AIF-C01 Applications of Foundation Models Practice Question

A company fine-tunes a foundation model on SageMaker using a custom dataset. They notice the training job takes too long. Which optimization technique is specifically designed to reduce training time for foundation models?

⚠ Common exam trap

AWS often tests the misconception that cost-saving techniques like Spot Instances or smaller instances also improve performance, but the question specifically asks for optimization to reduce training time, not cost.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Distributed training using SageMaker Data Parallelism

SageMaker Data Parallelism distributes the training workload across multiple GPUs or instances, splitting the data and synchronizing gradients using optimized all-reduce algorithms. This specifically reduces training time for large foundation models by enabling parallel computation, which is the most direct technique for accelerating training at scale.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✓
Distributed training using SageMaker Data Parallelism
Why this is correct
SageMaker distributed data parallelism shards the training dataset and gradients across multiple GPUs or instances, so each worker processes a subset and synchronises updates. This cuts wall-clock training time for large foundation models that cannot fit efficiently on a single device.
✗
Using a smaller instance type
Why it's wrong here
A smaller instance type provides less compute and memory, so training runs slower or fails outright. It is tempting because it is a cost-saving change to the training environment, and would be correct when the workload is small enough that a lower-cost instance adequately meets requirements without extending duration.
✗
Using Spot Instances
Why it's wrong here
Spot Instances reduce compute cost through spare capacity pricing, not training duration; they can also be interrupted mid-job. It is tempting because it optimises the training job economically, and would be correct when the priority is lowering SageMaker spend on interruptible, checkpointed workloads rather than shortening training time.
✗
Reducing batch size
Why it's wrong here
Reducing batch size lowers memory per step but increases the number of steps, typically lengthening training. It is tempting because it is a common tuning lever when GPU memory is exhausted, and would be correct when fitting a large model onto limited accelerator memory rather than accelerating the job.

Go deeper

AIF-C01 Applications of Foundation Models Practice Question

Why each option matters

Option-by-option breakdown

Related to this question