AIP-C01 · Question #90
Your team is working on fine-tuning a pre-trained model for a generative AI application. What should you do to minimize training time and costs while ensuring the model performs well?
The correct answer is C. Use SageMaker managed Spot Training to save costs. SageMaker Managed Spot Training is purpose-built for exactly this scenario-it uses EC2 Spot Instances (which can cost up to 90% less than On-Demand) and integrates automatic checkpointing so training progress is preserved if a Spot Instance is interrupted. This minimizes both…
Question
Your team is working on fine-tuning a pre-trained model for a generative AI application. What should you do to minimize training time and costs while ensuring the model performs well?
Options
- AUse AWS Batch for training large datasets in parallel.
- BFine-tune the model on Amazon EC2 instances with GPUs.
- CUse SageMaker managed Spot Training to save costs.
- DUse the AWS Deep Learning AMIs and train locally.
How the community answered
(35 responses)- A3% (1)
- B17% (6)
- C71% (25)
- D9% (3)
Explanation
SageMaker Managed Spot Training is purpose-built for exactly this scenario-it uses EC2 Spot Instances (which can cost up to 90% less than On-Demand) and integrates automatic checkpointing so training progress is preserved if a Spot Instance is interrupted. This minimizes both training time (via managed orchestration) and cost. AWS Batch (A) is a general-purpose batch compute service and lacks native ML training optimizations. EC2 with GPUs (B) works but is expensive at On-Demand pricing and requires manual management. Deep Learning AMIs on local infrastructure (D) eliminates cloud scalability advantages and cannot leverage distributed or Spot-based training.
Topics
Community Discussion
No community discussion yet for this question.