Câu đố mạch lạc

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

1. A retrieval system needs to find semantically similar passages from a large corpus. Which component supplies numerical representations for that task?
   • An embedding model
   • A reranker alone
   • A text generator
   • A tokenizer only
2. A reranker receives a candidate set and changes its order. What can it not do by itself?
   • Reorder supplied candidates
   • Recover a relevant document missing from the candidate set
   • Score candidate relevance
   • Support a retrieval pipeline
3. An assistant answer omits a policy passage. What should be checked first?
   • Only the final answer’s formatting
   • Only the generator’s tone
   • Whether the evidence entered the candidate set and was ranked highly enough
   • Only which generator brand was selected
4. What role does a generator play in the described pipeline?
   • Verify source permissions
   • Create source-language labels
   • Retrieve every document automatically
   • Produce an answer or other text
5. Why read documentation for the specific model version?
   • All models in a family are identical
   • Input limits, supported features, language behavior, or deployment terms may differ
   • A model card guarantees answer correctness
   • Third-party hosting always matches direct API behavior
6. A source passage is relevant to a question. What else should the pipeline review?
   • Whether it contains a citation-shaped phrase
   • Only its length
   • Whether the source is authorized and current
   • Whether the generator is newer
7. How should an organization test an unanswerable query?
   • Include it in evaluation and record whether the system handles it appropriately
   • Remove it from testing
   • Assume the generator will refuse
   • Replace it with a normal query
8. What should accompany model scores when evaluating a retrieval system?
   • A sample answer without measurement
   • Only parameter count
   • The API company’s age
   • Final task success and cost