179 questions
No questions match those filters.
← All questions
Retrieval & EmbeddingsHard· Quantization· Reranking· Memory Optimization
PQ made the index 30x smaller and recall@10 dropped six points. You add exact reranking to recover it. What just happened to your budget?
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansReranking requires full-precision vectors to be available. Consequently, the 1.09 GB index acquires a 15.36 GB companion in float16, causing the real compression ratio to fall from 30.5x to 2.0x. You must decide whether to keep the refine set resident in RAM, push it to SSD and accept the latency, or widen the PQ ‘m’ parameter to reduce the number of candidates needing refinement.