IntelLabs/BlueLens
收藏Hugging Face2025-05-15 更新2025-05-31 收录
下载链接：
https://hf-mirror.com/datasets/IntelLabs/BlueLens
下载链接
链接失效反馈官方服务：
资源简介：
BlueLense数据集是由Intel Labs创建的，包含从transformer模型的不同层级提取的中间特征。该数据集主要用于可解释AI的洞察和探索性分析，以及训练临时模型如线性探针、稀疏自动编码器或转码器。数据集以pyarrow格式存储，拥有超过1亿个训练样本和50万个测试样本。
The BlueLense dataset, created by Intel Labs, consists of intermediate features extracted from various layers of transformer models. It is primarily used for insights and exploratory analysis in explainable AI, as well as for training ad-hoc models such as linear probes, Sparse AutoEncoders (SAEs), or transcoders. The dataset is stored in pyarrow format and contains over 100 million training samples and 500,000 test samples.
提供机构：
IntelLabs
