Determine the Requirements Needed for Inferencing CERTIFIED-AI-EXPERT Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A ride-sharing company deploys a route-optimization model that receives a burst of requests during rush hour. The model currently runs on a fixed cluster of 10 GPU servers, each handling 50 requests per second. During peak, the demand reaches 1,000 requests per second, and the p95 latency must stay under 200 ms. The team has a limited budget and cannot add more GPU servers permanently. Which strategy should they use?

   • AUse a serverless inference platform that scales automatically during peak hours.
   • BPre-provision additional GPU servers and keep them idle for peak periods.
   • CIncrease the batch size on the existing GPU servers to improve throughput.
   • DApply quantization to reduce model size and speed up inference on the existing cluster.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   A healthcare organization must keep patient data within its own facilities due to strict data privacy regulations. Which deployment environment best meets this requirement?

   • APublic cloud
   • BOn-premises data center
   • CHybrid cloud
   • DEdge devices
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A startup is developing a voice-assistant application that runs on users' smartphones. The speech-recognition model must run entirely on-device to protect user privacy. The team has a model that is 500 MB in size and requires 2 GB of RAM, but the target smartphones have only 1 GB of available RAM. Which approach should the team take?

   • AIncrease the model's batch size to improve efficiency and reduce memory usage.
   • BUse a larger model with higher accuracy, assuming the OS will manage memory.
   • CApply pruning and quantization to reduce the model's size and memory footprint.
   • DMove the model to a cloud server and stream audio for processing.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A music-streaming service uses a model to generate personalized playlists. The model is invoked when a user opens the app, and the response must be delivered within 1 second. The service experiences 10x traffic spikes during new album releases. Which combination of requirements should the team define?

   • AA p95 latency target of 1 second and the ability to scale horizontally to handle spikes.
   • BA p99 latency target of 1 second and a fixed number of servers.
   • CA throughput target of 100 requests per second and no latency requirement.
   • DA batch processing approach that processes requests every 5 seconds.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A company is deploying a natural-language-processing model for customer-service chatbots. The model has 1 billion parameters and runs on a GPU server. The team wants to reduce inference cost without significantly affecting accuracy. Which optimization should they apply?

   • AApply knowledge distillation to create a smaller student model.
   • BUse a larger GPU with more memory to allow higher batch sizes.
   • CDeploy the model on a CPU-only server to reduce GPU costs.
   • DIncrease the model's batch size to improve GPU utilization.
   Reveal Solution Discussion Select an answer first