science-of-finetuning/diffing-stats-gemma-2-2b-crosscoder-l13-mu4.1e-02-lr1e-04
收藏Hugging Face2025-03-10 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/science-of-finetuning/diffing-stats-gemma-2-2b-crosscoder-l13-mu4.1e-02-lr1e-04
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了在使用gemma 2 2B模型的第13层训练的crosscoder上特征激活程度最高的例子。具体包括三个文件：base_examples.pt包含在fineweb验证测试子集上的特征最大激活例子，chat_examples.pt包含在lmsys聊天数据上的特征最大激活例子，而chat_base_examples.pt是前两个文件的合并。所有文件存储格式为字典，键为特征索引，值为按激活程度排序的样本列表。
This dataset contains the most activating examples for features on a crosscoder trained with gemma 2 2B model at layer 13. It includes three files: base_examples.pt with the maximum activation examples on a subset of the fineweb validation test, chat_examples.pt with the maximum activation examples on lmsys chat data, and chat_base_examples.pt which is a merge of the first two files. All files are stored in a dictionary format where the keys are feature indices and the values are lists of samples sorted by activation level.
提供机构：
science-of-finetuning
