Gemma-2-2b-it
收藏arXiv2025-09-30 收录
下载链接：
https://www.neuronpedia.org/gemma-2-2b-it/steer
下载链接
链接失效反馈官方服务：
资源简介：
该数据集基于Gemma-2-2b-it模型的一个参考特征集，旨在对各种概念执行可扩展的注意力模块发现（SAMD）。它包含了语义上不同的代表性概念，如简单名词、专有名词、动词以及互动语言等。该数据集大约为每个所选的4个概念提供了大约100个提示。其任务是发现并干预变压器模型中不同概念的注意力模块。
This dataset is developed based on a reference feature set of the Gemma-2-2b-it model, with the goal of performing scalable attention module discovery (SAMD) across various concepts. It includes semantically distinct representative concepts such as simple nouns, proper nouns, verbs, interactive language, and more. Approximately 100 prompts are provided for each of the 4 selected concepts in this dataset. The task of this dataset is to discover and intervene in attention modules of Transformer models that correspond to different concepts.
提供机构：
Unknown (based on referenced works)
