lekdan/osu-everything-tools

osu-everything数据集包含用于维护紧凑v1 osu!数据集的源工具和运行手册。数据集存储了原始.osz存档和紧凑的Parquet格式元数据，不发布提取的blob、清单、训练/测试分割或全局blob注册表。数据集还包含用于机器学习的派生特征存储。数据加载可以通过PyArrow进行，更新流程包括克隆或更新源代码、构建工具、配置本地osu!凭证以及运行更新脚本。数据集中的beatmaps经常包含受版权保护的歌曲、背景、视频和故事板资源，因此在没有适当权利的情况下不应公开分享。

The `osu-everything` dataset contains the source tools and runbooks for maintaining the compact v1 osu! dataset. It stores raw `.osz` archives and compact Parquet metadata, and does not publish extracted blobs, manifests, train/test splits, or a global blob registry. The dataset also includes a derived ML feature store. Data can be loaded using PyArrow, and the update process involves cloning or updating the source code, building tools, configuring local osu! credentials, and running update scripts. Beatmaps in the dataset frequently include copyrighted songs, backgrounds, videos, and storyboard assets, so the dataset should not be shared publicly without the appropriate rights.