Exam Professional Machine Learning Engineer topic 1 question 144 discussion

Custom trainer , don't overthink 1000%, this is google recommendation. you don't need Vertex AI Workbench user-managed notebooks,Google Kubernetes Engine, Compute Engine at at all , it is a waste of your effort https://cloud.google.com/vertex-ai/docs/training/configure-compute#specifying_gpus You can choose as your want

Why in the world would you setup a Compute engine VM, when your custom training job on vertex runs "serverless" atleast for the user side don't have to maintain the vm. You literally just have to select the region , machine type and accelerators that's all.

Pre-built container: Utilizing a pre-built PyTorch container image eliminates the need to manage dependencies within your container, saving time and simplifying the process. Vertex AI custom tier: Vertex AI custom tiers allow you to configure a machine type with the desired GPUs (4 V100 in this case) and pay only for the resources you use. This is more cost-effective than managing a dedicated VM instance. Setuptools packaging: Packaging your code with tools like Setuptools ensures all necessary libraries and scripts are included within the container, creating a self-contained training environment.

Using Vertex AI allows you to easily leverage multiple GPUs without managing infrastructure yourself. The custom tier gives you control to specify 4 V100 GPUs. Packaging with Setuptools and using a pre-built container ensures a consistent and portable environment with all dependencies readily available. This approach minimizes overhead and cost by relying on Vertex AI's managed service instead of setting up your own Kubernetes cluster or VMs.

Correct Ans is C. Below mentioned why B is incorrect.

Option B (using a Vertex AI Workbench user-managed notebooks instance with 4 V100 GPUs) is more suitable for interactive data exploration and experimentation rather than large-scale model training. Vertex AI Workbench is designed for collaborative data science, but using it for model training might not be the most efficient approach.

What is Supetools?

“Vertex AI provides flexible and scalable hardware and secured infrastructure to train PyTorch based deep learning models with pre-built containers and custom containers. (…) use PyTorch ResNet-50 as the example model and train it on ImageNet validation data (50K images) to measure the training performance for different training strategies”: https://cloud.google.com/blog/products/ai-machine-learning/efficient-pytorch-training-with-vertex-ai

Can someone explain why is B wrong?

Answer C Option A involves using Google Kubernetes Engine, which is a platform for deploying, managing, and scaling containerized applications. However, it requires more setup time and knowledge of Kubernetes, which might not be ideal for quickly scaling up training workloads. Furthermore, the use of the TensorFlow Job operator seems inappropriate for a PyTorch-based model.

Select C

C. Package your code with Setuptools, and use a pre-built container. Train your model with Vertex AI using a custom tier that contains the required GPUs. The recommended approach to scale the training workload while minimizing cost would be to package the code with Setuptools and use a pre-built container, then train the model with Vertex AI using a custom tier that contains the required GPUs. This approach allows for quick and easy scaling of the training workload while minimizing infrastructure management costs.

Vote for A, as you need to scale

It's B according to me, since VertexAI Notebook has alla dependencies for PyTorch that is the fastest solution

Google Kubernetes Engine (GKE) is a powerful and easy-to-use platform for deploying and managing containerized applications. It allows you to create a cluster of virtual machines that are pre-configured with the necessary dependencies and resources to run your machine learning workloads. By creating a GKE cluster with a node pool that has 4 V100 GPUs, you can take advantage of the powerful processing capabilities of these GPUs to train your model quickly and efficiently. You can then use the Kubernetes Framework such as TFJob operator to submit the job of training your model, which will automatically distribute the workload across the available GPUs. References: Google Kubernetes Engine TFJob operator Vertex Al