AI-300 Practice Exam 3 — 20 Questions

Question 1

You are managing a multi-workspace environment. You need to restrict the ability to create new compute clusters to only specific users. Where should you apply this control?

Accepted Answer

Azure RBAC on the Workspace. RBAC (Role-Based Access Control) at the workspace level determines which users can interact with compute resources.

Answer

Network Security Group

Answer

Azure Storage Access Policies

Answer

Azure ML Datastore configuration

Question 2

In Azure ML, what is a 'Component'?

Accepted Answer

A reusable pipeline step. A component is a reusable, versioned building block for an ML pipeline.

Answer

A storage account

Answer

A user role

Answer

A physical server

Question 3

Which Azure feature is used to define infrastructure as code for your Azure ML workspace?

Accepted Answer

Azure Bicep. Azure Bicep is the recommended tool for defining and deploying Azure resources as code.

Answer

Azure Portal

Answer

PowerShell

Answer

Azure CLI

Question 4

You want to automate the testing of your prompt flows as part of your CI/CD pipeline. Which tool should you use?

Accepted Answer

Prompt Flow CLI. The 'pf' (Prompt Flow) CLI is designed to run flows and evaluations programmatically in CI/CD pipelines.

Answer

Azure Logic Apps

Answer

Azure Functions

Answer

Azure AI Search test console

Question 5

Which THREE steps are necessary to successfully fine-tune an Azure OpenAI model?

Accepted Answer

Prepare and upload a training dataset in JSONL format.. Option B is correct because Azure OpenAI fine-tuning requires a training dataset uploaded as a JSONL file with chat-completion formatted examples (messages with role/content pairs), which the service uses to adapt the model. Option D is correct because you must choose a base model that supports fine-tuning (for example gpt-35-turbo, gpt-4o-mini, or babbage-002/davinci-002) and that is available in the region where you create the fine-tune job. Option E is correct because after submitting the fine-tuning job you track its status and metrics (training loss, token usage, and completion state) via the Azure OpenAI Studio or the REST API, and you only deploy the resulting model once the job succeeds. Option A is not needed because Azure OpenAI manages the weight updates internally; you never edit model weights in code. Option C is not required because content filtering is a separate safety configuration and disabling the input filter is neither necessary nor recommended for fine-tuning.

Answer

Manually update the model weights in the code.

Answer

Disable the input filter in the portal.

Question 6

When implementing LLM-as-a-judge for evaluation, which TWO factors can influence the reliability of your results?

Accepted Answer

The specific prompt instructions provided to the judge LLM. Option B is correct because the prompt instructions given to the judge LLM directly shape its evaluation criteria, rubric interpretation, and output format, so vague or biased instructions produce inconsistent, unreliable judgments. Option E is correct because the judge LLM's own capability—its reasoning ability, context window, training quality, and susceptibility to position or verbosity bias—bounds how accurately it can assess responses, making a weaker judge less reliable than a stronger one. Options A, C, and D do not belong: dashboard color is purely cosmetic, workspace token quota affects capacity/cost rather than judgment reliability, and storage account location affects latency, residency, and compliance but not the judge's evaluation quality.

Answer

The color of the application dashboard

Answer

The number of tokens available in the workspace quota

Answer

The location of the storage account

Question 7

A data scientist needs to access data in an Azure Data Lake Storage Gen2 account from Azure ML. You need to ensure the workspace uses the most secure method to access this data. What should you configure?

Accepted Answer

User-Assigned Managed Identity. A User-Assigned Managed Identity lets the Azure ML workspace authenticate to ADLS Gen2 via Microsoft Entra ID without storing credentials, and it can be scoped to specific resources with least-privilege RBAC (e.g., Storage Blob Data Reader). This is Microsoft's recommended, most secure access pattern for workspace-to-storage connectivity.

Answer

Shared Key

Answer

Public Access

Answer

Account SAS

Question 8

You need to monitor the 'latency' and 'token usage' of your model deployments in real-time. Which tool provides the most granular view?

Accepted Answer

Azure Monitor with diagnostic logs. Azure Monitor with diagnostic logs is the correct choice because it captures granular, per-request telemetry from Azure OpenAI and Azure AI model deployments, including latency, token usage (prompt/completion tokens), HTTP status codes, and model-level metrics. Diagnostic settings route these logs to Log Analytics, where KQL queries can slice latency and token consumption by deployment, model, and time window in near real-time. This is the only option that exposes token-level metrics rather than aggregate cost or activity metadata.

Answer

Azure Billing dashboard

Answer

Azure Activity Log

Answer

Azure AI Foundry 'Overview' tab

Question 9

Which THREE security features are essential for a production Azure ML deployment?

Accepted Answer

Key Vault integration.. Key Vault integration (A) is essential because it securely stores and centrally manages secrets, keys, and certificates used by the Azure ML workspace, compute, and endpoints, keeping credentials out of code and configuration files. Managed Identity (B) is essential because it gives the workspace and its compute resources an automatically managed identity in Microsoft Entra ID, enabling passwordless, least-privilege access to Key Vault, Storage, and ACR without embedding credentials. Virtual Network (VNet) (D) is essential because it provides network isolation for the workspace, compute clusters/instances, and private endpoints, restricting traffic to approved subnets and preventing exposure over the public internet. Opening inbound ports in the NSG (C) is not a security feature but a weakening of the perimeter, since production workloads should minimize inbound exposure and instead use private endpoints or Azure Bastion/JIT access. Public IP exposure (E) is not a security feature either, as assigning public IPs increases the attack surface and contradicts the private-networking posture expected in production.

Answer

Open inbound ports in NSG.

Answer

Public IP exposure.

Question 10

Your Azure ML training job fails with an 'Out of Memory' error on a GPU cluster. You need to ensure the job runs successfully without modifying the model code. What should you do?

Accepted Answer

Change the VM size to one with higher memory. The 'Out of Memory' error on a GPU cluster indicates the VM size has insufficient GPU memory (VRAM) for the training job. Changing to a VM size with higher memory (e.g., from NC6 to NC12 or ND series) directly addresses the root cause without modifying model code. Azure ML supports resizing the compute cluster by selecting a different VM SKU, which provides more GPU memory and allows the job to run successfully.

Answer

Increase the idle time for the compute cluster

Answer

Enable auto-scale to zero

Answer

Update the workspace storage account

Question 11

You are deploying a model via a Managed Online Endpoint. You want to implement a Canary deployment where 10% of traffic is sent to a new model version. How do you configure this?

Accepted Answer

Configure the 'traffic' parameter in the 'az ml online-endpoint update' command.. You use the 'update' command on the endpoint to set the 'traffic' parameter, distributing weights across deployment names.

Answer

Modify the 'inference_config.json' file to include weights.

Answer

Create two endpoints and use an Azure Load Balancer.

Answer

Set the 'replica_count' to 10% of total capacity.

Question 12

You have an automated model training pipeline that is failing intermittently due to compute availability. What should you configure to ensure the pipeline is more resilient?

Accepted Answer

Configure 'retry' settings for the step.. You should configure 'retry' settings in the 'PipelineStep' definition to handle transient compute errors.

Answer

Increase the 'max_nodes' of the cluster.

Answer

Set the 'allow_reuse' parameter to True.

Answer

Use 'LowPriority' compute.

Question 13

What is the recommended approach for managing configuration settings for different environments (Dev, Test, Prod) in an AI application?

Accepted Answer

Use 'App Configuration' service to manage environment-specific variables.. Using environment variables or app configuration services is the standard way to inject configuration without hardcoding.

Answer

Deploy different model versions for each environment.

Answer

Create separate Azure subscriptions for every environment.

Answer

Hardcode values in the application code.

Question 14

Which technique is most appropriate for optimizing RAG performance when the vector database returns too much noisy information?

Accepted Answer

Implementing a re-ranking stage.. A re-ranking stage applies a more expensive cross-encoder or semantic ranker to the initial candidate set returned by the vector database, reordering results by true relevance to the query. This filters out semantically similar but contextually irrelevant chunks that vector similarity alone cannot distinguish, directly addressing the noisy-retrieval problem. It is the standard second-stage pattern in production RAG pipelines.

Answer

Changing the embedding model to a smaller one.

Answer

Increasing the number of chunks retrieved.

Answer

Reducing the temperature to 0.

Question 15

You are using Azure AI Studio to evaluate your model. Which tool allows you to perform batch testing on a large dataset of prompts?

Accepted Answer

Evaluation tab. The Evaluation tab in Azure AI Studio (now Azure AI Foundry) provides batch evaluation capabilities, allowing you to run a model or prompt flow against a large dataset of prompts and compute metrics like groundedness, relevance, and fluency. This is the designated tool for batch testing at scale, as opposed to interactive testing in the playground.

Answer

Prompt Playground

Answer

Model catalog

Answer

Deployment logs

Question 16

You are implementing a custom container for model training. You need to push the image to the 'Azure Container Registry' (ACR) linked to your workspace. What is the correct command?

Accepted Answer

'docker build' and 'docker push' to the ACR URL.. The standard approach is 'docker build' followed by 'docker push' to the ACR URL, ensuring authentication via 'az acr login'.

Answer

'az ml environment create --custom'.

Answer

'az container create'.

Answer

'az ml model deploy --image'.

Question 17

You are building an Azure Machine Learning pipeline. You need to ensure that the pipeline components are reusable and versioned independently. What should you use?

Accepted Answer

Azure ML Components. Azure ML components allow for independent versioning and reusability across different pipelines.

Answer

Azure ML Datasets

Answer

Azure ML Pipelines Steps

Answer

Azure Container Registry images

Question 18

Which TWO of the following are benefits of using Prompt Flow for orchestration?

Accepted Answer

Integrated testing and evaluation.. Option B is correct because Prompt Flow provides built-in tooling for testing and evaluating flows, including the ability to run batch evaluations against datasets and compute metrics such as groundedness, relevance, and coherence, which supports iterative quality improvement. Option E is correct because Prompt Flow offers a visual designer (in Azure AI Foundry / Azure Machine Learning) where you can author and connect nodes—LLM, Python, prompt, and tool nodes—into directed acyclic graphs, making complex orchestration easier to design and understand. The unmarked options do not belong: A is wrong because Prompt Flow orchestrates calls to LLMs rather than removing the need for one, C is wrong because authentication and authorization are still required for accessing models and connected resources, and D is wrong because Prompt Flow does not automatically manage hardware—compute provisioning and scaling remain the user's responsibility.

Answer

Removal of the need for an LLM.

Answer

No need for authentication.

Answer

Automatic hardware management.

Question 19

What is the benefit of using 'Azure Machine Learning Datasets' (or Data Assets) over raw storage paths?

Accepted Answer

They provide versioning and lineage tracking.. Data Assets allow for versioning, lineage tracking, and simplified usage across different compute targets.

Answer

They convert data to Parquet format.

Answer

They automatically remove duplicate records.

Answer

They are faster to read.

Question 20

What is the primary benefit of using Azure ML compute clusters instead of compute instances for training?

Accepted Answer

Support for multiple nodes. Compute clusters are designed for distributed, scalable training across multiple nodes.

Answer

Clusters have more RAM per node

Answer

Instances are deprecated

Answer

Cost is always lower