Expand description
vecq — training-free vector quantization and search at configurable width (4/5/6-bit, default 5-bit).
The “SQLite profile” for vector storage: single file, embedded,
deterministic, no training pass. Vectors are rotated with a randomized
Hadamard transform (seeded, stored in the file header) so coordinates
become approximately N(0,1), then quantized with precomputed Lloyd-Max
tables and bit-packed LSB-first. view::VecqView serves the same
file format zero-copy from any byte owner (e.g. an mmap).
Re-exports§
pub use store::cosine_f32;pub use store::PreparedQuery;pub use store::VecqIndex;pub use view::VecqView;
Modules§
- format
- Single-file persistence format.
- lloyd
- Lloyd-Max optimal scalar quantizer tables for N(0,1), precomputed offline.
- rhdh
- Randomized Hadamard Transform (RHDH): R = (1/sqrt(d’)) · H · D.
- store
- Quantized store: nibble-packed 4-bit codes + per-vector norm correction, with asymmetric inner-product scoring (query f32, database 4-bit).
- view
- Zero-copy read-only index view (
VecqView) — issue #25.