Cost Control and Usage Tracking GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 4)

Questions 16–20

1. 16expert · medium

   A team has multiple serving endpoints for different models. They want to identify which endpoints are underutilized and could be downsized or moved to serverless. Which data should they analyze?

   • AThe cluster logs for each endpoint.
   • BThe number of models registered in the MLflow registry.
   • CThe number of users in the workspace.
   • DToken usage and request volume per endpoint from the system tables, along with the provisioned throughput setting.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   A team wants to deploy a fine-tuned LLM on Databricks for a production workload with predictable, high-volume traffic. Which Databricks feature is specifically designed to provide reserved capacity and predictable performance for such a workload?

   • ADelta Lake storage
   • BProvisioned throughput on a model serving endpoint
   • CServerless model serving with auto-scaling
   • DDatabricks SQL warehouse
   Reveal Solution Discussion Select an answer first
4. 18expert · medium

   A company has a model serving endpoint with provisioned throughput. They are considering switching to serverless to reduce costs. They have analyzed their usage and found that the endpoint is used heavily during business hours but has very low usage at night. What is the most important factor to consider in this decision?

   • AThe latency requirements of the application.
   • BThe number of models deployed on the endpoint.
   • CThe total cost of provisioned throughput versus the estimated serverless cost based on actual token usage.
   • DThe team's familiarity with serverless technology.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   An application sends many small, independent requests to an LLM endpoint. Which strategy would most effectively reduce per-request overhead and lower overall cost?

   • AIncreasing the number of concurrent requests
   • BBatching multiple requests into a single call
   • CEnabling more detailed logging
   • DSwitching to a larger model
   Reveal Solution Discussion Select an answer first
6. 20expert · medium

   A team runs a real-time chatbot that uses a large LLM. They want to reduce costs by batching requests. However, they are concerned about the impact on user experience. What is the best way to implement batching without degrading the user experience?

   • AIncrease the batch size to the maximum possible.
   • BBatch requests only during off-peak hours when users are less sensitive to latency.
   • CUse a smaller model for all requests.
   • DBatch all requests regardless of time, accepting higher latency.
   Reveal Solution Discussion Select an answer first