What compression formats (binary, scalar quant) are available in VDB?

In vector databases (VDB), two main types of compression formats are often used for vector compression:

1. Binary Compression — compression where vectors are transformed into binary codes. This significantly reduces data size and speeds up search by using bitwise operations. An example is Product Quantization (PQ), where vectors are divided into sub-vectors, each encoded as an index from a limited set of centroids.

2. Scalar Quantization — scalar quantization, where each element of the vector is rounded or encoded with less precision (for example, from 32-bit float to 8-bit integer). This reduces the memory volume needed for storage with some loss of accuracy.

Both approaches are often combined to achieve a balance between search speed, memory volume, and result accuracy. Modern VDBs may also use other methods, such as Product Quantization, Optimized Product Quantization, Residual Quantization, etc., which are based on the ideas of binary and scalar quantization.