# UMM-RM: An Upcycle-and-Merge MoE Reward Model for Mitigating Reward Hacking

arXiv: https://arxiv.org/abs/2512.00724 · HTML: https://mlsanity.com/paper/2512.00724/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reward modeling and RLHF, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 31 of its 32 references that are in the map.
- First posted on arXiv 30 Nov 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)
- Mixture-of-experts / [MoE in reinforcement learning and robotics](https://mlsanity.com/moe/moe-in-reinforcement-learning-and-robotics/)

## Builds on (most cited first, 30 of 31)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling](https://mlsanity.com/paper/2304.01373/) (2023)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback](https://mlsanity.com/paper/2305.14387/) (2023)
- [TinyLlama: An Open-Source Small Language Model](https://mlsanity.com/paper/2401.02385/) (2024)
- [LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion](https://mlsanity.com/paper/2306.02561/) (2023)
- [Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts](https://mlsanity.com/paper/2406.12845/) (2024)
- [Reward Model Ensembles Help Mitigate Overoptimization](https://mlsanity.com/paper/2310.02743/) (2023)
- [Sparse Upcycling: Training Mixture-of-Experts from Dense Checkpoints](https://mlsanity.com/paper/2212.05055/) (2022)
- [Helping or Herding? Reward Model Ensembles Mitigate but do not Eliminate Reward Hacking](https://mlsanity.com/paper/2312.09244/) (2023)
- [MoEfication: Transformer Feed-forward Layers are Mixtures of Experts](https://mlsanity.com/paper/2110.01786/) (2021)
- [WARM: On the Benefits of Weight Averaged Reward Models](https://mlsanity.com/paper/2401.12187/) (2024)
- [Regularizing Hidden States Enables Learning Generalizable Reward Model for LLMs](https://mlsanity.com/paper/2406.10216/) (2024)
- [InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling](https://mlsanity.com/paper/2402.09345/) (2024)
- [Improving Reinforcement Learning from Human Feedback with Efficient Reward Model Ensemble](https://mlsanity.com/paper/2401.16635/) (2024)
- [One Student Knows All Experts Know: From Sparse to Dense](https://mlsanity.com/paper/2201.10890/) (2022)
- [DMoERM: Recipes of Mixture-of-Experts for Effective Reward Modeling](https://mlsanity.com/paper/2403.01197/) (2024)
- [Fine-Tuning Language Models with Reward Learning on Policy](https://mlsanity.com/paper/2403.19279/) (2024)
- [Scalable Ensembling For Mitigating Reward Overoptimisation](https://mlsanity.com/paper/2406.01013/) (2024)
- [Towards Understanding the Influence of Reward Margin on Preference Model Performance](https://mlsanity.com/paper/2404.04932/) (2024)
- [XFT: Unlocking the Power of Code Instruction Tuning by Simply Merging Upcycled Mixture-of-Experts](https://mlsanity.com/paper/2404.15247/) (2024)
- [Generalizing Reward Modeling for Out-of-Distribution Preference Learning](https://mlsanity.com/paper/2402.14760/) (2024)

## Abstract (arXiv)

Reward models (RMs) are critical to reinforcement learning from human feedback (RLHF). However, policy models can exploit biases or spurious correlations in dense RMs, causing reward hacking: RM scores rise while alignment with human preferences deteriorates, especially under distribution shift. We propose UMM-RM (Upcycle-and-Merge MoE Reward Model), which copies the feed-forward layers of an SFT checkpoint to initialize shared and routed experts, then trains the model on preference data. The always-active shared expert captures instruction-agnostic preference signals, while routed experts model fine-grained preferences across instructions or tasks. The experts are then merged into a single dense RM using learnable weights, preserving the robustness and exploitation resistance of expert diversity without MoE or ensemble inference overhead. Across multiple base models and preference datasets, UMM-RM improves preference accuracy, reduces reward hacking during PPO training, and achieves more stable preference alignment than standard dense RMs.