SQLite-vec v0.1.0: a vector search SQLite extension that runs everywhere
sqlite-vec v0.1.0 is a new SQLite extension for vector search, supporting multiple programming languages and operating systems. It focuses on brute-force search, with future updates planned for ANN indexing.
Read original article
sqlite-vec v0.1.0 is a new SQLite extension designed for vector search, developed entirely in C without dependencies and licensed under MIT/Apache-2.0. This stable release allows users to integrate it into various projects, with installation options available for multiple programming languages including Python, Node.js, Ruby, Rust, and Go. The extension operates similarly to SQLite's full-text search, enabling users to create virtual tables with vector columns and perform KNN-style queries efficiently. It is compatible with major operating systems, including MacOS, Linux, and Windows, and can run in web browsers via WebAssembly.
Currently, sqlite-vec focuses on brute-force vector search, which is effective for datasets typically ranging from thousands to hundreds of thousands of vectors. Future updates are expected to introduce approximate nearest neighbor (ANN) indexing. The extension supports vector quantization techniques to reduce storage space and improve query speed, with minimal loss in accuracy. Benchmarks indicate that sqlite-vec performs well in various scenarios, particularly with smaller dimensions and bit vectors, achieving fast query times. However, performance may decline with larger datasets, especially with float vectors. Overall, sqlite-vec is positioned as a practical solution for local AI applications, with ongoing developments anticipated to enhance its capabilities further.
Related
Optimizing Large-Scale OpenStreetMap Data with SQLite
The article discusses optimizing large-scale OpenStreetMap data with SQLite. Converting OSMPBF to SQLite enhanced search functionalities. Indexing, full-text search, and compression improved query performance, despite some false positives.
DuckDB: Vector Similarity Search Extension
The vss extension in DuckDB enhances vector similarity search with HNSW indexing for ARRAY columns. Users can optimize queries with distance metrics but should be cautious due to limitations and experimental features.
Vectorlite: Fast Vector Search for SQLite
Vectorlite is a runtime-loadable extension for SQLite enabling fast vector search with hnswlib on Windows, MacOS, and Linux. It supports SIMD acceleration, various distance types, and customizable HNSW parameters. Installation via `pip install vectorlite-py` in Python is suggested for usage. The GitHub page offers examples, API references, benchmarks, and more for detailed exploration.
First Contact with SQLite
The article explores surprising aspects of SQLite, like limitations in altering columns and data types. It notes the recent jsonb support and handling date/time values, praising SQLite's streaming features but favoring PostgreSQL.
SQLite: 35% Faster Than the Filesystem
SQLite outperforms traditional file I/O by achieving 35% faster reads of small blobs and using 20% less disk space for 10-kilobyte blobs, despite some latency issues in specific scenarios.
3 commentsRelated
Optimizing Large-Scale OpenStreetMap Data with SQLite
The article discusses optimizing large-scale OpenStreetMap data with SQLite. Converting OSMPBF to SQLite enhanced search functionalities. Indexing, full-text search, and compression improved query performance, despite some false positives.
DuckDB: Vector Similarity Search Extension
The vss extension in DuckDB enhances vector similarity search with HNSW indexing for ARRAY columns. Users can optimize queries with distance metrics but should be cautious due to limitations and experimental features.
Vectorlite: Fast Vector Search for SQLite
Vectorlite is a runtime-loadable extension for SQLite enabling fast vector search with hnswlib on Windows, MacOS, and Linux. It supports SIMD acceleration, various distance types, and customizable HNSW parameters. Installation via `pip install vectorlite-py` in Python is suggested for usage. The GitHub page offers examples, API references, benchmarks, and more for detailed exploration.
First Contact with SQLite
The article explores surprising aspects of SQLite, like limitations in altering columns and data types. It notes the recent jsonb support and handling date/time values, praising SQLite's streaming features but favoring PostgreSQL.
SQLite: 35% Faster Than the Filesystem
SQLite outperforms traditional file I/O by achieving 35% faster reads of small blobs and using 20% less disk space for 10-kilobyte blobs, despite some latency issues in specific scenarios.