xl-zhao/PromptCoT-2.0-SFT-4.8M

PromptCoT-2.0-SFT-4.8M数据集是使用PromptCoT 2.0技术生成的最大规模的合成提示数据集，包含480万个完全合成的推理轨迹提示，用于监督微调（SFT）实验。该数据集证明了纯合成数据在PromptCoT 2.0的生成下，可以训练出竞争力强的推理模型，这些模型的表现超过了人工编写的基线，如OpenMathReasoning和OpenCodeReasoning。数据集具有以下特点：规模大、质量和难度高、完全合成、推理轨迹较短、微调性能强。此外，该数据集和相应的教师响应完全开源，为推理研究提供了可扩展的资源。The PromptCoT-2.0-SFT-4.8M dataset is the largest dataset released with PromptCoT 2.0, containing 4.8 million fully synthetic prompts with reasoning trajectories for supervised fine-tuning (SFT) experiments. The dataset demonstrates that purely synthetic data, when generated with PromptCoT 2.0, can train competitive reasoning models that outperform human-curated baselines such as OpenMathReasoning and OpenCodeReasoning. The dataset is characterized by its large scale, high quality and difficulty, full synthesis, shorter reasoning traces, and strong SFT performance. Additionally, both the 4.8M prompts and the corresponding teacher responses are fully open-sourced, providing a scalable resource for reasoning research.