MLA-C01 Question 414
Select 2A data scientist is training a machine learning model on a large dataset that requires extensive matrix computations and high single-threaded performance. The team is also planning to deploy this model for inference in a production environment with low latency requirements. Considering the training and inference needs, which combination of AWS instance types should the team select?
- A
Compute Optimized (C5) for training and Inference Optimized (Inf1) for inference
- B
Memory Optimized (R5) for training and Compute Optimized (C5) for inference
- C
Compute Optimized (C5) for both training and inference
- D
Memory Optimized (R5) for training and Inference Optimized (Inf1) for inference
- E
General Purpose (M5) for training and Inference Optimized (Inf1) for inference
Show answer and explanation
Correct answers: A, D
Explanation
Selecting the right instance type is critical for optimizing performance and cost in machine learning workflows. Compute Optimized (C5) instances are ideal for training tasks requiring high compute power, such as matrix computations. For inference, Inference Optimized (Inf1) instances provide low-latency performance and are specifically designed for deploying ML models. While Memory Optimized (R5) instances can also be used for training if the workload is memory-intensive, General Purpose (M5) instances are not ideal for the compute-heavy requirements of this scenario.
- A. Correct.
Correct: Compute Optimized (C5) instances are well-suited for tasks requiring high single-threaded computation, such as matrix operations during training. Inference Optimized (Inf1) instances are designed for low-latency inference tasks, making them ideal for deploying ML models in production.
- B. Incorrect.
Incorrect: Memory Optimized (R5) instances are ideal for applications requiring significant memory, such as processing large datasets in memory, but they are not optimal for compute-intensive tasks like training a model with extensive matrix computations. Additionally, Compute Optimized (C5) instances are not specialized for low-latency inference.
- C. Incorrect.
Incorrect: Compute Optimized (C5) instances are suitable for training due to their high compute power, but using them for inference is not ideal when low latency is a priority. Inference Optimized (Inf1) instances are better suited for inference tasks.
- D. Correct.
Correct: Memory Optimized (R5) instances can be beneficial for training if the dataset is memory-intensive, and Inference Optimized (Inf1) instances are suitable for low-latency inference. This combination is valid depending on the specific workload.
- E. Incorrect.
Incorrect: General Purpose (M5) instances provide a balanced mix of compute, memory, and networking but are not specialized for compute-intensive training tasks. Inference Optimized (Inf1) instances are suitable for inference, but pairing them with M5 instances for training is suboptimal for this scenario.