Module embed_compression 

Embedding Compression: Quantization and Product Quantization for efficient storage and approximate nearest-neighbor search.

Provides:

• QuantizedEmbedding — 8-bit scalar quantization via min-max
• EmbeddingQuantizer — batch quantization for 4-bit and 8-bit
• ProductQuantizer — product quantization with per-subspace k-means

Structs

EmbeddingQuantizer

Batch quantizer supporting 4-bit and 8-bit precision.

ProductQuantizer

Product quantizer: divides the embedding into subspaces and learns a codebook per subspace for efficient approximate nearest-neighbor search.

QuantizedEmbedding

A single embedding quantized to 8-bit precision via min-max scaling.