Compression

Uncompressed vectors can be large. Compressed vectors lose some information, but they use fewer resources and can be very cost effective.

To balance resource costs and system performance, consider one of these options:

• Rotational Quantization (RQ) (recommended)
• Product Quantization (PQ)
• Binary Quantization (BQ)
• Scalar Quantization (SQ)

You can also disable quantization for a collection.

Aside from quantization, Weaviate also offers encodings for multi-vector embeddings:

• MUVERA encoding