179 questions
No questions match those filters.
← All questions
EvaluationMedium· Benchmarks· Data Generation· Retrieval
A model performs differently on Natural Questions (NQ)...
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 benchmark’s collection protocol determines which retriever wins. SQuAD’s surface-level similarity makes it a ‘lexical matching’ task, while NQ’s paraphrase-heavy nature makes it a ‘semantic retrieval’ task. A benchmark that does not match how your users write queries will recommend the wrong retriever.