A company uses Amazon SageMaker with EC2 instances for training and inference and wants to reduce training costs without changing the current architecture. Training jobs can tolerate interruptions and results can be waited for days. Which combination of resources should the company use to lower costs most cost-effectively? (Choose two.)

Spot Instances are the most cost-effective choice for training jobs that can tolerate interruptions, as they offer significant discounts compared to On-Demand instances. Since the training jobs can wait for days, the company can leverage the cost savings of Spot Instances without impacting critical deadlines. Checkpoints are crucial when using Spot Instances because they allow the training process to save its state periodically. If a Spot Instance is interrupted, the training can resume from the last saved checkpoint on a new instance, preventing the loss of progress and wasted computation, thus making the overall process more cost-effective. On-Demand Instances are more expensive and don't offer cost savings. Reserved Instances are for predictable, long-term workloads, not interruptible training. Incremental training is a technique to improve model performance or adapt to new data, not primarily a cost-saving mechanism for instance types.