Expand description
Scalar Quantization (SQ) for efficient vector compression
This module implements Scalar Quantization, a simpler and faster alternative to Product Quantization (PQ) that quantizes each vector dimension independently.
SQ is particularly useful when:
- Training time needs to be minimal
- Simple, predictable compression is preferred
- Memory reduction is more important than extreme accuracy
- Real-time index updates are required (no retraining needed)
Structs§
- Quantization
Params - Quantization parameters for a single dimension
- SqConfig
- Scalar quantization configuration
- SqIndex
- Scalar quantization vector index
- SqStats
- Scalar quantization index statistics
Enums§
- Quantization
Mode - Quantization mode