Expand description
Scalar Quantization (SQ) for memory-efficient vector storage
Scalar Quantization reduces memory usage by quantizing each float dimension to a smaller integer type (typically uint8). This provides ~4x memory reduction with minimal accuracy loss for most use cases.
§Supported Quantization Types
- SQ4: 4-bit quantization (8x compression, lower accuracy)
- SQ8: 8-bit quantization (4x compression, good accuracy)
- SQ16: 16-bit quantization (2x compression, high accuracy)
Structs§
- SQConfig
- Configuration for Scalar Quantization
- SQIndex
- Scalar Quantization Index
- SQSearch
Result - Search result from SQ index
- SQStats
- Statistics for quantization
Enums§
- Quantization
Type - Quantization bit depth