Why A is the answer

Distributed training

Distributed training spreads the work across nodes and GPUs.

In detail

• Distributed training spreads the work across nodes.
• Scale only when one node is truly not enough.

See it in different ways

mindmap
  root((Reproducible run))
    Code
      Git
    Data
      Versioned data assets
    Environment
      Registered environment
    Compute
      Cluster that scales to zero
    Results
      MLflow

mindmap
  root(("Distributed training"))
    n1["Node 1"]
      n2["Share of the work"]
    n3["Node 2"]
      n4["Share of the work"]
    n5["Node 3"]
      n6["Share of the work"]

Why the other options are wrong

B. A higher logging level on the same single node

Logging does not add capacity.

C. A smaller test set for the final evaluation

A test set has no effect on training size.

D. A batch endpoint that scores the training data

An endpoint scores data, and does not train.

Related questions

• Question 2Your team wants to compare parameters and metrics across hundreds of training runs. Which tool fits?
• Question 4Model accuracy is falling because incoming data no longer looks like the training data. What is this, and what should be automated?
• Question 17Which tool logs parameters, metrics and artifacts for each run in Azure Machine Learning?
• Question 22A model needs its feature retrieval logic at inference. What do you do?

Learn more

• Our lesson: Model lifecycle
• MLflow in Azure Machine Learning
• Model monitoring