homerquan/boardgamebench-answer-sft

BoardGameBench Answer SFT数据集是一个用于监督微调（SFT）的棋盘游戏推理数据集，包含1,282,766个仅包含答案的示例，这些示例由BoardGameBench项目生成，旨在评估语言模型在结构化棋盘游戏决策中的表现。数据集覆盖多种经典棋盘游戏，如Connect Four、Gomoku、Othello、Hex、Breakthrough和Dots and Boxes，每个示例要求模型根据合法棋盘位置选择最佳移动，并以简单格式（id、prompt、answer）呈现，其中answer字段是目标移动标签（例如C4或e2-d3）。该数据集设计用于帮助语言模型学习解析棋盘状态、遵守合法移动规则，并选择更强动作，适用于监督微调、精确匹配评估或作为偏好学习实验的种子数据。The BoardGameBench Answer SFT Dataset is a supervised fine-tuning (SFT) corpus for board-game reasoning, containing 1,282,766 answer-only examples generated from the BoardGameBench project, which is a benchmark and data-generation initiative for evaluating language models on structured board-game decision making. It includes positions from classic board games such as Connect Four, Gomoku, Othello, Hex, Breakthrough, and Dots and Boxes. Each example prompts a model to inspect a legal board position and return the best move, with a simple format (id, prompt, answer) where the answer field is the target move label (e.g., C4 or e2-d3). This dataset is designed to help language models learn to parse board state, respect legal moves, and select stronger actions, and it is suitable for supervised fine-tuning, exact-match evaluation, or as seed data for preference-learning experiments.