PotatoPulse/sae-unlearning-contrastive-small

该数据集包含10,000个训练示例，主要用于生成模型（如扩散模型）的相关研究。每个示例包括以下特征：activations（一个形状为256x1280的浮点数组，可能表示神经网络层的输出激活），timestep（一个无符号整数，表示生成过程的时间步），object_class（字符串，表示对象类别），style_class（字符串，表示风格类别），base_prompt_idx（整数，可能表示基础提示的索引），以及object_name（字符串，表示对象名称）。数据集总大小约为6.56 GB，适用于机器学习和人工智能任务，如模型激活分析、生成过程优化或类别控制研究。This dataset consists of 10,000 training examples, primarily designed for research on generative models, such as diffusion models. Each example includes the following features: activations (a float array with shape 256x1280, likely representing neural network layer output activations), timestep (an unsigned integer indicating the timestep in the generation process), object_class (a string denoting the object category), style_class (a string denoting the style category), base_prompt_idx (an integer possibly indexing a base prompt), and object_name (a string representing the object name). The total dataset size is approximately 6.56 GB, and it is suitable for machine learning and AI tasks, such as model activation analysis, generation process optimization, or category control studies.