Temperature-tiered vector quantization for RuVector Format.
rvf-quant provides quantization codecs that reduce vector storage size based on access temperature:
- f32 -- full precision for hot vectors
- f16 -- half precision for warm vectors
- u8 -- scalar quantization for cool vectors
- binary -- 1-bit quantization for cold/archive vectors
- Automatic tiering -- promote/demote vectors based on access patterns
[dependencies]
rvf-quant = "0.1"std(default) -- enablestdsupportsimd-- enable SIMD-accelerated quantization
MIT OR Apache-2.0