Training Models PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 5)

Questions 21–25

1. 21expert · medium

   A company wants to fine-tune a large language model for a specialized domain. They have a dataset of 10,000 examples. They are concerned about the cost of fine-tuning and the time it takes. They also need the model to be deployed in a specific region. Which approach should they take?

   • AUse AutoML for text classification with the dataset.
   • BUse a pre-trained model without fine-tuning and rely on prompt engineering.
   • CUse Vertex AI's fine-tuning service for a foundational model in the desired region.
   • DTrain a custom model from scratch on the dataset.
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   A training job fails because the training data contains missing values in a column that the model expects to be numeric. What type of error is this?

   • AModel convergence issue
   • BData error
   • CInfrastructure failure
   • DResource exhaustion
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A team is training a deep learning model on Vertex AI. They want to optimize hyperparameters such as learning rate, batch size, and number of layers. They have a budget of 50 training trials and want to maximize the chance of finding the best configuration. Which hyperparameter tuning strategy should they use?

   • AManual tuning by running a few experiments and selecting the best.
   • BGrid search over all possible combinations of hyperparameters.
   • CRandom search with a fixed number of trials.
   • DBayesian optimization (Vertex AI Vizier) to guide the search.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A team runs multiple training jobs on Vertex AI. They want to monitor the resource utilization (CPU, memory, GPU) of each job and set up alerts for abnormal behavior. Which tool should they use?

   • ACloud Monitoring, using the Vertex AI training job metrics.
   • BVertex AI TensorBoard, to visualize metrics.
   • CCloud Console's training job details page only.
   • DCloud Logging only, to view training logs.
   Reveal Solution Discussion Select an answer first
6. 25expert · medium

   A team is using Vertex AI hyperparameter tuning to optimize a model. They have a limited budget and want to minimize the number of trials. They also need to ensure that the tuning job does not exceed a certain cost. Which strategy should they use?

   • AUse Bayesian optimization with a small number of trials and early stopping.
   • BUse grid search with a coarse grid to limit trials.
   • CRun a single trial with manually chosen hyperparameters.
   • DUse a large search space with many trials to explore thoroughly.
   Reveal Solution Discussion Select an answer first