MENU
You are developing an image recognition model using PyTorch based on ResNet50 architecture. Your code is working fine on your local laptop on a small subsample. Your full dataset has 200k labeled images You want to quickly scale your training workload while minimizing cost. You plan to use 4 V100 GPUs. What should you do? (Choose Correct Answer and Give Reference and Explanation)
Suggested Answer: C
Get Premium Professional Machine Learning Engineer Questions
as Interactive Web-Based Practice Test or PDF
Real Exam Format
Detailed Explanations
Instant Access
Get Your Practice Test
Traffic splitting is a feature of Vertex AI that allows you to distribute the prediction requests among multiple models or model versions within the same endpoint. You can specify the percentage of traffic that each model or model version receives, and change it at any time. Traffic splitting can help you test the new model in production without creating a new endpoint or a separate service. You can deploy the new model to the existing Vertex AI endpoint, and use traffic splitting to send 5% of production traffic to the new model. You can monitor the end-user metrics, such as listening time, to compare the performance of the new model and the previous model. If the end-user metrics improve between models over time, you can gradually increase the percentage of production traffic sent to the new model. This solution can help you test the new model in production while minimizing complexity and cost.Reference:
Deploying models to endpoints | Vertex AI
33 comments · newest first
BE
Beatriz
10 months ago
I thought Vertex AI was only for smaller projects, this is surprising!
KA
Kallie
10 months ago
I agree, C is straightforward and efficient!
AI
Aileen
11 months ago
Wait, isn't using Kubernetes overkill for this?
SH
Shanda
11 months ago
B could work too, but I think C is more user-friendly.
MI
Minna
11 months ago
Option C seems like the best choice for quick scaling with those GPUs.
HE
Hershel
11 months ago
I lean towards option D because it aligns with what we learned about container orchestration, but I need to double-check the specifics of the TFJob setup.
JO
Jody
11 months ago
I feel like using a Compute Engine VM could work, but I can't recall if it would be the most cost-effective compared to the other options.
DA
Daren
12 months ago
I think we practiced a similar question where Kubernetes was mentioned. It seems like a good way to manage resources, but I'm unsure about the TFJob operator part.
LE
Leonard
12 months ago
I remember we discussed using Vertex AI for scaling, but I'm not sure if a user-managed notebook is the best option for cost efficiency.
KE
Kenneth
1 year ago
I'm leaning towards option D. Using a GKE cluster with the required GPU node pool and submitting a TFJob operator could be a more flexible and scalable solution compared to the Vertex AI options.
AR
Ardella
1 year ago
Based on the information provided, I believe option C is the best solution. Creating a Vertex AI Workbench instance with the necessary GPUs and training the model directly on that platform seems like the quickest and most cost-effective approach.
MO
Mona
1 year ago
Hmm, I'm a bit confused. Should I be looking at creating a custom Vertex AI tier or using the pre-built Vertex AI Workbench instance? I'm not sure which one would be the most efficient approach.
BL
Blair
1 year ago
This question seems straightforward. I think option B is the way to go - packaging the code and using a pre-built container to train the model on Vertex AI with the required GPUs.
NO
Norah
1 year ago
Wait, I can use my refrigerator as a GPU? *scratches head* Nah, that's probably not a good idea. Time to get serious and go with Option B. Gotta package that code up and let Vertex AI handle the heavy lifting!
HY
Hyun
1 year ago
That's right, Option B is the best choice for scaling your training workload efficiently.
NI
Nicholle
1 year ago
Option B sounds like the way to go. Package your code and let Vertex AI do the heavy lifting.
RO
Ronny
1 year ago
Yeah, using your refrigerator as a GPU is definitely not a good idea.
RH
Rhea
1 year ago
Hmm, I'm not sure I trust any of these options. I think I'll just train my model on my local laptop and hope it scales eventually. Who needs fancy cloud infrastructure anyway? *laughs nervously*
CA
Carmelina
1 year ago
Training on your local laptop may work for small datasets, but for 200k labeled images, utilizing cloud resources like V100 GPUs will definitely improve your model's performance and efficiency.
EV
Eva
1 year ago
Using cloud infrastructure like Vertex AI can greatly speed up your training process and save you time in the long run. It's worth considering!
JE
Jessenia
1 year ago
Option A seems like the best choice for scaling your training workload with minimal cost. You should configure a Compute Engine VM with the necessary dependencies and use Vertex AI with custom tier containing the required GPUs.
LA
Lashandra
1 year ago
Option D is intriguing, but it sounds a bit more complex than I'd like to deal with. Setting up a GKE cluster and submitting a TFJob operator seems like a lot of work. I'm more interested in a simpler, managed solution like Vertex AI.
ER
Ernestine
2 years ago
I'm leaning towards Option C. Creating a Vertex AI Workbench instance with the required GPUs sounds like a great way to get my model training up and running quickly. Plus, I don't have to worry about managing the infrastructure myself.
TH
Theodora
1 year ago
Absolutely, Vertex AI Workbench provides a hassle-free way to scale your training workload efficiently.
ME
Melvin
1 year ago
It's definitely a convenient option. Plus, you can focus more on optimizing your model rather than setting up the environment.
BE
Bev
1 year ago
I agree, managing the infrastructure myself can be time-consuming. Vertex AI Workbench takes care of that.
AS
Asha
1 year ago
Option C seems like a good choice. Using Vertex AI Workbench with 4 V100 GPUs can speed up training.
LA
Lauran
2 years ago
Option A seems like the easiest way to scale my training workload. I can simply configure a Compute Engine VM with the necessary dependencies and use Vertex AI to train my model. Definitely the fastest and most cost-effective solution.
LA
Layla
1 year ago
Option A seems like the easiest way to scale my training workload. I can simply configure a Compute Engine VM with the necessary dependencies and use Vertex AI to train my model. Definitely the fastest and most cost-effective solution.
AR
Arlette
1 year ago
A) Configure a Compute Engine VM with all the dependencies that launches the training Train your model with Vertex AI using a custom tier that contains the required GPUs.
JE
Jesusita
2 years ago
I'm not sure, but I think option D could also be a valid choice. Creating a GKE cluster with a node pool that has 4 V100 GPUs might be a good solution too.
PA
Paul
2 years ago
I agree with Robt. Option A seems like the most efficient way to scale the training workload while minimizing cost.
RO
Robt
2 years ago
I think the correct answer is A. It involves configuring a Compute Engine VM with the necessary dependencies and using Vertex AI with the required GPUs.