Microsoft AI-300 Online Practice Questions and Exam Preparation

Microsoft AI-300 Online Questions & Answers

• Question 31:

  An older registered model version must no longer be selected for new deployments. Compliance requires that the version remain available for review.

  What should you do?

  A. Archive the model version
  B. Delete the model version
  C. Delete the workspace
  D. Move the model file to a notebook folder
  A. Archive the model version

  ---

  Explanation

  Archiving keeps an older model version available for governance while removing it from normal deployment use. Deleting or unregistering would undermine the compliance review requirement. Moving files manually does not manage the registered model lifecycle. The exam takeaway is to use lifecycle state, not deletion, when retention is required.

• Question 32:

  DRAG DROP

  A team deploys a classification model to production and monitors performance and data changes.

  The team wants to ensure that significant drops in prediction accuracy automatically trigger the following:

  1. Stakeholders must be notified of the drops.

  2. Retraining must be initiated when thresholds are exceeded.

  You need to configure monitoring to meet the requirements.

  Which four actions should you perform in sequence? To answer, move the appropriate actions from the list of actions to the answer area and arrange them in the correct order.

  Select and Place:

  

  

  ---

  The answer orders the monitoring work around baselines, thresholds, and alerts. A production model needs baseline metrics first, then threshold definitions, then alerting so stakeholders or automation can respond when accuracy or drift exceeds limits. Other actions such as manual inspection after failures are reactive and do not satisfy the requirement for automated monitoring.

• Question 33:

  A data science team completes multiple training runs within an experiment by using MLflow. The team wants to store a selected model in Azure Machine Learning so that it can be versioned and deployed later.

  The model must be versioned centrally for reuse across environments.

  You need to version the trained model.

  Which two actions should you perform? (Choose two.)

  A. Locate and capture the model artifacts from the outputs of the training run.
  B. Register the model in the Azure Machine Learning workspace.
  C. Tag the training experiment with a name.
  D. Export the model files to local storage.
  A. Locate and capture the model artifacts from the outputs of the training run.
  B. Register the model in the Azure Machine Learning workspace.

  ---

  Explanation

  The team should log or package the selected MLflow model and register it in Azure Machine Learning so it can be versioned, discovered, and deployed. Experiment runs alone are not enough for lifecycle management. Other actions such as endpoint traffic changes or data asset creation do not create the versioned model record required by the scenario.

• Question 34:

  A data science team trains a classification model that predicts loan approval outcomes. Before registering the model, the team must ensure the following:

  1. Predictions must not disproportionately impact protected groups.

  2. Prediction errors can be evaluated across different data segments.

  You need to assess whether the model meets Responsible AI expectations.

  Which two approaches should you use? (Choose two.)

  A. Analyze error rates across the global cohort.
  B. Measure endpoint latency under load.
  C. Validate inference schema compatibility.
  D. Evaluate feature importance for prediction transparency.
  E. Analyze error rates across defined demographic cohorts.
  D. Evaluate feature importance for prediction transparency.
  E. Analyze error rates across defined demographic cohorts.

  ---

  Explanation

  Responsible AI review before model registration should include fairness and error analysis so the team can detect biased outcomes and problematic behavior before production use. Those checks support the requirement for responsible evaluation. Other actions may support deployment or monitoring, but they do not directly evaluate the model against responsible AI principles before registration.

• Question 35:

  HOTSPOT

  You are reviewing a dataset that will be used for an advanced fine-tuning job in Microsoft Foundry.

  The fine-tuning job uses preference comparison data.

  You review the following dataset excerpt.

  

  For each of the following statements, select Yes if the statement is true. Otherwise, select No.

  NOTE: Each correct selection is worth one point.

  

  

  ---

  The preference dataset distinguishes records that contain a usable comparison signal from records that do not. Preference fine-tuning needs paired or ranked outputs so the training job can learn which response is preferred. Other records may contain prompts and completions, but without a clear preferred-versus-rejected signal they do not satisfy the preference-comparison requirement.

• Question 36:

  A training pipeline must read data from an existing Azure Storage account without embedding connection secrets in notebooks or scripts.

  What should you configure in Azure Machine Learning?

  A. A datastore that uses managed identity or credential-based access
  B. A model registry entry for the storage account
  C. A batch endpoint that points to the container
  D. A prompt variant that contains the storage path
  A. A datastore that uses managed identity or credential-based access

  ---

  Explanation

  A datastore gives Azure Machine Learning jobs a managed reference to external storage without embedding secrets directly in training scripts. It can use controlled credentials or managed identity depending on configuration. Other alternatives such as registries, endpoints, or prompt variants serve different lifecycle purposes and do not solve secure data access for training.

• Question 37:

  Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.

  After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.

  You manage an Azure Machine Learning workspace. The Python script named script.py reads an argument named training_data. The training_data argument specifies the path to the training data in a file named dataset1.csv.

  You plan to run the script.py Python script as a command job that trains a machine learning model. You need to provide the command to pass the path for the dataset as a parameter value when you submit the script as a training job.

  Solution: python script.py --trainingdata ${{inputs.training_data}} Does the solution meet the goal?

  A. Yes
  B. No
  A. Yes

  ---

  Explanation

  The solution meets the goal when it uses the appropriate Azure Machine Learning experiment tracking capability rather than ad hoc logs. Consistent experiment results depend on repeatable run metadata and metrics, not only on application monitoring. Other alternatives can observe systems, but they do not provide the structured experiment history required by model training workflows.

• Question 38:

  DRAG DROP

  An organization operates a generative AI application in production by using Microsoft Foundry. The application serves live user traffic and is updated by a data scientist team regularly as prompts and models evolve.

  The application intermittently times out during production use, which requires ongoing visibility into runtime behavior.

  The team must also validate model quality and safety before releasing new updates to avoid introducing regressions.

  You need to apply the correct mechanisms for continuous runtime monitoring and for release time validation.

  Which mechanisms should you use for each requirement? To answer, move the appropriate mechanisms to the correct requirements. You may use each mechanism once, more than once, or not at all. You may need to move the split bar between panes or scroll to view content.

  NOTE: Each correct selection is worth one point.

  Select and Place:

  

  

  ---

  The diagram maps runtime monitoring to observability metrics and release validation to an evaluation workflow. Observability metrics track production behavior such as latency and throughput, while evaluation workflows validate quality and safety before a new prompt or model update is released. Other mechanisms, such as diagnostic logs or blocklist updates, are narrower supporting controls.

• Question 39:

  A new model version must be introduced to a production real-time endpoint. The team wants a controlled rollout and a quick rollback path if errors increase.

  What should you configure?

  A. Traffic splitting between endpoint deployments
  B. A new datastore for the training data
  C. A single replacement deployment that receives all traffic immediately
  D. A notebook-only validation step
  A. Traffic splitting between endpoint deployments

  ---

  Explanation

  Traffic splitting between deployments supports progressive rollout because a limited percentage of production requests can reach the new model while the old deployment remains available. That gives the team a rollback path if errors increase. Other alternatives either do not affect traffic or replace production too abruptly for the requirement.

• Question 40:

  You need to isolate training workloads while remaining cost-aware to address Fabrikam Inc.'s issues, constraints, and technical requirements.

  What should you implement?

  A. Training jobs that run on a single shared compute cluster
  B. Fixed-size compute cluster
  C. Dedicated compute clusters per experiment
  D. Managed compute targets with autoscaling
  D. Managed compute targets with autoscaling

  ---

  Explanation

  Managed compute targets with autoscaling isolate training workloads while scaling down when demand drops, which controls cost. A fixed-size cluster or dedicated clusters can leave unused capacity running, and a single shared cluster weakens workload isolation. The exam rule is to match compute configuration to both isolation and cost requirements.