179 questions
No questions match those filters.
← All questions
RAGVery Hard· End-to-End Training· Vector Indexing· Data Freshness
Research wants joint end-to-end training. Platform poin...
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 plansThe decision hinges on the operational cost of index maintenance. With the encoder (phi) frozen, a 5% weekly turnover requires encoding only 500,000 new chunks, which is a manageable task taking approximately 83 seconds. However, if you proceed with joint end-to-end training, the moment phi updates, every vector in the existing index becomes stale. This necessitates a full re-encoding of the entire corpus, which is computationally prohibitive compared to the incremental updates possible with a frozen encoder.