pgvector: Fewer dimensions are better

Summary

Embeddings support semantic search, retrieval augmented generation, and clustering in pgvector, but high-dimensional vectors increase storage, memory, and distance-computation costs as datasets grow. The post explains how IVF indexes reduce the records examined and compares embedding models using Massive Text Embedding Benchmark results, noting little correlation between dimension count and performance. In a 224,482-vector Wikipedia dataset, all-MiniLM-L6-v2 with 384 dimensions delivered 78% higher queries per second than OpenAI text-embedding-ada-002 with 1536 dimensions at accuracy@10 of 0.99, while using 4GB rather than 7.5GB of RAM. The 384-dimensional gte-small model produced more than a 200% queries-per-second improvement over all-MiniLM-L6-v2 at the same accuracy, though model choice still depends on task performance, sequence length, language support, and operational costs.

Takeaways

1. 1

   IVF indexes can reduce nearest-neighbor distance computations from the full dataset to roughly 5–20% in real-world use, but a 1-million-record dataset may still require 50,000–200,000 calculations per query.

2. 2

   MTEB results show models with fewer dimensions can perform comparably or better than text-embedding-ada-002; gte-small ranks above it while producing 384-dimensional vectors instead of 1536.

3. 3

   PCA and t-SNE may discard semantic information and become less valid when the distribution of embedded text shifts, because their learned representations are not inherently adaptable.