MLA-C01 · Question #2
Case Study A company is building a web-based AI application by using Amazon SageMaker. The application will provide the following capabilities and features: ML experimentation, training, a central mod
The correct answer is B. Use SageMaker managed warm pools.. SageMaker managed warm pools keep compute infrastructure provisioned and "warm" between training jobs for a configurable period, so consecutive jobs skip the cold-start provisioning delay entirely - directly addressing the goal of minimizing startup times during ML experimentatio
Question
Case Study A company is building a web-based AI application by using Amazon SageMaker. The application will provide the following capabilities and features: ML experimentation, training, a central model registry, model deployment, and model monitoring. The application must ensure secure and isolated use of training data during the ML lifecycle. The training data is stored in Amazon S3. The company is experimenting with consecutive training jobs. How can the company MINIMIZE infrastructure startup times for these jobs?
Options
- AUse Managed Spot Training.
- BUse SageMaker managed warm pools.
- CUse SageMaker Training Compiler.
- DUse the SageMaker distributed data parallelism (SMDDP) library.
How the community answered
(52 responses)- A15% (8)
- B71% (37)
- C10% (5)
- D4% (2)
Explanation
SageMaker managed warm pools keep compute infrastructure provisioned and "warm" between training jobs for a configurable period, so consecutive jobs skip the cold-start provisioning delay entirely - directly addressing the goal of minimizing startup times during ML experimentation.
Why the others are wrong:
- A (Managed Spot Training): Uses spare EC2 capacity to cut costs, but can actually increase wait times due to spot instance availability and potential interruptions - the opposite of minimizing startup time.
- C (Training Compiler): Optimizes GPU utilization by compiling the training computation graph, reducing training duration - but it doesn't touch infrastructure provisioning time at all.
- D (SMDDP library): Distributes training workloads across multiple GPUs/nodes to handle large-scale models faster - again, about training throughput, not cluster startup latency.
Memory tip: Associate "warm pools" with a car left running between errands - the engine (infrastructure) stays warm so you never wait for it to start. Any question mentioning consecutive or iterative training jobs and startup time should immediately point you to warm pools.
Topics
Community Discussion
No community discussion yet for this question.