用迭代优化方法生成更难更广的推理题，提升大模型自研能力
PromptCoT 2.0: Scaling Prompt Synthesis for Large Language Model Reasoning
- 通过期望最大化循环自动优化提示中的推理链，替代人工规则
- 生成题目在多个竞赛数据集上超越人类数据，最高提升5.3分
- 适合想提升模型推理能力的研究者和开源项目开发者
大语言模型正从对话系统演变为解决奥数和编程竞赛等复杂任务的强推理工具。尽管参数量和推理计算量的扩展推动了进展，但高质量训练题稀缺成为关键瓶颈：人工标注数据成本高、范围有限，现有合成数据集则往往过于简单或单一。PromptCoT 1.0证明，在提示生成中注入推理链可提高题目难度。在此基础上，我们提出PromptCoT 2.0，一个可扩展的框架，用期望最大化（EM）循环替代手工启发式规则，通过迭代精炼推理链来指导提示构建。该方法生成的问题比以往更难且更具多样性。这些合成提示支持两种后训练策略：(1) 自对弈（Self-Play），强模型通过可验证反馈自主改进，无需更强教师；(2) 监督微调（SFT），弱模型从教师蒸馏的推理轨迹中学习。实验表明，将PromptCoT 2.0应用于Qwen3-30B-A3B-Thinking-2507，在30B规模上取得新SOTA：AIME 24/25和HMMT 25分别提升+4.4、+4.8、+5.3，LiveCodeBench v5/v6提升+6.1和+5.0，Codeforces提升+35 Elo。在SFT中，仅用合成提示训练Qwen2.5-7B-Instruct，准确率达73.1（AIME 24）、65.6（AIME 25）、53.4（LiveCodeBench v5），超过使用人工或混合数据训练的模型。分析进一步证实，PromptCoT 2.0生成的问题本质更难，分布也更独特。这些结果确立提示合成作为推理能力扩展的新维度，并使PromptCoT 2.0成为未来开源模型的重要基础。代码已开源：https://github.com/inclusionAI/PromptCoT。
原文摘要 · Abstract (English)
Large language models (LLMs) are evolving from conversational systems into strong reasoners for tasks such as Olympiad mathematics and competitive programming. While scaling parameters and test-time computation has driven progress, a key bottleneck is the lack of high-quality training problems: human-curated datasets are costly and limited, while existing synthetic corpora are often too easy or narrow. PromptCoT 1.0 showed that injecting rationales into prompt synthesis increases problem difficulty. Building on this, we present PromptCoT 2.0, a scalable framework that replaces hand-crafted heuristics with an expectation-maximization (EM) loop, where rationales are iteratively refined to guide prompt construction. This produces problems that are both harder and more diverse than prior corpora. The synthetic prompts support two post-training regimes: (1) Self-Play, where strong models improve autonomously via verifiable feedback without stronger teachers; and (2) Supervised Fine-Tuning (SFT), where weaker models learn from teacher-distilled traces. Extensive experiments demonstrate the effectiveness of this approach. In self-play, applying PromptCoT 2.0 to Qwen3-30B-A3B-Thinking-2507 sets new state-of-the-art results at the 30B scale, with +4.4, +4.8, and +5.3 on AIME 24/25 and HMMT 25, +6.1 and +5.0 on LiveCodeBench v5/v6, and +35 Elo on Codeforces. In SFT, training Qwen2.5-7B-Instruct solely on synthetic prompts boosts accuracy to 73.1 (AIME 24), 65.6 (AIME 25), and 53.4 (LiveCodeBench v5), surpassing models trained on human or hybrid data. Analyses further confirm that PromptCoT 2.0 yields fundamentally harder and distributionally distinct problems. These results establish prompt synthesis as a new axis for scaling reasoning and position PromptCoT 2.0 as a scalable foundation for future open-source models. The implementation is available at https://github.com/inclusionAI/PromptCoT.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。