Reproducing Task Runs

You can reproduce task runs on local or remote machines in one of the following ways:

• Cloning any task - Make an exact copy, while maintaining the original task
• Resetting a task whose status is not Published - Delete the previous run's logs and output

After cloning or resetting, enqueue the reset or newly cloned task for execution by a worker.

Tasks can also be modified before execution. See Tuning Tasks.

Cloning

To clone a task:

1. In the task table, right-click the task to reproduce and click Clone.
2. In the Clone Task modal, set the following:

   • Project - The project where the task will be saved

   • Name - The name for the new task.

     New task name template

     By default, the new task is named Clone of <original_task_name>.

     This default name can be customized:

     • In ClearML Enterprise, via UI Settings page > Task clone name template
     • In ClearML Open Source, via server configuration setting. For more information, see Default Task Clone Name.

   • Set <cloned_task> as parent - Select to set this task as the new task's parent

   • Clear Dataviews (Enterprise feature) - If selected, the new task will start with no Dataviews

   • Force Original Packages - If selected, use the parent task's original pip/original conda for the new task’s Python Packages.

   • Description (optional)

3. Click Clone

note

By default, the new task's parent task is set to the original task's parent, unless the original task does not have a parent, in which case the original task is set as the parent. Select Set <cloned_task> as parent to force the original task to become the clone's parent.

Resetting

To reset a task:

1. In the task table, right-click the relevant task and click Reset.
2. In the Reset Task modal, if you want the task's artifacts and debug samples to be deleted from the ClearML file server, click the checkbox
3. Click Reset

Final Steps

At the end of the process you are left with a Draft task, meaning that it is editable.

Re-execute the new task:

1. If desired, modify the task's configuration (see Tuning Tasks).

2. Enqueue the task for execution. Right-click the task > Enqueue > Select a queue > ENQUEUE.

   note

   Make sure that a ClearML Agent has been assigned to the selected queue

A ClearML Agent will fetch the task from the queue and execute it. The task can now be tracked and its results visualized.