LieUr/Qwen2-0.5B-Instruct_gsm8k_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(可能为Transformer架构)的中间层表示。keys和values配置包含label字段(int64类型)和24个layer_*字段(每个字段为float32类型的嵌套列表,表示多层结构),而residuals配置包含label字段和24个layer_*字段(每个字段为float32类型的单层列表)。所有配置仅包含训练集(train),各有400个示例,总数据大小从约5MB到34MB不等。数据集可能用于分析或训练与模型层表示相关的任务,如特征提取或可视化。
This dataset includes three configurations: keys, residuals, and values, each designed to store intermediate layer representations of a neural network (likely a Transformer architecture). The keys and values configurations contain a label field (int64 type) and 24 layer_* fields (each as a nested list of float32, representing multi-layer structures), while the residuals configuration contains a label field and 24 layer_* fields (each as a single-layer list of float32). All configurations consist only of a training set (train) with 400 examples each, and total data sizes range from approximately 5MB to 34MB. The dataset may be used for analysis or training tasks related to model layer representations, such as feature extraction or visualization.




