To begin using Gradient, follow these preliminary steps:
Create a team if higher tiers of service and collaboration features are desired
Now you can create Notebooks, Jobs, Projects, Experiments, Deployments, and more!
Note: if you are a Gradient Enterprise customer, or a Gradient Private Cloud user, please visit the Gradient Private Cloud section for more info on how to utilize private processing site clusters.
Notebooks can be created by clicking Create Notebook button on the Notebooks tab. You can stop, start, fork, and swap out the instance type anytime. Choose from a wide selection of pre-configured templates or bring your own. See more info here.
You can run Experiments from the web interface or CLI:
Before creating an experiment using the CLI, you must first install the CLI and create a Project for your Experiments to live in. To create a Project, navigate to Gradient > Projects in the UI and click Create Project. Then select Create Standalone Project and provide a project name. Now, you can use the created Project's Project ID in order to create Experiments in that Project via the CLI.
The following command will work and will create and start an Experiment that will display properties of the attached GPU. Be sure to replace
<your-project-id> with your Project ID.
gradient experiments run singlenode --projectId <your-project-id> --container 'Test-Container' --machineType P4000 --command 'nvidia-smi' --name 'test-01' --workspace none --apiKey <your-api-key>
You can also create
hyperparameter Experiments, and you can use the
create command to simply create Experiments that can be started later. Explore all the advanced options here.
Behind the scenes, your Experiment will be uploaded and executed on our cluster of machines starting with the command you provided. There are several optional Experiment parameters, such as to specify your workspace (the additional files to be used in your experiment). You can always use the
--help option after any command in the CLI for more info.
Experiments states transition from Queued > Pending > Running. Once the Experiment is in the Running state, you can watch your Experiment run in the CLI and web UI. An Experiment can complete with the following states: Success, Cancelled, Error, or Failed.
Congratulations! You ran your first Experiment on Gradient 🚀