medium10 pts · first answer countsragvector-databasesquantization
Binary quantisation with rescoring
You turn on binary quantisation (1 bit per dimension) for 1024-dimension vectors to cut memory 32×. Recall drops. What is the standard way to recover most of it?
Discussion 0
Join free to post in the discussion.
Related questions
- What RAG doesAI Engineer (LLM & Generative AI)
- What a vector database doesAI Engineer (LLM & Generative AI)
- What semantic search adds over keywordsAI Engineer (LLM & Generative AI)
- What an embedding dimension isAI Engineer (LLM & Generative AI)
- What top-k means in retrievalAI Engineer (LLM & Generative AI)
- Why vector search is approximateAI Engineer (LLM & Generative AI)
- When a flat index is fineAI Engineer (LLM & Generative AI)
- Why store metadata with each chunkAI Engineer (LLM & Generative AI)