A team is deploying a model that has a very spiky traffic pattern: 100 requests per minute for a few hours, then zero for several hours. They need low latency during the spikes. They are considering provisioned throughput vs. serverless. What is the most important trade-off to evaluate?

The team's preference for a specific cloud provider.

The number of tokens per request.

The cost of idle provisioned throughput versus the potential for cold starts in serverless.

A company runs a sentiment-analysis pipeline that processes millions of short social media posts daily. The pipeline is not latency-sensitive. They want to minimize cost. Which approach should they take?

Use batch inference jobs instead of real-time serving.

Increase the model's max tokens to capture more context.

Use a serverless real-time endpoint.

Use a real-time serving endpoint with provisioned throughput.

A company uses a large LLM for a real-time translation feature. They notice that many requests are for the same short phrases. They want to reduce costs while keeping response times low. Which approach is most effective?

Increase the batch size for the serving endpoint

Move to a provisioned throughput endpoint

Switch to a smaller model for all requests

Implement response caching for repeated phrases

Which Databricks feature allows you to monitor the number of requests and tokens processed by a model serving endpoint in near real-time?

Serving endpoint metrics in the Databricks UI

A company runs multiple LLM workloads on Databricks, including real-time chat, batch summarization, and a nightly ETL that uses an LLM for data cleaning. They want to optimize overall costs. Which combination of strategies is most effective?

Use serverless for all workloads.

Use the largest model for all workloads to ensure quality.

Use serverless for chat, batch inference for summarization, and a smaller model for the ETL cleaning task.

Use provisioned throughput for all workloads to ensure performance.

DatabricksCertified Generative AI Engineer Associate

Domain 6Objective 4

Cost Control and Usage Tracking GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 3)

Part of the Section 6: Evaluation and Monitoring domain, which makes up ~17% of our current practice bank. Databricks does not publish an official question count, but from its 90-minute exam (~35–60 total, ~6–10 in this domain), expect 1–2 from this objective — we provide 21 practice questions to prepare you well beyond it. (estimate)

21questions here

5free pages

3concepts

Questions 11–15

11expert · medium
A team is deploying a model that has a very spiky traffic pattern: 100 requests per minute for a few hours, then zero for several hours. They need low latency during the spikes. They are considering provisioned throughput vs. serverless. What is the most important trade-off to evaluate?
Select an answer first
12application · medium
A company runs a sentiment-analysis pipeline that processes millions of short social media posts daily. The pipeline is not latency-sensitive. They want to minimize cost. Which approach should they take?
Select an answer first
13application · medium
A company uses a large LLM for a real-time translation feature. They notice that many requests are for the same short phrases. They want to reduce costs while keeping response times low. Which approach is most effective?
Select an answer first
14foundation · easy
Which Databricks feature allows you to monitor the number of requests and tokens processed by a model serving endpoint in near real-time?
Select an answer first
15expert · medium
A company runs multiple LLM workloads on Databricks, including real-time chat, batch summarization, and a nightly ETL that uses an LLM for data cleaning. They want to optimize overall costs. Which combination of strategies is most effective?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to GENERATIVE-AI-ENGINEER-ASSOCIATE

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by Databricks. “GENERATIVE-AI-ENGINEER-ASSOCIATE” is a trademark of its owner, used for identification only.