lowry02/hs-layer-10

该数据集包含语言模型隐藏状态分析数据，用于研究模型内部表示。特征包括样本ID、数据分片（训练/验证/测试）、层索引、令牌位置、令牌ID、令牌标签、逻辑标签和隐藏状态向量。隐藏状态以float16列表形式存储，可能来自Transformer模型的特定层。数据集规模较大，总大小约1.79GB，包含训练集351242个示例、验证集39890个示例和测试集39892个示例，适用于机器学习任务如模型解释或逻辑分析。

This dataset contains hidden state analysis data from language models, designed for studying internal model representations. Features include sample ID, data split (train/validation/test), layer index, token position, token ID, token label, logic label, and hidden state vectors. Hidden states are stored as float16 lists, likely derived from specific layers of Transformer models. The dataset is large, with a total size of approximately 1.79GB, comprising 351242 training examples, 39890 validation examples, and 39892 test examples, suitable for machine learning tasks such as model interpretation or logical analysis.