cohere_large_10m
收藏资源简介:
该数据集为VectorDBBench中的Cohere Large 10M数据集预构建的DiskANN索引,专为分布式向量搜索设计。数据集包含10,000,000个768维的float32类型向量和10,000个查询向量,使用L2距离度量。数据集提供了多种分片配置(3、5、7、10分片)和不同的DiskANN参数(R=16,32,64;L=100;PQ bytes=192)。文件结构包括原始向量数据(fbin/)、原始Parquet格式数据(parquet/)以及分片后的DiskANN索引文件(diskann/)。数据集适用于向量搜索基准测试和最近邻搜索任务。
This dataset is a pre-built DiskANN index for the Cohere Large 10M dataset within VectorDBBench, tailored specifically for distributed vector search. It comprises 10,000,000 768-dimensional float32 vectors and 10,000 query vectors, utilizing the L2 distance metric. Multiple sharding configurations (3, 5, 7, 10 shards) and a range of DiskANN parameters (R=16, 32, 64; L=100; PQ bytes=192) are offered. The file structure encompasses raw vector data (fbin/), raw Parquet-formatted data (parquet/), and sharded DiskANN index files (diskann/). This dataset is applicable for vector search benchmarking and nearest neighbor search tasks.



