform limited types of evaluation tasks, limiting generalization. |
| Approach: | They propose to train judge models at large data scales with direct preference optimization . they use four training tasks to form three types of preference pairs targeting different aspects of evaluation . |
| Outcome: | The proposed model outperforms GPT-4o and other similar models on 13 benchmarks. |
Multi-Task Reinforcement Learning for Enhanced Multimodal LLM-as-a-Judge (2026.acl-industry)
Copied to clipboard
| Challenge: | Existing MLLMs are optimized for single-task scenarios and struggle to generalize to diverse contexts. |
| Approach: | They propose a framework that integrates multitask reinforcement learning and generalization capabilities of MLLMs to optimize the judge model across multiple tasks. |
| Outcome: | The proposed framework outperforms baseline models in judgment consistency and correlation with human preferences. |
GroUSE: A Benchmark to Evaluate Evaluators in Grounded Question Answering (2025.coling-main)
Copied to clipboard
| Challenge: | Existing automated RAG evaluation frameworks overlook important failure modes when using GPT-4 as a judge. |
| Approach: | They propose a novel pipeline to assess the calibration and discrimination capabilities of judge models by using a meta-evaluation benchmark of 144 unit tests to identify key failure modes. |
| Outcome: | The proposed pipeline improves on existing frameworks, while state-of-the-art open-source judges do not generalize to their proposed criteria. |
Does Context Matter? ContextualJudgeBench for Evaluating LLM-based Judges in Contextual Settings (2025.acl-long)
Copied to clipboard
| Challenge: | Contextual evaluation is challenging for state-of-the-art judge models . evaluation criteria are often conditional and dependent on practitioner priorities . |
| Approach: | They propose a judge benchmark that evaluates large language models as judges in contexts . they use human annotations and model-based perturbations to build the benchmark . |
| Outcome: | The proposed benchmark aims to evaluate large language models in contexts with 2,000 challenging response pairs. |
Improve LLM-as-a-Judge Ability as a General Ability (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies focus on generative judges, but only on their judge ability. |
| Approach: | They propose a method that leverages the generative and reasoning capabilities of large language models to evaluate LLM responses across diverse scenarios, providing accurate preference signals. |
| Outcome: | The proposed model performs on RewardBench with only 2% to 40% of the data required by other training frameworks. |
Triviality Corrected Endogenous Reward (2026.acl-long)
Copied to clipboard
Xinda Wang, Zhengxu Hou, Yangshijie Zhang, null Yanbingren, Jialin Liu, ChenZhuo Zhao, Zhibo Yang, Bin-Bin Yang, Feng Xiao
| Challenge: | Recent work on unsupervised reinforcement learning for mathematical reasoning using confidence-based endogenous rewards focuses on open-ended text generation, requiring either annotated data or powerful closed-source models. |
| Approach: | They propose a method that rewards the relative information gain between a specialist and a generalist reference policy, modulated by a probability-dependent correction mechanism. |
| Outcome: | The proposed model improves on multiple writing benchmarks and model architectures without external supervision and validates generality across different generation tasks. |
IF-RewardBench: Benchmarking Judge Models for Instruction-Following Evaluation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks for instruction-following lack data coverage and oversimplified pairwise evaluation paradigms that misalign with model optimization scenarios. |
| Approach: | They propose a meta-evaluation benchmark for instruction-following that covers diverse instruction and constraint types and a preference graph for each instruction. |
| Outcome: | Extensive experiments on IF-RewardBench show that the proposed benchmark achieves a stronger positive correlation with downstream task performance compared to existing benchmarks. |