MLA-C01 Question #39: Real Exam Question with Answer & Explanation

Question

A company wants to improve the sustainability of its ML operations. Which actions will reduce the energy usage and computational resources that are associated with the company's training jobs? (Choose two.)

Options

• AUse Amazon SageMaker Debugger to stop training jobs when non-converging conditions are
• BUse Amazon SageMaker Ground Truth for data labeling.
• CDeploy models by using AWS Lambda functions.
• DUse AWS Trainium instances for training.
• EUse PyTorch or TensorFlow with the distributed training option.

Explanation

Option A is correct because SageMaker Debugger can automatically halt training jobs that show non-converging patterns (e.g., loss plateauing or diverging), preventing wasted compute cycles and energy on runs that will never produce useful models. Option D is correct because AWS Trainium instances are purpose-built ML training chips (Trn1/Trn2) designed to deliver higher performance-per-watt than general-purpose GPU instances, directly reducing energy consumption for the same workload.

Why the distractors are wrong:

• B (Ground Truth): Data labeling is a pre-training activity - it has no effect on the energy consumed during training jobs themselves.
• C (Lambda): Lambda is used for inference/deployment, not training; optimizing deployment doesn't touch training resource usage.
• E (Distributed training): Distributed training speeds up wall-clock time but spreads the job across more instances simultaneously - it typically increases total resource consumption, not decreases it.

Memory tip: Think "Stop waste, use specialists." SageMaker Debugger stops wasteful jobs (A), and Trainium is a specialist chip built for efficient training (D). Both act at the training stage - anything about labeling, deployment, or simply adding more machines is off-scope.

No community discussion yet for this question.