Prepstellar
Get it on Google Play

AI-103 · Practice set 12 of 13

Configuring Model Deployments: 10 practice questions

10 questions · Untimed · Free

10 free AI-103 practice questions on Configuring Model Deployments, with an explanation for every answer. Untimed. The full mock exam and the timed version are in the app.

Set 12 · Configuring Model Deployments · 10 questions Read the lesson

  1. Question 1 of 10

    An application calls a model deployed in a Foundry resource. Which value in the request routes it to a specific deployment?

    1. AThe SKU code of the deployment type
    2. BThe model name shown on the catalog model card
    3. CThe deployment name in the model parameter
    4. DThe model version number set at creation
    Show the answer

    The deployment name is the address. It defaults to the model name, but you can change it, and the request carries it in the model parameter. That is why two deployments of the same model can coexist and still be told apart.

    Next → 1 / 10

  2. Question 2 of 10

    Before deploying which kind of model must you subscribe to Azure Marketplace?

    1. AAny model deployed with the Global Standard type
    2. BAny Foundry Model sold directly by Azure
    3. CA model from partners and community
    4. DAn Azure OpenAI model such as gpt-4o-mini
    Show the answer

    The Marketplace step depends on who sells the model. Partner and community models, such as Llama-4-Scout, use Azure Marketplace for licensing and price, so you subscribe first. Models sold by Azure, including Azure OpenAI, skip that step.

    Next → 2 / 10

  3. Question 3 of 10

    A developer's attempt to create a model deployment fails with an authorization error. Which role assignment fixes it for creating and managing deployments?

    1. AFoundry User on the Foundry project scope
    2. BReader on the Foundry resource's resource group
    3. CStorage Blob Data Contributor on its storage account
    4. DCognitive Services Contributor on the Foundry resource
    Show the answer

    Creating and managing deployments is a management task on the Foundry resource, and the role named for it is Cognitive Services Contributor, or equivalent permissions. Assign it at the Foundry resource where the deployment will live.

    Next → 3 / 10

  4. Question 4 of 10

    A team wants to compare two content filter configurations on the same model, side by side, in one Foundry resource. What can it do?

    1. ARedeploy the model with the same name and a new filter
    2. BCreate a second Foundry resource for the second filter
    3. CDeploy the model twice under two deployment names
    4. DAttach two filters to one single deployment
    Show the answer

    One model can back several deployments, as long as each has a different deployment name. Testing different configurations, including content filters, is exactly the use case for that, and each request picks its deployment by name.

    Next → 4 / 10

  5. Question 5 of 10

    In az cognitiveservices account deployment create, which parameter sets the deployment type, such as GlobalStandard?

    1. A--model-format
    2. B--sku-capacity
    3. C--deployment-name
    4. D--sku-name
    Show the answer

    Each deployment type is represented by a SKU code in the CLI and ARM, so the type goes in --sku-name. The neighbors are easy to mix up: --sku-capacity sets how much capacity, and --model-format names the provider.

    Next → 5 / 10

  6. Keep the ones you got wrong

    In the app, every question you miss comes back exactly when you’re about to forget it.

    Get it on Google Play

  7. Question 6 of 10

    An application uses the OpenAI v1 API against https://<resource-name>.openai.azure.com/openai/v1/. What must each request include?

    1. AThe deployment type's SKU code as a header
    2. BThe model version in the request path
    3. CThe deployment name in the model field
    4. DAn api-version query parameter
    Show the answer

    The /openai/v1/ route uses implicit versioning, so no api-version is passed. What the request does need is the deployment name in the model field, because that name routes it to the right deployment in the resource.

    Next → 6 / 10

  8. Question 7 of 10

    A subscription has used all its TPM quota for a model in East US 2. The team needs one more deployment of that model in that region. What are its two options?

    1. APick a new deployment name, or a new model version
    2. BCreate a second resource, or a new project in it
    3. CRequest more quota, or free TPM from other deployments
    4. DLower the new deployment's TPM, or deploy it twice
    Show the answer

    Quota is assigned per subscription, per region, and per model in TPM, and each deployment takes a share. At the limit, either submit a quota increase request or reduce the TPM allocated to existing deployments of that model to free tokens.

    Next → 7 / 10

  9. Question 8 of 10

    A pipeline service principal holds Cognitive Services Contributor on the Foundry resource and deploys Phi-4-mini-instruct without trouble. Deploying a Cohere model with the same principal now fails with a Marketplace subscription error. What should the admin do?

    1. AChange the deployment to the Global Standard SKU and retry
    2. BRepeat SaaS registration for the Cohere deployment
    3. CGive it Cognitive Services Contributor on the resource group
    4. DGrant the required Marketplace and SaaS permissions
    Show the answer

    A partner model needs two permission sets: Marketplace agreement actions on the Azure subscription, and SaaS resource read and write actions on the resource group. Subscription Owner or Contributor includes them; a custom role can grant the listed actions more narrowly.

    Next → 8 / 10

  10. Question 9 of 10

    A team's gpt-4o-mini quota in one region is fully allocated. A low-traffic test deployment holds a large TPM share, and policy blocks any quota increase this quarter. A new production deployment of the same model is needed in that region. What works?

    1. ARequest a quota increase through the form
    2. BLower the test deployment's TPM first
    3. CDeploy it in a new resource in that region
    4. DDeploy it under a new name with fresh quota
    Show the answer

    With no new quota allowed, the only lever is reallocation: quota is TPM per subscription, region, and model, and each deployment holds a share. Reducing the test deployment's allocation frees tokens for the production deployment.

    Next → 9 / 10

  11. Question 10 of 10

    An app calls the /openai/v1/ route with the deployment name chat-prod. The team creates a second deployment, chat-v2, in the same resource and must switch traffic without touching the endpoint URL or authentication. What changes in the request?

    1. AThe URL path gains a chat-v2 segment
    2. BThe SKU code of chat-v2 goes in a header
    3. CThe model field value becomes chat-v2
    4. DA chat-v2 api-version gets added
    Show the answer

    Every deployment in a resource sits behind the same /openai/v1/ route, and the deployment name in the model field picks the target. Switching to chat-v2 means changing that one value, with the URL and authentication untouched.

    Next → 10 / 10

  12. You’ve finished this set

    That’s 10 questions on Configuring Model Deployments. In the app the ones you miss come back exactly when you’re about to forget them.

    Back to the course Next lesson: Configuring Agent Deployments Get it on Google Play
Get it on Google Play

The whole course, on your phone

Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.

Get it on Google Play

“This app is the ‘One Lake’ for preparing the certification exams. Really worth it.”

Abdelhak Y. Google Play review