突破探索瓶颈：基于评分标准的强化学习助力通用大语言模型推理

摘要

近期，大型语言模型（LLMs）的进展凸显了强化学习（RL）在促进推理能力涌现方面的潜力。尽管取得了令人鼓舞的成果，但一个根本性难题依然存在：RL的改进依赖于从高质量样本中学习，而此类样本的探索却受限于LLMs的固有局限。这实际上形成了一个不良循环，即无法探索的内容也就无法学习。在本研究中，我们提出了“Rubric-Scaffolded Reinforcement Learning”（RuscaRL），一种新颖的教学支架框架，旨在突破通用LLM推理的探索瓶颈。具体而言，RuscaRL引入了清单式评分标准作为：（1）在生成阶段为探索提供显式支架，通过任务指令中提供不同的评分标准作为外部指导，引导多样化的高质量响应。这种指导随时间逐渐衰减，鼓励模型内化底层的推理模式；（2）在模型训练期间为利用提供可验证的奖励，通过以评分标准为参考，获得稳健的LLM-as-a-Judge评分，从而在通用推理任务上实现有效的RL。大量实验证明了所提出的RuscaRL在多个基准测试中的优越性，有效扩展了在最佳N评估下的推理边界。值得注意的是，RuscaRL将Qwen-2.5-7B-Instruct在HealthBench-500上的得分从23.6显著提升至50.3，超越了GPT-4.1。此外，我们在Qwen3-30B-A3B-Instruct上的微调变体在HealthBench-500上达到了61.1分，超越了包括OpenAI-o3在内的领先LLMs。

English

Recent advances in Large Language Models (LLMs) have underscored the potential of Reinforcement Learning (RL) to facilitate the emergence of reasoning capabilities. Despite the encouraging results, a fundamental dilemma persists as RL improvement relies on learning from high-quality samples, yet the exploration for such samples remains bounded by the inherent limitations of LLMs. This, in effect, creates an undesirable cycle in which what cannot be explored cannot be learned. In this work, we propose Rubric-Scaffolded Reinforcement Learning (RuscaRL), a novel instructional scaffolding framework designed to break the exploration bottleneck for general LLM reasoning. Specifically, RuscaRL introduces checklist-style rubrics as (1) explicit scaffolding for exploration during rollout generation, where different rubrics are provided as external guidance within task instructions to steer diverse high-quality responses. This guidance is gradually decayed over time, encouraging the model to internalize the underlying reasoning patterns; (2) verifiable rewards for exploitation during model training, where we can obtain robust LLM-as-a-Judge scores using rubrics as references, enabling effective RL on general reasoning tasks. Extensive experiments demonstrate the superiority of the proposed RuscaRL across various benchmarks, effectively expanding reasoning boundaries under the best-of-N evaluation. Notably, RuscaRL significantly boosts Qwen-2.5-7B-Instruct from 23.6 to 50.3 on HealthBench-500, surpassing GPT-4.1. Furthermore, our fine-tuned variant on Qwen3-30B-A3B-Instruct achieves 61.1 on HealthBench-500, outperforming leading LLMs including OpenAI-o3.

突破探索瓶颈：基于评分标准的强化学习助力通用大语言模型推理

Breaking the Exploration Bottleneck: Rubric-Scaffolded Reinforcement Learning for General LLM Reasoning

摘要

Support