unum-cloud/USearchWiki
收藏资源简介:
USearchWiki是一个基于HuggingFace FineWiki构建的多模型嵌入数据集,专为近似最近邻(ANN)搜索基准测试设计。该数据集包含来自多种模型的嵌入,预计算的地面真实k近邻,以及用于过滤搜索的图结构元数据。源语料库来自维基百科,经过处理和丰富,包含各种注释。数据集结构支持多种嵌入模型,详细描述了所使用的模型、其性能以及数据集的布局。README还涵盖了下载选项、嵌入和处理的工作流程、托管详情以及许可信息。
USearchWiki is a multi-model embedding dataset built on HuggingFace FineWiki, specifically designed for approximate nearest neighbor (ANN) search benchmarking. This dataset contains embeddings from multiple models, pre-computed ground-truth k-nearest neighbors, and graph-structured metadata for search filtering. The source corpus is derived from Wikipedia, processed and enriched with various annotations. The dataset structure supports multiple embedding models, and details the employed models, their performance, as well as the dataset layout. The accompanying README also covers download options, embedding and processing workflows, hosting details, and licensing information.




