LieUr/Qwen2-0.5B-Instruct_piqa_oai_contrastive
收藏资源简介:
该数据集包含三个配置:keys、residuals和values,每个配置用于存储神经网络模型的层特征数据。每个样本包括一个整数标签(label)和24个层(layer_0到layer_23)的特征,其中keys和values配置的层特征为二维浮点数列表(list of list of float32),而residuals配置的层特征为一维浮点数列表(list of float32)。数据集仅包含训练集,共400个样本,总大小约为50.3 MB(keys和values)和34.4 MB(residuals)。这些数据可能用于模型分析、特征提取或机器学习任务,但具体应用场景未在README中说明。
This dataset includes three configurations: keys, residuals, and values, each designed to store layer feature data from a neural network model. Each sample consists of an integer label and features from 24 layers (layer_0 to layer_23), where the keys and values configurations have two-dimensional floating-point lists (list of list of float32) as layer features, while the residuals configuration has one-dimensional floating-point lists (list of float32). The dataset contains only a training split with 400 samples, with total sizes of approximately 50.3 MB for keys and values, and 34.4 MB for residuals. These data may be used for model analysis, feature extraction, or machine learning tasks, though the specific application context is not detailed in the README.




