EpistemeAI/plan-reason-deep-reasoning-fix-openai

Next-Gen推理循环数据集是基于Nathan Lambert的演讲“下一代推理模型的特征”而受启发的。该数据集为大型语言模型引入了一个结构化的多阶段推理循环。它不仅包括简单的问题-答案对，还增加了明确的推理阶段：计划、回答、复核和自信度。这种结构鼓励模型进行更透明的推理、自我纠正和校准自信度。每个样本包含用户的输入或问题、模型的推理步骤、提出的解决方案和原始模型输出。

The Next-Gen Reasoning Cycle Dataset is inspired by Nathan Lamberts talk Traits of Next Generation Reasoning Models. This dataset introduces a structured multi-phase reasoning cycle for large language models (LLMs), which includes planning, answering, double-checking, and confidence phases. This structure promotes transparent reasoning, self-correction, and confidence calibration in models. Each sample contains the users input or question, the models reasoning steps, the proposed solution, and the raw models output.