Jonathangrossman/chess-premium-dataset

这是一个来自NEXUS Engine的早期测试版本数据集，名为NEXUS Premium Chess RL Trace Dataset，旨在展示捕获、过滤和结构化高质量人类决策数据的能力，以用于AI对齐和行为建模。数据集包含475,989条记录，每条记录编码为一个人类决策，格式为(state, action, reward, next_state)，并与标准监督学习和强化学习训练管道兼容。它专门设计用于行为克隆、基于人类反馈的强化学习(RLHF)以及训练认知规划和空间推理的基础模型。数据经过质量过滤，确保来自国际象棋专家级玩家（ELO评分至少1800），并避免了标准开局记忆偏差。数据集以压缩Tarball形式提供，包含JSON Lines文件，每条记录一行。This is an early testing release from the NEXUS Engine, named NEXUS Premium Chess RL Trace Dataset, designed to demonstrate the capability to capture, filter, and structure high-quality human decision data for AI alignment and behavioral modeling. The dataset contains 475,989 records, each encoding a single human decision as (state, action, reward, next_state), and is drop-in compatible with standard supervised-learning and RL training pipelines. It is built specifically for Behavioral Cloning, Reinforcement Learning from Human Feedback (RLHF), and training Foundation Models on cognitive planning and spatial reasoning. Quality filters are applied to ensure data from expert-level chess players (minimum ELO 1800) and prevent standard opening-book memorization bias. The dataset is provided as a compressed Tarball containing JSON Lines files, with one record per line.