cross-modal-multivector-search - Multi-Vector Search

Overview

The SearchMultiVectorOnRoarGraph function introduces efficient multi-query nearest neighbor search on a Roar graph, leveraging dynamic priority queue management for improved accuracy and scalability.

Key Features

• Multi-Query Support: Handles simultaneous processing of multiple query vectors.
• Dynamic Priority Queue: Adjusts queue sizes based on result quality and resource constraints.
• Efficient Traversal: Uses prefetching and visited lists to optimize memory and computation.
• Scalable and Adaptive: Expands queries selectively to balance precision and cost.

Example usage

```cpp std::vector queries = {...}; sizet k = 10; Parameters params = {{"minpq", 50}, {"maxpq", 200}, {"maxpqsizebudget", 5000}}; std::vector> indices(queries.size()); std::vectorstd::vector<float> res_dists(queries.size());

auto results = index.SearchMultiVectorOnRoarGraph(queries, k, queryid, params, indices, resdists); ```

- C++
Published by hyojeongyunn over 1 year ago