blythet/deepseek-v4-pro-math-cot-1k
收藏资源简介:
DeepSeek V4 Pro Math CoT 1K 是一个小型、高质量的监督微调(SFT)数据集,专注于数学推理轨迹。数据集中的问题来自 Nemotron 数学问题集(最初来源于 StackExchange-Math 和 AoPS),由 DeepSeek V4 Pro 在高推理努力下启用思考功能回答,并由 DeepSeek V4 Flash 独立审查其正确性。病态推理轨迹(如循环、长度失控、过多的 `Wait` 式回溯)被隔离。该数据集旨在为小型/中型开放模型(如 Qwen3.5-9B)提供一个干净、紧凑的长篇数学推理微调语料库。数据集设计小巧,每一行都经过更强审查模型的单独检查。
DeepSeek V4 Pro Math CoT 1K is a small, high-signal supervised-fine-tuning (SFT) dataset of math reasoning traces. Problems were sampled from a Nemotron math problem set (originally sourced from StackExchange-Math and AoPS), answered by DeepSeek V4 Pro with thinking enabled at high reasoning effort, then independently reviewed by DeepSeek V4 Flash for correctness against the expected answer. Pathological reasoning traces (looping, run-away length, excessive `Wait`-style backtracking) were quarantined. This dataset is intended as a clean, compact corpus for fine-tuning small/mid-size open models (e.g. Qwen3.5-9B) on long-form mathematical reasoning. It is small by design: every row has been individually inspected by a stronger reviewer model.




