MLA-C01 Question 436
Select 3A company is building a machine learning model to predict the likelihood of customers purchasing a product. The team wants to use Amazon SageMaker and has already prepared the training and validation datasets in Amazon S3. They need to ensure that the training process leverages distributed training across multiple GPUs for faster execution. Which of the following actions should the team take to achieve this?
- A
Choose a SageMaker training instance type that supports multiple GPUs, such as ml.p3.16xlarge.
- B
Use a built-in SageMaker algorithm that supports distributed GPU training, such as XGBoost.
- C
Implement a data parallelism or model parallelism strategy for distributed training.
- D
Ensure that the training script uses the SageMaker Training Toolkit to manage distributed training automatically.
- E
Use SageMaker Clarify to pre-process the dataset before training.
Show answer and explanation
Correct answers: A, C, D
Explanation
To leverage distributed training across multiple GPUs in SageMaker, you need to select an appropriate instance type that supports multiple GPUs (e.g., ml.p3.16xlarge), use a distributed training strategy such as data or model parallelism, and ensure your training script is configured to handle distributed training through the SageMaker Training Toolkit. These steps work together to optimize the training process for large datasets and complex models. Using tools like SageMaker Clarify is unrelated to distributed training.
- A. Correct.
This is correct. Choosing a training instance type that supports multiple GPUs, such as ml.p3.16xlarge, is necessary for distributed training.
- B. Incorrect.
This is incorrect. Although XGBoost is a built-in algorithm in SageMaker, it does not inherently support distributed GPU training. Distributed GPU training requires specific algorithms or frameworks designed for such purposes.
- C. Correct.
This is correct. Implementing a data parallelism or model parallelism strategy is a key step in enabling distributed training across multiple GPUs.
- D. Correct.
This is correct. The SageMaker Training Toolkit simplifies distributed training by providing utilities that manage distribution across GPUs.
- E. Incorrect.
This is incorrect. SageMaker Clarify is used for detecting bias and explaining model predictions; it does not assist in distributed training.