lowry02/hs-layer-7
收藏官方服务：
资源简介：
该数据集是一个用于分析语言模型内部表示的结构化数据集，包含多个特征：数据集样本ID、数据集分割（如训练、验证、测试）、层索引、令牌位置、令牌ID、令牌标签、逻辑标签以及隐藏状态（以float16列表形式存储）。数据分为训练集（351,242个示例）、验证集（39,890个示例）和测试集（39,892个示例），总大小约1.78 GB，适用于NLP任务中的模型解释或特征分析。
This is a structured dataset for analyzing internal representations of language models, which contains multiple features: dataset sample ID, dataset split (e.g., training, validation, test), layer index, token position, token ID, token label, logical label, and hidden states stored as a float16 list. The dataset is divided into training set (351,242 examples), validation set (39,890 examples) and test set (39,892 examples), with a total size of approximately 1.78 GB. It is suitable for model interpretation or feature analysis in NLP tasks.
提供机构：
lowry02
