Manage Model Lifecycle and Governance GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   An organization has a model serving endpoint that is used by both internal data scientists and external partners. The internal team needs full access to update the endpoint, while partners should only be able to query it. The model is registered in Unity Catalog. What is the most secure way to configure access?

   • AGrant the internal team CAN_MANAGE and the partners CAN_QUERY on the serving endpoint.
   • BGrant both groups CAN_QUERY and rely on the model's schema permissions to restrict management.
   • CGrant the partners CAN_MANAGE and the internal team CAN_QUERY, then use a service principal for management.
   • DCreate two separate endpoints, one for each group, with the same model.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard · select all that apply

   A team is developing an agent that uses a memory store and a language model. They want to implement component testing to ensure reliability. Which of the following are appropriate component tests? Select all that apply.

   • ATest the deployment pipeline's ability to deploy the agent.
   • BTest the retrieval component's ability to return relevant documents.
   • CTest that the memory store correctly saves and retrieves data.
   • DTest the language model's response quality with a set of sample prompts.
   • ETest the entire agent's response to a user query end-to-end.
   Check answers Discussion Select an answer first
4. 8expert · hard

   A prompt has been in production for a while, but a new version is being developed. The team wants to test the new version with a small percentage of users before fully rolling it out. They also need to be able to revert to the old version if issues arise. What is the best approach?

   • ADeploy the new prompt to all users immediately and monitor for issues.
   • BUse a canary deployment strategy, where a small percentage of traffic is routed to the new prompt version, and monitor for issues.
   • CUse a blue-green deployment, where the new prompt is deployed to a separate environment and then all traffic is switched at once.
   • DKeep the old prompt and the new prompt in the same environment and let users choose.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   An agent application needs to store conversation history for each user to provide personalized responses. The team is deciding between using a Delta table and a Vector Search index for this purpose. The history is structured and needs to be updated frequently. Which storage solution is more appropriate?

   • AStore the history in MLflow artifacts for each user.
   • BUse a Delta table to store the conversation history, as it supports frequent updates and structured queries.
   • CUse a Vector Search index to store the conversation history, as it supports semantic search.
   • DStore the history in a Python dictionary and serialize it to a file after each session.
   Reveal Solution Discussion Select an answer first
6. 10foundation · easy

   What information does a version control system like Git provide for a prompt file?

   • AA complete history of changes, including who made them and when
   • BThe current performance metrics of the prompt in production
   • CThe cost of running the prompt in the production environment
   • DThe list of users who have queried the model using this prompt
   Reveal Solution Discussion Select an answer first