Module 3: Vector Quantization Techniques

This notebook demonstrates how to use scalar and binary quantization with multi-vector collections in Qdrant.

Install the Qdrant client and fastembed libraries.

Load the ColPali vision-language model for generating multi-vector embeddings from document images.

Embed sample document images. Each image produces 1030 vectors of 128 dimensions - 1024 image patches plus 6 instruction tokens.

Scalar Quantization

Scalar quantization converts float32 values to 8-bit integers (uint8), reducing memory by 4x.

Binary Quantization

Binary quantization represents each component as a single bit (positive/negative), achieving 32x compression.

Search with Rescoring

Qdrant provides automatic rescoring: the quantized index quickly finds candidates, then re-ranks them using the original float32 vectors for accuracy.

Impact of Rescoring

Let's compare results with and without rescoring to see the impact on result quality.