GambitFlow/Elite-Data

该数据集是用于训练强大的俱乐部级别国际象棋评估模型（如Nexus-core CE）的高度精选输入数据。通过仅包含ELO评分超过2000的精英级别对局，数据集旨在最大化国际象棋数据中的信噪比，避免模型学习业余玩家的常见错误。数据来源于Lichess公共数据库（2017年1月），经过严格筛选和处理，最终存储在SQLite文件中，包含超过500万个棋局位置（248万多个唯一位置），文件大小为882MB。数据集采用CC BY-NC 4.0许可证，禁止商业使用。

This dataset is the highly curated input required to train strong, club-level chess evaluation models like the Nexus-core CE. It is designed to maximize the signal-to-noise ratio in chess data by removing moves made by lower-rated players. By exclusively training on Elite-level games, the resulting AI avoids learning common amateur mistakes and focuses on solid positional principles. The data was sourced from the Lichess Public Database (January 2017), filtered to include only games with White ELO > 2000 AND Black ELO > 2000, and processed into a SQLite file containing over 5,000,000 total positions (2,488,753 unique positions), with a file size of 882 MB. The dataset is licensed under CC BY-NC 4.0, with commercial use strictly prohibited.