Research map / Vision-language and multimodal / Multimodal LLMs
Omni-RRM: Advancing Omni Reward Modeling via Automatic Rubric-Grounded Preference Synthesis
Zicheng Kong, Dehua Ma, Zhenbo Xu, Anwen Yang, Yiwei Ru, Haoran Wang, Zixuan Zhou, Fuqing Bie, Liuyu Xiang, Huijia Wu, Jian Zhao, Zhaofeng He
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add GRPO after SFT (ablation) | Omni-RRM-7B · Overall (5 benchmarks) | preference accuracy (%): 62.6 -> 70.4 | Improved |
| add GRPO after SFT (ablation) | Omni-RRM-7B · MM-RewardBench | preference accuracy (%): 61.0 -> 72.9 | Improved |
| add GRPO after SFT (ablation) | Omni-RRM-7B · ShareGPT-Video | preference accuracy (%): 70.5 -> 80.2 | Improved |
| add rubric rationales during SFT+RL (ablation) | Omni-RRM-7B · Overall (4-benchmark ablation suite) | preference accuracy (%): 64.4 -> 71.8 | Improved |
| add rubric reward term in GRPO (ablation) | Omni-RRM-7B · Overall (4-benchmark ablation suite) | preference accuracy (%): 68.1 -> 71.8 | Improved |
| add rubric rationales during SFT+RL (ablation) | Omni-RRM-3B · Overall (4-benchmark ablation suite) | preference accuracy (%): 58.9 -> 65.0 | Improved |
| add rubric reward term in GRPO (ablation) | Omni-RRM-3B · Overall (4-benchmark ablation suite) | preference accuracy (%): 62.9 -> 65.0 | Improved |
| train Omni-RRM on omni-modal data instead of audio only (ablation) | Omni-RRM-3B · Audio-HH-RLHF | preference accuracy (%): 63.8 -> 65.1 | Improved |
| use Omni-RRM-7B as Best-of-N selector instead of self-consistency | Qwen2.5-Omni-7B · MMMU | accuracy (%): 54.5 -> 56.4 | Improved |
| use Omni-RRM-7B as Best-of-N selector instead of self-consistency | Qwen2.5-Omni-7B · Video-MME | accuracy (%): 52.4 -> 54.1 | Improved |
| use Omni-RRM-7B as Best-of-N selector instead of self-consistency | Qwen2.5-Omni-7B · AVQA | accuracy (%): 77.7 -> 80.3 | Improved |
| apply Omni-RRM-7B post-training to Qwen2.5-Omni-7B backbone | Qwen2.5-Omni-7B · Overall (5 benchmarks) | preference accuracy (%): 60.2 -> 70.4 | Improved |
Closest papers by meaning
- Omni-Reward: Towards Generalist Omni-Modal Reward Modeling with Free-Form Preferences (ICLR 2026 · similarity 0.93)
- BaseReward: A Strong Baseline for Multimodal Reward Model (ICLR 2026 · similarity 0.90)
- RoVRM: A Robust Visual Reward Model Optimized via Auxiliary Textual Preference Data (AAAI 2025 · similarity 0.89)
- Auto-Rubric as Reward: From Implicit Preference to Explicit Generative Criteria (NeurIPS 2026 · similarity 0.89)
- Think, then Score: Decoupled Reasoning and Scoring for Video Reward Modeling (NeurIPS 2026 · similarity 0.88)
- Omni-R1: Reinforcement Learning for Omnimodal Reasoning via Two-System Collaboration (NeurIPS 2025 · similarity 0.88)
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.
