Domain 3: Scaling Prototypes into ML Models¶
โ Back to PMLE Learning Path
Official exam weight: approximately 21% โ the highest-weighted domain.
Official objectives¶
3.1 Build models for cost, complexity, latency, and scalability¶
- Choose an appropriate model type, such as ARIMA, deep neural networks (DNNs), or large language models (LLMs).
- Select a suitable product, such as Agent Platform AutoML, BigQuery ML, or Agent Platform Pipelines.
- Choose a deployment strategy.
- Select modeling techniques that meet interpretability requirements.
3.2 Train models¶
- Organize tabular, text, speech, image, and video training data in services such as Cloud Storage and BigQuery.
- Ingest structured and unstructured data into training pipelines.
- Understand Agent Platform custom training, Kubeflow on GKE, Agent Platform AutoML, and Tabular Workflows.
- Troubleshoot model training failures.
- Perform hyperparameter tuning.
- Understand when and how to fine-tune foundation models from Agent Platform and Model Garden.
3.3 Choose appropriate training hardware¶
- Evaluate CPU, GPU, and TPU options.
- Understand distributed training on GPUs and TPUs using data and model parallelism.
Guided labs to build¶
| Status | Lab | Objective |
|---|---|---|
| Planned | Move from notebook prototype to reproducible training code | 3.1, 3.2 |
| Planned | Submit a custom training job | 3.2 |
| Planned | Configure hyperparameter tuning | 3.2 |
| Planned | Diagnose common training-job failures | 3.2 |
| Planned | Fine-tune a foundation model and assess whether tuning is justified | 3.2 |
| Planned | Compare CPU, GPU, and TPU training options | 3.3 |
| Planned | Understand data parallelism versus model parallelism | 3.3 |
Google Skills¶
Open the official PMLE learning path for the current training, model development, and accelerator activities.
Exam focus¶
This is the largest exam domain. Practise selecting a model and training approach from requirements rather than memorizing product descriptions. Be able to discuss cost, complexity, latency, scalability, interpretability, data shape, and accelerator constraints.
๐ฌ DevopsPilot Weekly โ Learn DevOps, Cloud & Gen AI the simple way.
๐ Subscribe here