Weaviate: 8-bit Rotational Quantization achieves 4x compression
AI Impact Summary
Weaviate has released a new 8-bit Rotational Quantization algorithm that achieves 4x vector compression while maintaining comparable search speeds and quality. This technique leverages random rotations and scalar quantization to reduce memory footprint and accelerate distance computations, particularly beneficial for large vector embeddings like those from OpenAI (1536 or 3072 dimensions). The algorithm’s robustness and speed-quality tradeoff make it a superior default for Weaviate users compared to uncompressed vectors or other quantization methods.
Affected Systems
- Date
- Date not specified
- Change type
- capability
- Severity
- info