問題文
A retrieval augmented generation service is being moved from a laptop prototype to a shared data center platform. Which statement best describes what this workload adds to the infrastructure compared with a plain generative service?
選択肢
- It adds a lookup step against a stored index before generation, so the platform now needs a low latency read path to that index in addition to accelerator capacity.
- It removes the need for accelerators, as the answer is copied directly out of the retrieved documents without any generation step.
- It adds a training step that must run on every request, so the platform needs the same sustained collective traffic pattern as a distributed training cluster, with the accelerators exchanging updates after every answer has been produced for a user.
- It removes the latency budget, as retrieval makes the total response time irrelevant to the user and the request can be answered whenever the index is free.