Why Vector Size Matters
We’re so glad you’re here. You can expect all the best TNS content to arrive Monday through Friday to keep you on top of the news and at the top of your game.
Vector indexes are the hottest topic in databases because approximate nearest neighbor (ANN) vector search puts the “R” in RAG (retrieval-augmented generation). “Nearest neighbor” for text embedding models is almost always measured with angular distance — for instance, the cosine between two vectors. Getting the retrieval accurate and efficient is a critical factor for the entire application; failing to find relevant context — or taking too long to find it — will leave your large language model (LLM) prone to hallucination and your users frustrated.
Every general-purpose ANN index is built on a graph structure. This is because graph-based indexes allow for incremental updates, good recall and low-latency queries. (The one exception was pgvector, which started with a partition-based index, but its creators switched to a graph approach as fast as they could because the partitioning approach was far too slow.)
