PROFESSIONAL-MACHINE-LEARNING-ENGINEER · Question #349
You built a custom Vertex AI pipeline job that preprocesses images and trains an object detection model. The pipeline currently uses 1 n1-standard-8 machine with 1 NVIDIA Tesla V100 GPU. You want to…
The correct answer is D. Update the to use a machine with 24 vCPUs and 3 NVIDIA Tesla V100 GPUs. To reduce training time without sacrificing accuracy, you need more compute power-not less data or a simplified model. Adding more GPUs (D: 3x V100s) enables data parallelism or model parallelism, distributing training workload and reducing wall-clock time while training on the…
Question
Options
- AReduce the number of layers in your object detection model.
- BTrain the same model on a stratified subset of your dataset.
- CUpdate the WorkerPoolSpec to use a machine with 24 vCPUs and 1 NVIDIA Tesla V100 GPU.
- DUpdate the to use a machine with 24 vCPUs and 3 NVIDIA Tesla V100 GPUs.
How the community answered
(60 responses)- A3% (2)
- B15% (9)
- C7% (4)
- D75% (45)
Explanation
To reduce training time without sacrificing accuracy, you need more compute power-not less data or a simplified model. Adding more GPUs (D: 3x V100s) enables data parallelism or model parallelism, distributing training workload and reducing wall-clock time while training on the full dataset at full quality. More vCPUs also help with data preprocessing. Reducing model layers (A) would compromise accuracy. Training on a subset (B) reduces accuracy. Increasing only vCPUs without more GPUs (C) won't significantly speed up GPU-bound deep learning training, since the bottleneck is GPU compute, not CPU.
Topics
Community Discussion
No community discussion yet for this question.