Determine the Requirements Needed for Inferencing CERTIFIED-AI-EXPERT Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A real-time translation app runs a neural machine translation model on a GPU server. The model currently has a p95 latency of 300 ms, but the app requires under 100 ms. The team has already quantized the model to INT8, but latency is still too high. Which additional optimization should they apply?

   • ADeploy the model to a cloud GPU with higher clock speed.
   • BUse a smaller, distilled model that has lower latency.
   • CMove the model to a CPU with more cores.
   • DIncrease the batch size to improve GPU utilization and reduce per-request latency.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A bank has deployed a loan-approval model that must be audited for fairness and accuracy. The model's performance is monitored, but the team needs to ensure that any degradation is caught quickly and that the model remains compliant with regulations. Which monitoring practice should they implement?

   • ATrack accuracy and fairness metrics over time and set alerts for significant deviations.
   • BRetrain the model every month to ensure it stays current.
   • CReview model predictions manually on a quarterly basis.
   • DMonitor server CPU and memory usage to ensure the model is running efficiently.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A healthcare startup is deploying an image-classification model for radiology images. The application experiences unpredictable spikes in demand, with some periods of near-zero traffic. The team needs to keep costs low while ensuring that inference latency stays under 2 seconds. Which deployment strategy best meets these requirements?

   • AUse a single cloud VM with a powerful GPU and manually scale up when traffic increases.
   • BProvision a fixed number of on-premises GPU servers to handle peak load at all times.
   • CDeploy the model to a serverless inference platform that scales to zero during idle periods.
   • DDeploy the model on edge devices at each hospital to avoid cloud latency.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   Which optimization technique reduces the precision of model weights to lower memory usage and speed up inference?

   • AQuantization
   • BPruning
   • CDistillation
   • DBatching
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A team is deploying a large language model for a document-summarization service. The model has 7 billion parameters and requires 16 GB of GPU memory in FP16. The team has access to GPU nodes with 24 GB of memory. They need to serve multiple concurrent requests and want to maximize throughput. Which action should they take?

   • AIncrease the batch size to fill the GPU memory, even if it causes out-of-memory errors.
   • BDeploy the model on a single GPU and rely on the operating system's swap space to handle memory overflow.
   • CQuantize the model to INT8, reducing memory footprint to about 8 GB, and deploy on a single GPU.
   • DUse model parallelism to split the model across multiple GPUs, even though each GPU has enough memory.
   Reveal Solution Discussion Select an answer first