Chapter 3 of 6 · Google Cloud Professional Machine Learning Engineer
Scaling prototypes into ML models
Model and product choice, training, tuning and hardware.
- 4topics
- 20 minfast track
- 1diagrams and exhibits
- 21%of the exam
Topics in this chapter
Take them in order, or jump to the one you need.
- 1Choosing the Model Type, Product and DeploymentChoose ARIMA, DNN, tree or LLM models, pick AutoML, BigQuery ML or Pipelines, plan the deployment and meet interpretability needs…5 min
- 2Organizing Training Data and Running Training JobsOrganize training data in Cloud Storage and BigQuery, choose custom training, Kubeflow on GKE, AutoML or Tabular Workflows, and…5 min
- 3Hyperparameter Tuning and Fine-Tuning Foundation ModelsRun hyperparameter tuning jobs with search algorithms and trial limits, and decide when and how to fine-tune Gemini and open…5 min
- 4Training Hardware and Distributed TrainingChoose CPUs, GPUs or TPUs for training, and use data parallelism, model parallelism and worker pools to train large models across…Diagram5 min
About this chapter
What this chapter covers
Scaling prototypes into ML models is about 21% of the Google Cloud Professional Machine Learning Engineer exam, the largest share of any section. It covers turning a notebook prototype into a model trained properly at scale on Gemini Enterprise Agent Platform (the platform formerly called Vertex AI), with the right trade-offs between cost, complexity, latency and scalability.
It covers building models for the task (choosing a model type such as ARIMA, DNN or LLM; choosing Agent Platform AutoML, BigQuery ML or Agent Platform Pipelines; choosing a deployment strategy; meeting interpretability requirements), training models (organizing training data in Cloud Storage and BigQuery; ingesting structured and unstructured data; Agent Platform custom training, Kubeflow on GKE, AutoML and Tabular Workflows; troubleshooting failures; hyperparameter tuning; fine-tuning foundation models) and choosing training hardware (CPU, GPU and TPU; distributed training with data and model parallelism).
The exam has 50 to 60 multiple choice and multiple select questions in 2 hours. Questions in this section usually describe a model, a data size and a constraint such as budget, deadline or explainability, then ask "What should you do?".
How the topics fit together
Constraints decide the model and product, the product decides how training runs, and the model size decides hardware and distribution. A wrong first choice makes every later step more expensive.
Read the constraint words first. "Explain each decision to regulators" favours interpretable models, "least effort" favours AutoML or BigQuery ML, and "model too large for one accelerator" means model parallelism.
How to use these pages
Each topic page has the same shape, so you can skim for what you need.
- What it is: the idea in plain words.
- How it works: the mechanism, with tables and code examples.
- What to remember for the exam: the facts to memorise and how questions are phrased.
- Common mistakes: the classic traps.
- Check yourself: three questions to test your understanding.
Ready for more practice?
The full Google Cloud Professional Machine Learning Engineer question bank has 749 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Google Cloud. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.