Millionbase

该数据集名为Millionbase，包含了约290万场高质量的国际象棋对局，在去除重复对局以及不符合特定步数限制的对局后，数据集被精简至约250万场。数据集被划分为训练集（20万场对局）、开发集（1.5万场对局）和测试集（1.5万场对局），此外还包含了5万场对局作为评估集，用于探测棋盘状态。经过筛选后，数据集规模大约为250万场对局。该数据集的任务是用于国际象棋状态跟踪的语言建模。

This dataset is named Millionbase. It contains approximately 2.9 million high-quality chess games. After removing duplicate games and filtering out those that do not meet the specified move limit, the dataset is reduced to approximately 2.5 million games. The dataset is split into a training set (200,000 games), a development set (15,000 games) and a test set (15,000 games). Additionally, 50,000 games are included as an evaluation set for probing chess board states. After the filtering process, the total scale of the dataset remains approximately 2.5 million games. The task of this dataset is language modeling for chess state tracking.