SQLite-Vector adds support for float16 and bfloat16 (CPU, NEON, AVX2 and SSE2)
2 days ago
- #Edge AI
- #Vector Search
- #SQLite
- SQLite Vector is a cross-platform SQLite extension for vector search, supporting iOS, Android, Windows, Linux, and macOS with a 30MB memory footprint.
- Supports multiple data types including Float32, Float16, BFloat16, Int8, and UInt8, optimized for Edge AI applications.
- No need for virtual tables—vectors are stored as BLOBs in ordinary tables.
- Features include blazing-fast performance with SIMD acceleration, low memory usage, and no preindexing required.
- Works offline, making it ideal for privacy-preserving, on-device AI workloads.
- Plug-and-play integration with existing SQLite workflows.
- Comparison table highlights advantages over traditional solutions like no external server requirement and memory efficiency.
- Easy setup with pre-built binaries for various platforms and simple SQL commands for integration.
- Supports various distance functions (L2, L1, Cosine, Dot Product) optimized for performance.
- Ideal for applications like semantic search, image retrieval, recommendation systems, and anomaly detection.
- Designed for Edge AI, with offline capabilities and mobile-friendly operation.
- Compatible with other SQLite extensions like SQLite-AI, SQLite-Sync, and SQLite-JS.
- Licensed under Elastic License 2.0, with commercial licenses available for production use.