My Role
UX Designer — Interaction Design, User Flows, Prototyping
Team
Grace, PM
Qi, SWE
Fanny, SWE
Timeline
November 2023 - January 2024
Overview
Deep Learning VM Images are VMware's tailored solution for deploying high-performance virtual environments optimized for deep learning tasks. These VM images are designed to support popular machine learning frameworks and tools such as TensorFlow, PyTorch, and JupyterLab, making it easier for data scientists to start working with their models quickly. By streamlining the deployment process and enabling easy GPU resource allocation, Deep Learning VM Images eliminate the complexities of setting up and configuring deep learning environments.
I spearheaded the design of these VM images, focusing on simplifying the setup and providing pre-configured, optimized environments. This approach enhances deployment efficiency, allowing data scientists to rapidly access and utilize powerful deep learning tools.
Understanding their roles and needs
Anita,
Virtual Infrastructure (VI) Admin
JTBD
I want to efficiently allocate hardware resources, so that I can ensure optimal VM performance without delays
Goals
Streamline VM deployment while ensuring available resources like GPUs and NICs are allocated efficiently
Pain points
Manually assigning hardware after deployment wastes time and adds complexity
Lack of real-time resource visibility makes configuration difficult
Juggling multiple workflows for hardware adjustments slows down operations
Bob,
Data Scientist
JTBD
I want to quickly start my deep learning / machine learning work, so that I can focus on training models rather than configuring environments.
Goals
Start work immediately with pre-configured environments that include all necessary frameworks.
Pain points
Installing frameworks manually wastes time and risks version conflicts.
Inconsistent environments across VMs hinder collaboration and reproducibility.
Workflow 1: Deploy VM from template
Anita follows the 9-step process to deploy a VM using a deep learning template from her content library.
Workflow 2: Add GPU resources
Step 1. Edit VM settings
After completing the deployment workflow, Anita must go to the VM details page to modify the settings and add GPU resources.
Workflow 2a: Add GPU resources
Step 2. Add Peripheral Component Interconnect (PCI) device
Anita selects “PCI Device” from the “Add New Device” dropdown.
Workflow 2b: Add GPU resources
Step 3. Select vGPU profile
Anita chooses the appropriate vGPU profile that provides the necessary resources for the VM.
Workflow 2c: Add GPU resources
Step 4. Confirm GPU resources have been added
Anita confirms that the selected vGPU profile has been successfully applied to her VM.
1. Deploy Private AI Foundation Workload Domain
Anita follows the 9-step process to deploy a VM using a deep learning template from her content library.
2. Subscribe to content library
She subscribes to the content library in vCenter, which includes VMware's Deep Learning VM image, and is now ready to create her own VM.
3. Deploy VM from template
She launches the “Deploy from Template” workflow and completes steps 1-8.
4. Customize hardware
She no longer needs to deploy the VM first before adding hardware resources like GPUs; instead, she can allocate them directly during the deployment process.
5. Select GPU device
After choosing "PCI Device" from the dropdown menu, she selects the necessary GPU resources for her VM.
6. Confirm desired GPU device is selected
Once she selects the desired GPU resources for her VM, she verifies that the correct profile is applied and proceeds to the next step.
7. Select data science toolkit
She chooses the software bundle that defines the data science toolkit for the VM, then inputs the required details, such as container versions and configuration tokens.
8. Deep Learning VM is deployed!
She deploys the deep learning VM and provides the login details to the data scientist, allowing them to begin training models and working within the environment.
