chimbiwide/databricks-thinking
收藏Hugging Face2025-12-16 更新2025-12-20 收录
下载链接：
https://hf-mirror.com/datasets/chimbiwide/databricks-thinking
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为databricks-thinking，是通过从databricks-dolly数据集中提取问题，并使用Qwen3-14b模型生成推理痕迹和答案而创建的。其目的是解决现有公开数据集中推理痕迹过长、且主要集中在数学、编程和科学领域的问题，特别关注创意任务如写作、头脑风暴和摘要。数据集展示了模型如何生成创意写作的推理痕迹和最终答案，适用于增强小型模型在创意任务中的表现。
The dataset named databricks-thinking was created by extracting questions from the databricks-dolly dataset and using the Qwen3-14b model to synthetically generate reasoning traces and answers. Its purpose is to address the issues of overly long reasoning traces in existing public datasets and their focus on math, coding, and science, with particular attention to creative tasks such as writing, brainstorming, and summarization. The dataset demonstrates how the model generates reasoning traces and final answers for creative writing, suitable for enhancing the performance of smaller models in creative tasks.
提供机构：
chimbiwide
