AI-300 Practice Test 14 — 15 Questions

Question 1

When sharing an AI project with a team, which resource group architecture is best practice for lifecycle management?

Accepted Answer

Keep the project and its dependencies in a single resource group.. Keeping an AI project and its dependencies (model deployments, prompt flows, connections, storage, and search resources) in a single resource group is best practice because a resource group is the unit of lifecycle management in Azure. Deleting or exporting the resource group removes or captures the entire project atomically, RBAC and cost tracking are simplified, and deployment templates can be scoped to that group. This aligns resource boundaries with project boundaries.

Answer

Use a public resource group for all team members.

Answer

Spread resources across multiple regions.

Answer

Place each component in a different subscription.

Question 2

Where do you define the model deployment settings, including instance count and version?

Accepted Answer

Azure AI Foundry -> Project -> Deployments. Model deployment settings — including the deployed model version, instance count (capacity), deployment type, and rate limits — are configured in Azure AI Foundry under the project's Deployments blade. This is where you create and manage deployments that expose a model endpoint for inference, and where you scale capacity and pin versions. It is the operational control plane for serving models.

Answer

Azure Portal -> Subscriptions

Answer

Azure Key Vault -> Secrets

Answer

Azure Monitor -> Alerts

Question 3

Which TWO actions can you perform in the Azure Machine Learning studio?

Accepted Answer

View run history.. Option C (View run history) is correct because the Azure Machine Learning studio provides an Experiments and Runs interface where you can inspect each run's metrics, logs, and outputs for tracking and comparison. Option E (Register a new dataset) is correct because the studio's Data section (Datastores/Datasets) lets you create and register datasets from datastores, local files, or web URLs for reuse in experiments and pipelines. Options A (Modify global azure subscriptions) and D (Configure networking in the tenant) are incorrect because subscription-level and tenant-wide networking settings are managed in the Azure portal or Microsoft Entra admin center, not in the ML studio. Option B (Manage office 365 users) is incorrect because Office 365 user administration is handled through the Microsoft 365 admin center, not Azure Machine Learning studio.

Answer

Modify global azure subscriptions.

Answer

Manage office 365 users.

Answer

Configure networking in the tenant.

Question 4

You are troubleshooting a deployment. Which THREE logs are most helpful to check?

Accepted Answer

Deployment status logs.. Deployment status logs (A) are essential because they show the orchestration-level events and errors from the deployment controller, such as failed rollouts, image pull errors, or readiness probe failures, directly indicating why the deployment did not succeed. Application output logs (B), typically captured from stdout/stderr, reveal runtime errors, exceptions, and startup failures inside the deployed containers that would not appear in orchestration logs. Container build logs (D) are critical when the failure originates during image construction, showing failed build steps, dependency errors, or Dockerfile issues before the image ever reaches the deployment stage. User browser logs (C) reflect client-side rendering or network issues and are not authoritative for server-side deployment failures, while Azure billing logs (E) only track cost and usage data and provide no diagnostic value for a failed deployment.

Answer

User browser logs.

Answer

Azure billing logs.

Question 5

Which THREE tools in Azure can be used to monitor the health and performance of your GenAI infrastructure?

Accepted Answer

Azure Log Analytics. Azure Log Analytics (B) is correct because it is the workspace-based service that ingests and queries telemetry such as logs and metrics from GenAI resources using Kusto Query Language (KQL), enabling health and performance analysis. Azure Application Insights (C) is correct because it provides application performance monitoring (APM) for GenAI apps, capturing request rates, latency, failures, dependency calls to model endpoints, and distributed traces. Azure Monitor (D) is correct because it is the umbrella monitoring platform that collects metrics, activity logs, and diagnostics, and drives alerts, dashboards, and autoscale for GenAI infrastructure. Azure Front Door (A) is a global HTTP load balancer and CDN for web traffic, not a health/performance monitoring tool, and Microsoft Entra ID (E) is an identity and access management service, so neither belongs here.

Answer

Azure Front Door

Answer

Azure Active Directory

Question 6

Your Azure OpenAI deployment is experiencing high latency during peak hours. You observe that input tokens are consistently high. Which strategy is most effective for reducing latency while maintaining quality?

Accepted Answer

Implement prompt caching for static instructions.. Prompt caching stores the KV cache for static prompt prefixes (system instructions, few-shot examples) so the model doesn't recompute attention over those tokens on every request. When input tokens are consistently high and dominated by static content, caching eliminates redundant prefill computation, directly reducing time-to-first-token and overall latency without changing output quality.

Answer

Increase the top_p parameter.

Answer

Enable dynamic scaling in Azure App Service.

Answer

Decrease the number of parallel requests.

Answer

Use a higher model version (e.g., o1-preview).

Question 7

Which TWO of the following strategies are commonly used to optimize for cost in Generative AI?

Accepted Answer

Selecting the smallest capable model for the task.. Option A is correct because choosing the smallest capable model for a task directly reduces cost, since inference pricing scales with model size and capability—smaller models consume fewer compute resources per token. Option B is correct because caching frequent prompts or responses avoids redundant model invocations, so repeated queries are served from cache instead of incurring new token-generation charges. Option C is not a cost optimization: expanding the context window to 128k increases the number of input tokens processed per request, which typically raises cost. Option D is not a cost optimization: verbose logging increases storage and observability overhead without reducing inference spend. Option E is not a cost optimization: using the highest possible temperature affects output randomness, not token pricing, and can even increase retries or output length.

Answer

Increasing the context window size to 128k.

Answer

Enabling verbose logging.

Answer

Using the highest possible temperature.

Question 8

You need to update a prompt flow without downtime. What deployment strategy should you use?

Accepted Answer

Use a Blue-Green deployment pattern.. Blue-Green deployment is the correct strategy because it involves running two identical environments (blue and green) where the new version is deployed to the inactive environment, tested, and then traffic is switched over. This allows updates without downtime and provides a quick rollback if issues arise. It is a standard pattern for zero-downtime deployments in cloud-native applications.

Answer

Directly overwrite the existing deployment.

Answer

Stop the service during the update.

Answer

Delete the old deployment and create a new one.

Question 9

You are evaluating an LLM application using Prompt Flow. You want to measure the 'Groundedness' of the model response relative to the retrieved context. Which evaluator should you configure?

Accepted Answer

Groundedness evaluator. The Groundedness evaluator in Prompt Flow checks if the response is supported by the context provided.

Answer

Relevance evaluator

Answer

Coherence evaluator

Answer

Fluency evaluator

Question 10

You are using GitHub Actions. You want to authenticate to Azure without using hardcoded credentials. What should you use?

Accepted Answer

OIDC. OIDC (OpenID Connect) is the secure way for GitHub Actions to authenticate to Azure.

Answer

SSH keys

Answer

Azure Storage keys

Answer

User name and password

Question 11

You need to monitor the health of your deployed Azure ML models. Which service is integrated directly with Azure ML to provide automated metrics and logging?

Accepted Answer

Azure Application Insights. Azure Application Insights is the native service for monitoring logs, metrics, and telemetry from deployed model endpoints.

Answer

Azure Event Grid

Answer

Azure Policy

Answer

Azure Data Factory

Question 12

When configuring a compute instance, which setting allows you to automatically stop the instance during periods of inactivity?

Accepted Answer

Idle shutdown. Idle shutdown automatically turns off the compute instance when no active processes are detected, saving costs.

Answer

Provisioning timeout

Answer

Auto-scaling

Answer

Priority based scheduling

Question 13

You are creating a CI/CD pipeline in Azure DevOps. Which extension is essential to integrate Azure ML tasks into your build/release pipeline?

Accepted Answer

Azure Machine Learning extension. The Azure Machine Learning extension for Azure DevOps installs the pipeline tasks (such as AzureML job, model registration, and pipeline tasks) that let a build or release pipeline invoke Azure ML workspaces. Without this extension, the hosted agent has no task definitions to call Azure ML CLI v2 or SDK operations, so CI/CD integration with Azure ML cannot be orchestrated.

Answer

Visual Studio Code extension

Answer

Terraform extension

Answer

Kubernetes extension

Question 14

Which THREE metrics are critical for monitoring a production Generative AI system?

Accepted Answer

Latency (Time-to-first-token).. Option A (Latency, specifically Time-to-first-token) is critical because in a production Generative AI system, TTFT directly measures the perceived responsiveness of streaming responses, and high TTFT degrades user experience and can indicate model, network, or inference bottlenecks. Option D (Token usage per request) is critical because tokens drive both cost and throughput in LLM serving; tracking prompt and completion tokens per request enables accurate chargeback, quota enforcement, and capacity planning against rate limits such as tokens-per-minute. Option E (Error rates, HTTP 4xx/5xx) is critical because 4xx codes reveal client-side issues like authentication failures, throttling (429), or malformed prompts, while 5xx codes expose server-side faults such as model timeouts or backend unavailability, making them essential for reliability and SLO monitoring. Option B (hardware temperature of the Azure server) is not a critical application-level metric for a Generative AI system because Azure abstracts physical hardware and exposes health through platform metrics rather than requiring per-server temperature tracking. Option C (model training loss) is not appropriate here because it is an offline training-time metric and does not reflect the runtime health, cost, or reliability of a deployed production inference system.

Answer

The hardware temperature of the Azure server.

Answer

Model training loss.

Question 15

You are configuring a 'Managed Online Endpoint' for production. You want to ensure high availability. What should you configure?

Accepted Answer

Set the 'instance_count' to at least 2.. To ensure high availability, you should configure the 'instance_count' to be greater than 1, spanning multiple availability zones if supported.

Answer

Use 'LowPriority' compute.

Answer

Set the 'instance_count' to 1.

Answer

Enable 'Auto-scaling' only.