MLA-C01 Question 277
Select 3A data scientist at your organization is tasked with training a deep learning model using TensorFlow. The team wants to deploy the model on Amazon SageMaker using a container but is unsure whether to use a provided container or a customized container. Which of the following considerations should guide their choice?
- A
If the training requires a specific version of TensorFlow that is not supported by the provided containers, a customized container is needed.
- B
If the model includes custom libraries and dependencies not available in the provided containers, a customized container is required.
- C
Provided containers should be used if the required framework version and dependencies are supported out-of-the-box.
- D
Customized containers are mandatory for all deep learning training jobs on Amazon SageMaker.
- E
Provided containers should always be avoided when deploying models with TensorFlow.
Show answer and explanation
Correct answers: A, B, C
Explanation
The choice between provided and customized containers in Amazon SageMaker depends on the requirements of the training job. Provided containers are pre-configured for popular frameworks like TensorFlow and are suitable for standard use cases. However, customized containers are necessary when specific framework versions, custom libraries, or dependencies are required. Understanding these trade-offs ensures the appropriate container is used for efficient model training and deployment.
- A. Correct.
Correct: If the required version of TensorFlow is not supported by the provided containers, a customized container allows the team to include the specific version and its dependencies.
- B. Correct.
Correct: When the model relies on additional custom libraries or dependencies that are unavailable in the provided containers, a customized container is necessary.
- C. Correct.
Correct: Provided containers are optimized and pre-configured for common use cases, and they should be used when the required framework version and dependencies are already supported.
- D. Incorrect.
Incorrect: Customized containers are not mandatory for all deep learning training jobs. Provided containers are sufficient for many standard use cases.
- E. Incorrect.
Incorrect: Provided containers are often the best choice for TensorFlow models when the required versions and dependencies are supported, as they simplify deployment and reduce effort.