In this tutorial, we will learn how to set up a Project on Gradient, connect your GitHub account to Paperspace, and create a Workflow from one of the templates we provide, in this case StyleGan2.
Workflows are designed to help automate machine learning tasks. Workflows can be used independently or in conjunction with Notebooks and Deployments.
StyleGan2 is one of the premiere unconditional generative image modeling frameworks, and we will use this workflow to generate images of novel faces using our Workflow.
What we will learn
In this tutorial, we will apply Notebook findings to an automatic, iterative training process using Workflows and the accompanying YAML Workflow spec, and learn how to interact with the Gradient Workflows UI.
After running a successful Workflow, we will discuss possibilities for additional Workflow capabilities like modifying the Workflow YAML syntax for customization of our Workflow in the follow up tutorial: Getting Started: Build and Run a Custom Workflow.
In your browser of choice, head to the Paperspace Gradient homepage and sign up for an account. If you already have an account, log in. This will take you to the Gradient workspace where you will create your first project.
The projects homepage
Create a Project
From the Projects tab in the Gradient console, select Create A Project. Name your Project.
Users can create a Worfklow within a Project.
Create and Run a Sample Workflow
Navigate to the Workflows tab once you are in your new project. Select Create to make our Workflow. It will then ask to set up GitHub integration. New users should configure their Paperspace account to connect to their GitHub, at this time.
Next, we can either run an automatic sample Workflow, or we can set up our own custom Workflow. Before creating our own custom Workflow, we will select Automatically run a Workflow which will clone StyleGAN2 and run a simple face generator.
The Workflow UI
We will soon see a Workflow spinning up in the UI:
Each green box represents a different job in the YAML file, displayed under the Overview.
By clicking on the green box representing each job, we can look at the outputs, logs, compute metrics, and YAML for each component job in the Workflow. When the Workflow completes our logs will look something like this:
Logs from our first Workflow run.
It looks like our Workflow successfully completed! The logs allow us to see where any issues may have come up in the workflow. They also facilitate making those changes by showing us which step in the job was problematic.
In the same panel in the UI, let's select Data to see if we can view the results of our face generator.
If we click into our data folder we find a list of generated files:
The data tab for each Project contains the stored Datasets
And we can preview one of the faces we just generated using StyleGAN2!
Congratulations! You've run your first Workflow. At this point here is everything we've done:
Learned how to create a Project and Workflow on Gradient within your browser
How to connect your Github account to Gradient
How to interact with Workflows UI in the browser
Part 2 of this tutorial, detailing how to run a custom Workflow from the Gradient CLI, can be found here
If you feel comfortable with this classical ML implementation of Workflows and are ready to move on to using more advanced or deep learning techniques, work through the tutorial on creating a recommender system using Notebooks and Workflows