If you don't already have an account, create one.
First, log in using the client using vh login. Your credentials will be used to
acquire an authentication token. For your convenience, authentication tokens do not
expire, but you can audit your authentication tokens in the app at any time.
If you imported our example project over at the app, you should now clone it to a working copy of your own.
git clone https://github.com/valohai/tensorflow-example
cd tensorflow-exampleNow that you are in the working directory, you can link it with a Valohai project.
Invoke vh project link and enter the number (likely 1) of the project you'd like to link.
In a linked project directory, you can use vh execution run to start a run in the cloud.
For instance, using the Tensorflow example, vh execution run train will run the "Train model" step.
You can use vh execution run train --help (or vh ex r train --help; commands may be abbreviated to
unique prefixes) to see all of the available arguments. All parameters and inputs declared in the step's
configuration are available as command-line arguments.
By default, in the interest of reproduceability, executions are created from the newest commit
published in the repository the project is linked to. You can also create ad-hoc executions that upload
the contents of the project directory (sans any files ignored by .gitignore files) to the cloud before
creating an execution with the --adhoc flag; i.e. vh ex r --adhoc train.
To see the execution's progress in real time, you can add the --watch argument.
This is equivalent to invoking vh ex watch N, where N is the number of the execution.
In a linked project directory, you can start a pipeline run using vh pipeline run <pipeline-name>.
See vh --help. Each subcommand also has documentation of its own, also available via --help.
If you have any questions, please don't hesitate to contact us either via Github issues or the in-app support system!