Exam AWS Certified Machine Learning - Specialty topic 1 question 203 discussion

https://docs.aws.amazon.com/sagemaker/latest/dg/model-managed-spot-training.html Managed spot training can optimize the cost of training models up to 90% over on-demand instances. SageMaker manages the Spot interruptions on your behalf. "Spot instances can be interrupted, causing jobs to take longer to start or finish. You can configure your managed spot training job to use checkpoints. SageMaker copies checkpoint data from a local path to Amazon S3. When the job is restarted, SageMaker copies the data from Amazon S3 back into the local path. The training job can then resume from the last checkpoint instead of restarting."

It has to be D. With Spot training we can reduce the cost and save the model weights with checkpoint enabled.

A. NO - D is simpler B. NO - D is simpler C. NO - D is simpler D. YES - works out-of-the-box

Managed spot training in Amazon SageMaker uses Amazon EC2 Spot instances to run training jobs, which can optimize the cost of training models by up to 90% over on-demand instances 1. SageMaker manages the Spot interruptions on the company’s behalf 1. By enabling checkpointing, the company can ensure that if a Spot instance is interrupted, the training job can resume from the last checkpoint instead of restarting, avoiding loss of work and model retraining 1

Selected Answer: D Use managed spot training in Amazon SageMaker. Launch the training jobs with checkpointing enabled. Managed spot training in Amazon SageMaker can help minimize operational overhead and cost by using spot instances to perform the training. This can significantly reduce the cost of training, while still achieving the same accuracy. SageMaker provides built-in checkpointing capability, which allows saving model weights and progress to Amazon S3 periodically. This ensures that even if the spot instances are terminated, the training can resume from the last saved checkpoint. Additionally, SageMaker provides a managed service, so the ecommerce company does not need to worry about managing the infrastructure, and can focus on building and tuning their model.

Selected Answer: D The ML specialist should choose option D, which provides the training data to SageMaker with the least development overhead. This option involves putting the TFRecord data into an Amazon S3 bucket and pointing the SageMaker training invocation to the S3 bucket without reformatting the training data. Using SageMaker script mode is a convenient way to execute training scripts without any modification. Since the training script train.py already works with TFRecord data, it can be used as is without any changes. By storing the data in S3 and accessing it from there, the specialist can take advantage of SageMaker's built-in data distribution and parallelization capabilities, which can significantly speed up training. Rewriting the train.py script or using additional services like AWS Glue or Lambda would add unnecessary complexity and increase development overhead.

Managed spot training in Amazon SageMaker provides a cost-effective way to run large machine learning workloads. With managed spot training, the training jobs are executed using Amazon EC2 Spot instances, which can significantly reduce the cost of training. Additionally, by launching training jobs with checkpointing enabled, the work done up to the last checkpoint is saved to Amazon S3. This ensures that the training job can be resumed from the last checkpoint in case of instance failure or termination. This minimizes the risk of data loss and avoids the need for retraining the model from scratch. Using Amazon SageMaker also reduces the operational overhead required to set up and manage the training environment.