Vector Retrieval Service:What is DashVector?

DashVector is a vector search service based on Proxima, an efficient vector engine developed by Tongyi Lab. The service is cloud-native, fully-managed, and horizontally scalable. DashVector offers diverse capabilities, such as vector management and search. These capabilities are encapsulated in the SDK and API so that you can integrate them into your AI applications with ease. This way, you can guarantee efficient vector search in various scenarios, such as large models, multi-modal AI search, and molecular structure analysis.

• High precision for multi-dimensional data: selects or combines different algorithms based on data dimensions and distributions. This contributes to the balance between precision and performance based on the needs of specific scenarios.

• Online updates in real time: adopts a flat index structure to support the stream construction of online large-scale vector indexes from scratch. This facilitates real-time dynamic updates and allows vectors to be searched and stored in disks in an instant manner.

• High performance with low cost: maximizes performance and meets business needs at a limited cost.

• Adaptation to multiple scenarios: enhances automation capabilities and makes services easy-to-use through hyperparameter optimization and composite indexing.

• Ultra-large-scale index building and search: achieves high search efficiency at a low cost by introducing composite search algorithms combined with skillful project implementation and algorithm optimization. Indexing based on a microcontroller can reach billions in scale.

• Vector search with tags: enables conditional vector searches at the index algorithm layer, which solves the problem of unsatisfactory recalling results caused by multi-channel storage. This meets the requirements of composite searches.

• Index horizontal scalability: implements distributed searches by non-peer-to-peer sharding and supports the quick merge of indexes under a limited precision. This allows the service to be effectively combined with MapReduce calculation models.

• Heterogeneous computing: supports accelerating offline searches in large-scale, high-throughput scenarios and building peer graph indexes based on GPU. This allows resources to be used in small batches with low latency and high throughput.