Google Cloud Professional Machine Learning Engineer — Question 211

You are pre-training a large language model on Google Cloud. This model includes custom TensorFlow operations in the training loop. Model training will use a large batch size, and you expect training to take several weeks. You need to configure a training architecture that minimizes both training time and compute costs. What should you do?

Answer options

• A. Implement 8 workers of a2-megagpu-16g machines by using tf.distribute.MultiWorkerMirroredStrategy.
• B. Implement a TPU Pod slice with -accelerator-type=v4-l28 by using tf.distribute.TPUStrategy.
• C. Implement 16 workers of c2d-highcpu-32 machines by using tf.distribute.MirroredStrategy.
• D. Implement 16 workers of a2-highgpu-8g machines by using tf.distribute.MultiWorkerMirroredStrategy.

Correct answer: A

Explanation

The correct answer, A, is optimal because a2-megagpu-16g machines provide high memory bandwidth and GPU capabilities that are well-suited for training large models with substantial batch sizes. The other options either involve less efficient hardware configurations or a TPU strategy that may not fully leverage the benefits of the specific TensorFlow operations implemented in the model.