cetusian/chess-sft-lichess-2200

Lichess Chess SFT (≥2200 Elo) 是一个用于训练语言模型通过预测PGN前缀中的下一步走棋来下棋的监督微调数据集。该数据集源自Lichess上的高水平棋手对局，筛选条件为双方Elo评分均≥2200，并包含195,999个训练样本和4,000个验证样本。每个样本由一个PGN走棋前缀和下一步的标准代数记法（SAN）走棋组成。数据集的构建旨在防止模型记忆单个对局的长期片段，并确保开局、中局和残局的平衡表示。

Lichess Chess SFT (≥2200 Elo) is a supervised fine-tuning dataset for training language models to play chess by predicting the next move from a PGN prefix. Derived from strong-player Lichess games, the dataset is filtered for games where both players have an Elo rating of at least 2200, and includes 195,999 training and 4,000 validation examples. Each example consists of a PGN move prefix and the next move in Standard Algebraic Notation (SAN). The dataset is constructed to prevent memorization of individual games and to ensure a balance of openings, middlegame, and endgame positions.