On Designing Effective RL Reward at Training Time for LLM Reasoning

• URL: http://arxiv.org/abs/2410.15115v3
• Date: Wed, 27 Nov 2024 11:58:50 GMT
• Title: On Designing Effective RL Reward at Training Time for LLM Reasoning
• Authors: Jiaxuan Gao, Shusheng Xu, Wenjie Ye, Weilin Liu, Chuyi He, Wei Fu, Zhiyu Mei, Guangju Wang, Yi Wu,
• Abstract summary: We evaluate popular reward models for RL training, including the Outcome-supervised Reward Model (ORM) and the Process-supervised Reward Model (PRM) Surprisingly, even though these learned reward models have strong inference-time performances, they may NOT help or even hurt RL training. We introduce two novel reward refinement techniques, including Clipping and Delta.
• Score: 14.006845442313134
• License:
• Abstract: Reward models have been increasingly critical for improving the reasoning capability of LLMs. Existing research has shown that a well-trained reward model can substantially improve model performances at inference time via search. However, the potential of reward models during RL training time still remains largely under-explored. It is currently unclear whether these reward models can provide additional training signals to enhance the reasoning capabilities of LLMs in RL training that uses sparse success rewards, which verify the correctness of solutions. In this work, we evaluate popular reward models for RL training, including the Outcome-supervised Reward Model (ORM) and the Process-supervised Reward Model (PRM), and train a collection of LLMs for math problems using RL by combining these learned rewards with success rewards. Surprisingly, even though these learned reward models have strong inference-time performances, they may NOT help or even hurt RL training, producing worse performances than LLMs trained with the success reward only. Our analysis reveals that an LLM can receive high rewards from some of these reward models by repeating correct but unnecessary reasoning steps, leading to a severe reward hacking issue. Therefore, we introduce two novel reward refinement techniques, including Clipping and Delta. The key idea is to ensure the accumulative reward of any reasoning trajectory is upper-bounded to keep a learned reward model effective without being exploited. We evaluate our techniques with multiple reward models over a set of 1.5B and 7B LLMs on MATH and GSM8K benchmarks and demonstrate that with a carefully designed reward function, RL training without any additional supervised tuning can improve all the evaluated LLMs, including the state-of-the-art 7B LLM Qwen2.5-Math-7B-Instruct on MATH and GSM8K benchmarks.

Related papers

• On the Emergence of Thinking in LLMs I: Searching for the Right Intuition [34.32871896067864]
  We propose a post-training framework called Reinforcement Learning via Self-Play (RLSP) RLSP involves three steps: supervised fine-tuning with human or synthetic demonstrations of the reasoning process, using an exploration reward signal to encourage diverse and efficient reasoning behaviors, and RL training with an outcome verifier to ensure correctness while preventing reward hacking. Empirical studies in the math domain show that RLSP improves reasoning.
  arXiv  Detail & Related papers  (2025-02-10T18:52:04Z)
• Reusing Embeddings: Reproducible Reward Model Research in Large Language Model Alignment without GPUs [58.18140409409302]
  Large Language Models (LLMs) have made substantial strides in structured tasks through Reinforcement Learning (RL) Applying RL in broader domains like chatbots and content generation presents unique challenges. We show a case study of reproducing existing reward model ensemble research using embedding-based reward models.
  arXiv  Detail & Related papers  (2025-02-04T19:37:35Z)
• Process Reinforcement through Implicit Rewards [95.7442934212076]
  Dense process rewards have proven a more effective alternative to the sparse outcome-level rewards in the inference-time scaling of large language models (LLMs) Dense rewards also offer an appealing choice for the reinforcement learning (RL) of LLMs since their fine-grained rewards have the potential to address some inherent issues of outcome rewards. This can be primarily attributed to the challenges of training process reward models (PRMs) online, where collecting high-quality process labels is prohibitively expensive. We propose PRIME, which enables online PRM updates using only policy rollouts and outcome labels through implict process rewards
  arXiv  Detail & Related papers  (2025-02-03T15:43:48Z)
• RewardBench: Evaluating Reward Models for Language Modeling [100.28366840977966]
  We present RewardBench, a benchmark dataset and code-base for evaluation of reward models. The dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods.
  arXiv  Detail & Related papers  (2024-03-20T17:49:54Z)
• Bayesian Reward Models for LLM Alignment [26.612181012468167]
  We train a Bayesian reward model, which signals higher uncertainty further from the training data distribution. We find that the resulting uncertainty estimates can effectively mitigate reward overoptimization in BoN sampling.
  arXiv  Detail & Related papers  (2024-02-20T18:20:59Z)
• Self-Rewarding Language Models [105.6830788170348]
  We study Self-Rewarding Language Models, where the language model itself is used via LLM-as-a-Judge prompting to provide its own rewards during training. We show that during Iterative DPO training that not only does instruction following ability improve, but also the ability to provide high-quality rewards to itself.
  arXiv  Detail & Related papers  (2024-01-18T14:43:47Z)
• Improving Large Language Models via Fine-grained Reinforcement Learning with Minimum Editing Constraint [104.53687944498155]
  Reinforcement learning (RL) has been widely used in training large language models (LLMs) We propose a new RL method named RLMEC that incorporates a generative model as the reward model. Based on the generative reward model, we design the token-level RL objective for training and an imitation-based regularization for stabilizing RL process.
  arXiv  Detail & Related papers  (2024-01-11T17:58:41Z)
• Routing to the Expert: Efficient Reward-guided Ensemble of Large Language Models [69.51130760097818]
  We propose Zooter, a reward-guided routing method distilling rewards on training queries to train a routing function. We evaluate Zooter on a comprehensive benchmark collection with 26 subsets on different domains and tasks.
  arXiv  Detail & Related papers  (2023-11-15T04:40:43Z)
• Handling Sparse Rewards in Reinforcement Learning Using Model Predictive Control [9.118706387430883]
  Reinforcement learning (RL) has recently proven great success in various domains. Yet, the design of the reward function requires detailed domain expertise and tedious fine-tuning to ensure that agents are able to learn the desired behaviour. We propose to use model predictive control(MPC) as an experience source for training RL agents in sparse reward environments.
  arXiv  Detail & Related papers  (2022-10-04T11:06:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.