OpenSearch - Vector Search Edition launches QGraph vector algorithm
Dec 05 2024
OpenSearchContent
Target customers: 1. Customers who need to save costs 2. Customers with high requirements on vector query performance 3. Customers with requirements such as image-based search, AI semantic search, and RAG. Features released: Quantized Graph (QGraph) is an improved HNSW-based algorithm developed by OpenSearch. QGraph automatically quantifies the raw data of users during index construction and then builds a graph index. Compared with the HNSW algorithm, the QGraph algorithm can effectively reduce the index size and save memory overhead, and can reduce the index to up to 1/8 of the original one. At the same time, with the optimization of CPU instructions for integer computing, the performance of QGraph is improved by several times compared with HNSW. After quantization, the discrimination of vectors is reduced, and the recall rate of QGraph is lower than that of HNSW. In real scenarios, more recalls can be used to reduce this impact.
Help Document
https://www.alibabacloud.com/help/open-search/vector-search-edition/qgraph-quantized-graph-configuration?spm=a2c63.p38356.help-menu-29102.d_3_0_4.29a87f46S1c0PO
