brain-bzh/reve-dataset

该子集包含了用于训练REVE（脑电图表示学习）的开放访问数据，约占REVE预训练总数据量的4.85TB。数据集设计用于促进大规模脑电图基础模型的开发，支持自监督学习、基准测试和数据规模影响研究。数据格式为大规模NumPy二进制文件（.npy），优化了内存映射（memmap）访问方式，以高效处理多TB级数据而不需全部加载到RAM中。

This subset represents approximately 4.85TB of raw and preprocessed EEG data—roughly half of the total REVE pretraining volume. It is designed to facilitate the development of large-scale foundation models for electroencephalography. The dataset is in large-scale NumPy binaries (.npy) format, optimized for memory-mapping (memmap) to allow efficient access without loading the entire multi-terabyte corpus into RAM. Intended uses include pretraining for self-supervised learning, benchmarking, and researching scaling laws.