179 questions
No questions match those filters.
← All questions
Retrieval & EmbeddingsHard· Vector Databases· Memory Optimization· Quantization
Your embeddings are float32 at d = 768 and the index no...
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 plansA strong answer begins by calculating the footprint: 3,072 bytes per vector, which equals 30.72 GB for 10 million vectors. You should then separate the two axes of optimization: bits per dimension and the number of dimensions. It is critical to take the free 4x reduction from SQ8 before discussing any lossy methods. A weak answer proposes using a smaller embedding model, which conflates dimensionality reduction with quantization and pays in recall for a problem that a storage format can solve at no cost.