The Submit model run button appears on the Predict view.
Each model run requires a unique name. Unique names are generated by default, but you can enter a custom name if desired.
If there's a problem with the model run name, an error message describes the problem and the Submit model run button becomes inactive.
Adjust the name until the message disappears.
Select the Submit model run button to create predictions.
When you submit a model job, you're returned to the Catalog view with your selected data. A small banner appears while your model run is in progress.
To view model run progress, you can:
- Select View in notifications tab in the progress banner shown earlier.
- From the Labelbox app menu, open the Notification center to view current progress.
This example shows multiple model runs runs, including three in progress and one completed.
When finished with the notifications tab, select View in Catalog to return to the Predict screen. (You can also use your browser's Back button.)
When the model run completes, you can:
- View predictions results
- Use the the prediction results to prelabel your data by sending them to Project.
- Download the prediction results using Export v2 (coming soon)
It takes time to set a server up to process a model run, using a process similar to traditional provisioning: an instance is created to host your selected model, it's configured according to your settings, and each selected data row is processed.
In some cases, it's possible a model run may be able to use a server from an earlier model run. When this happens, the latter model run might appear to complete more quickly than the earlier run because the server doesn't need to be "spun up."
There are no visible indications showing start up times; the notifications tab provides current status.
In addition, each selected data row must be fetched and processed by the model. Expect the process to require an appropriate period of time.
Updated 2 months ago