curveball-steering/convsersations_humor_llama3.2-1B-it_large

该数据集包含对话提示和响应对，每个样本包括提示文本、响应文本、标签、主题和子主题，以及一个模型的多层残差向量（从第0层到第15层）。数据集分为训练集和测试集，训练集有7702个样本，测试集有1926个样本。它可能用于自然语言处理任务，如对话生成或模型分析，通过残差向量提供额外的模型内部表示信息。

This dataset contains dialogue prompt and response pairs, with each sample including prompt text, response text, label, topic, subtopic, and multi-layer residual vectors (from layer 0 to layer 15) of a model. The dataset is split into training and test sets, with 7702 examples in the training set and 1926 in the test set. It may be used for natural language processing tasks such as dialogue generation or model analysis, providing additional internal representation information through residual vectors.