Manage Runs

Monitor active training sessions, stop runs in progress, and delete completed runs in Datature Vi.

A run in Datature Vi is a single training session: one execution of a workflow that produces metrics, logs, and (if it completes) a trained model. This section covers how to track runs while they are active, stop them when you need to, and clean them up when they are done.

Run operations

Monitor A Run

Track real-time training progress, read loss charts, and catch configuration errors before they consume your compute credits.

Kill A Run

Stop an active training session immediately to free GPU resources or fix a configuration error.

Delete A Run

Permanently remove a completed, failed, or killed run along with its metrics, logs, and trained model.

Run lifecycle

Runs move through these states:

Run lifecycle

State
Meaning
Available actions
Queued
Waiting for GPU resources
Monitor, Kill
Starting
Allocating hardware and loading weights
Monitor, Kill
Running
Actively training
Monitor, Kill
Completed
Training finished successfully
View metrics, Download model, Delete
Failed
Stopped due to an error
View logs, Delete
Killed
Manually stopped
View partial progress, Delete

Active runs (Queued, Starting, Running) must be killed before they can be deleted. Completed runs cannot be killed.

Next steps

Monitor A Run

Watch training progress in real time and catch errors early.

Kill A Run

Stop a run that has a configuration error or is consuming credits unnecessarily.

Delete A Run

Remove completed or failed runs to keep the project list clean.