Answer-first summary for fast verification
Answer: MirroredStrategy; Custom tier with a single master node and four v100 GPUs., MirroredStrategy; Custom tier with a single master node and eight v100 GPUs.
**Correct Answers:** MirroredStrategy; Custom tier with a single master node and four v100 GPUs (B) and MirroredStrategy; Custom tier with a single master node and eight v100 GPUs (E). **Why?** - **MirroredStrategy:** This strategy is ideal for training large models like Inception-v3 on multiple GPUs, as it replicates the model on each GPU, allowing for parallel training and significantly reducing wall-clock time. - **Custom Tier with Four v100 GPUs (B):** Offers a cost-effective balance between performance and expense, suitable for projects with budget constraints. - **Custom Tier with Eight v100 GPUs (E):** While more expensive, this configuration provides higher computational power, further reducing training time, making it suitable for projects with tighter deadlines and higher budgets. **Other Options Considered:** - **Default Strategy (C):** Less efficient for large models and multiple GPUs compared to MirroredStrategy. - **One Device Strategy (A & D):** Limits training to a single GPU, which is inefficient for large-scale models and does not leverage the full potential of available hardware.
Author: LeetQuiz Editorial Team
Ultimate access to all questions.
No comments yet.
Your team is working on an image classification project involving 10,000 images using a TensorFlow Inception-v3 CNN model pretrained on ImageNet. The project has strict deadlines, requiring the training process to be as efficient as possible in terms of wall-clock time. The team has decided to use Google's AI Platform for model training. Given the constraints of cost-effectiveness and the need for rapid training, which TensorFlow distribution strategy and AI Platform training job configuration should you select? Choose the best option(s). (Choose two correct options if option E is available)
A
One Device Strategy; Custom tier with a single master node and eight v100 GPUs.
B
MirroredStrategy; Custom tier with a single master node and four v100 GPUs.
C
Default Strategy; Custom tier with a single master node and four v100 GPUs.
D
One Device Strategy; Custom tier with a single master node and four v100 GPUs.
E
MirroredStrategy; Custom tier with a single master node and eight v100 GPUs.