1. Which option is not a metric used to assess models in the Azure AI catalogue?
AAI quality
BEstimated cost
CThroughput
DModel size
Correct! Although model size can correlate to improved quality, It is not always the case that larger models are more accurate.
Incorrect: AI quality, estimated cost, Throughput are all metrics used in AI foundry to evaluate models.
True or false: Recall augmented generation (RAG) compares your prompt with your documents to change the way the model was programmed?
ATrue
DFlase
Correct! RAG does not change the way the model functions but instead adds the context of your documents to the prompt.
Incorrect. RAG does not change the way the model functions but instead adds the context of your documents to the prompt.
3. When deploying a standard model, what measure is used to measure the size of your deployment and how much of your quota is used?
ATokens per minute
BEnqueued token
CTotal Token quota
DProvisioned Throughput Units
Correct! Standard deployments are measured in tokens per minute. Different subscriptions provide you with different amounts of this quota, allowing you to customise how many requests your model can handle.
Incorrect. Although enqueued tokens and provisioned throughput units are real used to gage the size of the model, they are used to measure batched deployments and provisioned deployments respectively. There are no quotas set on the total number of tokens used as each token is charged to the credit on your account.
4. When designing a deployment with AI foundry, what should you do when you wish to upload sensitive data?
AUpload the data whenever, Microsoft’s security will take on the risk
BNever upload sensitive data
CConsult your someone at your institution or Microsoft before
Correct! when handling sensitive data, it’s always important to consult your institution and Microsoft representatives if needed .
Incorrect. When using sensitive data, it is extremely important that you consult your institutions IT team and relevant departments. If you are still unsure, you should consult the team at Microsoft to find a solution.
5. Which one of these statements is true for serverless API deployments rather than a managed compute?
AAccess to more models than managed compute deployments
BServerless API is more secure that managed compute
CYou are solely responsible for all data security provisions in serverless API deployments
DServerless API is more expensive
Correct! By default, serverless API is more secure, as Microsoft can guarantee that models will not learn from your data, and any data you upload will only stay in your Microsoft azure account’s storage.
Incorrect. Although serverless API deployments have access to less models, they are more secure and cheaper than managed compute deployments.
