4-bit Rotational Quantization
Weaviate 1.39 adds 4-bit rotational quantization, a way to compress vectors for search. Weaviate reports similar recall with a 45% heap reduction compared with 8-bit RQ.
To enable it on a new collection, set bits to 4 in the RQ quantizer. You can also update an existing uncompressed collection, and Weaviate will re-encode its vectors in the background.
4-bit RQ requires the hnsw index. A dynamic index can use it only after converting to HNSW. Once RQ is enabled, the bits value can’t be changed.
Loading replies…
Loading replies