This chapter concludes our exploration of training techniques. You learned about Managed Spot Training, a simple way to slash training costs by 70% or more. You also saw how checkpointing helps to resume jobs that have been interrupted. Then, you learned about Automatic Model Tuning, a great way to extract more accuracy from your models by exploring hyperparameter ranges. Finally, you learned about SageMaker Debugger, an advanced capability that automatically inspects training jobs for unwanted conditions and saves tensor collections to S3 for inspection and visualization.
In the next chapter, we'll study model deployment in more detail.