PMLE practice questions 351–360

1. Question 351 · 364 in bank

   You are developing an automated training workflow using Agent Platform Pipelines. The pipeline trains a new custom model and registers it to Model Registry. You need to validate the model’s performance on a held-out test dataset. You want to ensure that the model is deployed to a production Agent Platform endpoint only if its evaluation metrics meet a specific threshold. You want to minimize cost and administrative overhead. What should you do?

   • AConfigure the training component to calculate evaluation metrics on the training dataset during the training job. Output the metrics to the next step and use a conditional task to deploy the model if the evaluation metrics exceed the specified threshold.
   • BInsert a Python component that pulls the latest model version from Model Registry. Compare its training metrics against the current production model, and use a conditional task to deploy the new model if it performs better.
   • CDeploy the model to a temporary Agent Platform endpoint. Use a custom component to send the test dataset for online prediction, calculate the metrics, and deploy to the production endpoint if the evaluation metrics exceed the specified threshold.
   • DAdd a batch prediction component to generate predictions on the test dataset. Pass the results to a model evaluation component. Use a conditional task to deploy the model if the evaluation metrics exceed the specified threshold.

   Show answer

   Answer: D

   Next question

2. Question 352 · 364 in bank

   You are developing a machine learning pipeline using the Kubeflow Pipelines (KFP) SDK that will run on Agent Platform Pipelines. The pipeline includes a preprocessing component that takes two hours to transform a large dataset. You are currently iterating on the code for the subsequent model_training component and need to re-submit the pipeline frequently for testing. You need to ensure that the preprocessing component only re-runs if its inputs or parameters have changed. You want to minimize effort, costs, and total execution time. What should you do?

   • AEnsure that execution caching is enabled for the preprocessing component by including the set_caching_options(enable_caching=True) method.
   • BTrack the run lineage using Agent Platform Experiments, and pass the output artifact URI from the previous run into the training component.
   • CConfigure the pipeline to write the preprocessed data to a persistent volume claim (PVC) that is shared across pipeline runs.
   • DAdd a dsl.Condition step to the pipeline that checks if the preprocessing step has executed in the last hour.

   Show answer

   Answer: A

   Next question

3. Question 353 · 364 in bank

   You are an ML engineer at a large retail company. You have a BigQuery table containing five years of daily sales data for 50,000 distinct product SKUs. You need to generate a daily demand forecast for the next 30 days for each individual SKU. The solution must automatically account for holidays and weekly seasonality. You want to minimize data egress costs and operational complexity by keeping the data and model within the data warehouse. What should you do?

   • AUse the CREATE MODEL... OPTIONS(model_type='ARIMA_PLUS') statement in BigQuery to generate a forecast based on historical SKU sales.
   • BUse the gcloud ai custom-jobs create command to run a PyTorch LSTM container on Agent Platform that pulls data from a CSV into Cloud Storage.
   • CUse the CREATE MODEL... OPTIONS(model_type='LINEAR_REG') statement in BigQuery to predict sales using the transaction date as a feature.
   • DUse the google-cloud-aiplatform SDK to create a TimeSeriesDataset and trigger an AutoMLForecastingTrainingJob.

   Show answer

   Answer: A

   Next question

4. Question 354 · 364 in bank

   Your team is deploying a real-time personalized ad-ranking model. The model requires low-latency access for hundreds of user-level features to be fetched at request time. These features are stored in BigQuery and updated frequently by a streaming Dataflow pipeline. As your user base continues to grow, you discover that fetching these features from BigQuery at request time is causing the total inference latency to exceed your 100 ms SLA. You need a feature serving layer that provides low-latency retrieval and scales automatically while minimizing operational overhead. What should you do?

   • AUse the BigQueryReadClient to stream data directly into the model’s inference container.
   • BEnable Agent Platform Feature Store, and use the FetchFeatureValues API to fetch features for the model.
   • CImplement a Redis cache on Memorystore, and add a component to the Dataflow pipeline to write the feature to the Redis cache.
   • DUse BigQuery BI Engine to accelerate the SQL queries used to fetch features.

   Show answer

   Answer: B

   Next question

5. Question 355 · 364 in bank

   You are optimizing the training of a 175-billion parameter LLM on Gemini Enterprise Agent Platform. You have provisioned a TPU v5p Pod slice with 32 chips for the job. You initially ran the training using standard data parallelism, but the job immediately failed with an out-of-memory (OOM) error. You need to implement a training strategy that resolves the memory issue and minimizes training latency. What should you do?

   • AConfigure the training to use a parameter server architecture where the model weights are stored in the host CPU memory.
   • BEnable gradient checkpointing, and continue using the data parallelism strategy.
   • CConfigure the training to shard the model parameters and optimizer states, and continue using the data parallelism strategy.
   • DScale up the TPU Pod slice to include more chips, and continue using the data parallelism strategy.

   Show answer

   Answer: C

   Next question

6. Question 356 · 364 in bank

   You have deployed a custom classification model to an Agent Platform endpoint to predict the likelihood of customer churn in real-time. You need to collect the data to calculate the model’s exact precision and recall by comparing production predictions with ground truth data in BigQuery. You want to minimize development effort. What should you do?

   • AModify the prediction container to asynchronously write the prediction result to BigQuery after every request.
   • BSchedule a daily Agent Platform batch prediction job that uses the same model resource as the deployed endpoint and stores predictions in BigQuery.
   • CUpdate the model deployment parameters to enable request-response logging to BigQuery. Set the sampling rate to 1.0.
   • DCreate a model evaluation job by providing the model resource and the BigQuery ground truth table to calculate precision and recall.

   Show answer

   Answer: C

   Next question

7. Question 357 · 364 in bank

   You are in the exploratory phase of development of a demand forecasting model. Several terabytes of historical sales and inventory records are stored in a BigQuery table. You need to create a new, clean table to perform preliminary analyses on metrics, such as moving averages of sales figures and weekly aggregates, without additional data movement. You want to use the simplest approach and minimize overhead. What should you do?

   • ACreate a PreprocessingOp component in the Agent Platform Pipelines workflow to compute the moving averages and extract time-based features and write the processed data to a new destination table in BigQuery.
   • BRead the data from BigQuery into a Dask DataFrame using Agent Platform Workbench. Perform exploratory data analysis (EDA) and feature engineering transformations, and write the processed data back to a new destination table in BigQuery.
   • CWrite a SQL query in the BigQuery console. Use the ML.TRANSFORM function in BigQuery ML to compute the moving averages and extract time-based features. Write the processed data to a new destination table in BigQuery.
   • DWrite a SQL query in the BigQuery console. Use standard SQL functions like CAST, EXTRACT, and window functions with OVER() clauses to perform the transformations and calculations. Materialize the results into a new destination table in BigQuery.

   Show answer

   Answer: D

   Next question

8. Question 358 · 364 in bank

   You work for a healthcare organization and need to automate the patient triage process. You want to build a telephony agentic solution using the Gemini Live API to perform human-like conversations with patients and gather necessary patient information. You plan to pilot the solution in Japanese. You need to minimize latency and development time, and ensure high, native-like fluency in the model’s responses. What should you do?

   • AUse Gemini Live API Speech-to-Text functionality to capture the user’s input followed by Text-to-Speech streaming processing with Chirp to communicate with the patient in their language.
   • BUse the Gemini Live API, and configure the language_code field to ja-JP. Include a clear system instruction to respond in the patient’s language.
   • CFine-tune the Gemini Flash model on a dataset of Japanese medical transcripts to ensure native fluency. Deploy this custom model behind the Gemini Live API.
   • DUse the Gemini Pro model to automatically select the appropriate language.

   Show answer

   Answer: B

   Next question

9. Question 359 · 364 in bank

   You work for an international bank. Your team is developing a generative AI application that summarizes complex financial reports. You need to implement an automated, scalable solution to evaluate and quantify the quality of generated summaries against the previous version of the solution. What should you do?

   • AUse Model Monitoring to track the prediction drift of the token output length.
   • BRun an automatic side-by-side (AutoSxS) evaluation job in Agent Platform to compare outputs from the new and previous versions of the solution.
   • CDeploy both versions of the solution, and use A/B testing to route a portion of production traffic to the new version and route the remaining traffic to the previous version. Compare user click-through rates.
   • DRun an evaluation job in Agent Platform, and calculate a BLEU score to compare the new and previous versions of the solution.

   Show answer

   Answer: B

   Next question

10. Question 360 · 364 in bank

    You are using generative AI to build a conversational agent to support bank customers’ inquiries about their product base and transaction history. You must ensure that the agent does not include personally identifiable information (PII) in its responses. You want to implement this requirement with minimal effort. What should you do?

    • AConfigure the safety_settings thresholds within the Agent Platform model parameters to Block_Most.
    • BDevelop a post-processing wrapper using Python’s re module with patterns r'\b(?:\d{4}[- ]?){3}\d{4}\b' to identify and redact credit card numbers.
    • CImplement Model Armor with Sensitive Data Protection to detect and redact PII from the agent’s responses.
    • DUse sampled Shapley (SHAP) values to understand the model’s predictions and ensure that it is not biased toward revealing PII.

    Show answer

    Answer: C

    Next question

Keyboard:J Kmove between questionsA–Dselect an answerRshow or hide the answer← →change page

PMLE practice, question 351. 364 questions available.