DashVector:What is Vector Retrieval Service DashVector

What is Vector Retrieval Service DashVector

DashVector is a cloud-native, fully managed vector retrieval service built on the Proxima kernel by Tongyi Lab. It supports horizontal scaling and provides vector management and query capabilities through SDKs and APIs for quick integration into AI applications. Use cases include Large Language Model (LLM) ecosystems, multimodal AI search, and molecular structure analysis.

• Multi-dimensional precision: Selects algorithms based on data dimensions and distribution to balance precision and performance.

• Real-time online updates: Uses a flat index structure for stream-based building of large-scale vector indexes. Supports immediate search on data insertion, instant disk persistence, and real-time updates.

• High performance and low cost: Maximizes performance and meets business requirements at a limited cost.

• Multi-scenario adaptability: Uses hyperparameter optimization and composite indexes for improved automation and ease of use.

• Ultra-large-scale indexing: Combines composite retrieval algorithms with underlying optimization for cost-effective retrieval. A single shard supports up to several billion vectors.

• Tag-based and vector retrieval: Implements conditional filtering at the index layer, avoiding suboptimal results from traditional multi-way merge retrieval.

• Horizontal scaling: Uses non-peer sharding for distributed retrieval with fast index merging with limited precision loss. Compatible with the MapReduce computing model.

• Heterogeneous computing: Accelerates large-batch offline retrieval with high throughput. Supports GPU-based neighbor graph index construction for low-latency, high-throughput resource utilization.