Revisiting Reinforcement Learning for LLM Reasoning from A Cross-Domain Perspective

2025-06-17

AI Summary

Existing reinforcement learning (RL) approaches for enhancing large language model (LLM) reasoning focus predominantly on mathematics and coding, suffering from scarce and poorly understood cross-domain reward signals. Method: We introduce Guru, the first verifiable cross-domain RL reasoning corpus (92K samples spanning mathematics, code, science, logic, simulation, and tabular reasoning), and propose a domain-aware RL training paradigm integrating domain-specific rewards, intra- and inter-domain fine-tuning, and PPO optimization. Contribution/Results: We demonstrate that RL drives genuine skill acquisition—not mere knowledge activation—in pretraining-underexposed domains (e.g., logic, simulation, tabular reasoning). We release Guru-7B/32B models, achieving SOTA on 17 cross-domain reasoning benchmarks (+7.9%/+6.7% over baselines) and substantially improving Pass@k on complex tasks. All data, models, and code are open-sourced.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Reinforcement learning (RL) has emerged as a promising approach to improve large language model (LLM) reasoning, yet most open efforts focus narrowly on math and code, limiting our understanding of its broader applicability to general reasoning. A key challenge lies in the lack of reliable, scalable RL reward signals across diverse reasoning domains. We introduce Guru, a curated RL reasoning corpus of 92K verifiable examples spanning six reasoning domains--Math, Code, Science, Logic, Simulation, and Tabular--each built through domain-specific reward design, deduplication, and filtering to ensure reliability and effectiveness for RL training. Based on Guru, we systematically revisit established findings in RL for LLM reasoning and observe significant variation across domains. For example, while prior work suggests that RL primarily elicits existing knowledge from pretrained models, our results reveal a more nuanced pattern: domains frequently seen during pretraining (Math, Code, Science) easily benefit from cross-domain RL training, while domains with limited pretraining exposure (Logic, Simulation, and Tabular) require in-domain training to achieve meaningful performance gains, suggesting that RL is likely to facilitate genuine skill acquisition. Finally, we present Guru-7B and Guru-32B, two models that achieve state-of-the-art performance among open models RL-trained with publicly available data, outperforming best baselines by 7.9% and 6.7% on our 17-task evaluation suite across six reasoning domains. We also show that our models effectively improve the Pass@k performance of their base models, particularly on complex tasks less likely to appear in pretraining data. We release data, models, training and evaluation code to facilitate general-purpose reasoning at: https://github.com/LLM360/Reasoning360

Problem

• Lack of reliable RL rewards across diverse reasoning domains
• Limited understanding of RL's broader applicability beyond math and code
• Need for domain-specific training to achieve meaningful performance gains

Innovation

• Cross-domain RL training for LLM reasoning
• Domain-specific reward design in Guru corpus
• State-of-the-art Guru-7B and Guru-32B models