#Quantization
6 posts
BM25 straight from your payload, a tokenizer that finally understands camelCase, nested filters 7x faster, and two new TurboQuant datatypes. Here's what's actually new and when you should care.
One config line, nine times less disk, and a single knob for all your RAM decisions. Here's what's actually new and when you should care.
Millions of vectors, a fraction of the memory, and almost zero accuracy lost, here's the math behind it.
Dynamic schema updates, TurboQuant compression, better monitoring, and stronger operational safeguards for production vector search.
Because a robot shouldn't have to wait for a 5G signal to think: Decentralizing vector search for a local-first physical world
How quantization can reduce memory usage by 32x without losing accuracy using Qdrant.