Task cloning is also useful for many other things:
Basically if you want to change only one/several aspects of the task but leave the rest of the framework in your experiment the same.
You can use it for transfer learning, for example
You can use it for simple HPO - or use our example for more sophisticated HPO.
The premise of the system is that it creates an abstraction layer, so you can change anything from the UI without going back to the code.
Makes sense?

You can go to the UI and right-click the tasks, than select Dequeue

Where its enqueuing everything

what's an agent? it may be a powerful gpu that runs the task?

Finished running meaning the code has completed, but what do you see in the UI?

https://github.com/allegroai/clearml/blob/a29a655a6ea3424906392bafb8d2fab292397f3a/examples/automation/manual_random_param_search_example.py

this one

and what does it mean when a task status is "draft"

Basically it means the task was created. This is the initial Task state

So what is the purpose of task cloning? Its mostly useful for agents?

Exactly 😄

SweetGiraffe8 Don't think of the UI as the list of running (or to be run) experiments. For that you have the Workers & Queues screen. These are all the jobs on the queues which will be run by the clearml-agent
Best practice is to run all your experiments using the clearml-agent

I have a task which is running and cloned tasks, they all finished running but they are still pending

so i just should comment the enqueue of the task

The task status will return to Draft

Can you share the code that created the tasks? Or at least the part of it that created the tasks?

I took that from the example, cant find which one

Hi SweetGiraffe8 ,
What is exactly their status? Did they start running?

So your code probably enqueued them - they are essentially waiting in a queue for a ClearML Agent to pick them up for execution - I assume that's not what you wanted...?

See here: https://github.com/allegroai/clearml/blob/a29a655a6ea3424906392bafb8d2fab292397f3a/examples/automation/manual_random_param_search_example.py#L58

and what does it mean when a task status is "draft"

Cool!

You're right 🙂 - these are additional benefits 🙂

ExcitedReindeer30 Thanks Nir! That makes a lot of sense. In the example that I saw, I see that the parent task is only used as a template, which looks wierd at the UI cause it should not really be running..

I have a for loop running

SuccessfulKoala55 So what is the purpose of task cloning? Its mostly useful for agents?

Pending i.e. enqueued?

no, I dont have any agents..

This enqueues the task for execution by an Agent, so in your case nothing was running...

it may be a powerful gpu that runs the task?

Yes, that's an option 🙂
A ClearML Agent can run on any machine, and communicates with the server to obtain enqueued tasks and run them on the machine it's running on

what should I do?

yes