Welp, that seems to work!
However, the idea of executing the same script locally for creating a task and then executing it remotely feels a bit unintuitive.

Thought about it, but this is a thing I don't want to do manually.

Hi DullPeacock33 , I think what you're looking for is this:
https://clear.ml/docs/latest/docs/references/sdk/task#execute_remotely
This will initialize all the automagical stuff but won't require running the script locally.
What do you think?

I know.
Despite that, the execute_remotely() solution is good enough for me. Thanks!

DullPeacock33 , this is because you want the uncommitted changes attached to the task. Technically speaking you can do something like this yourself. A task's uncommitted changes are editable when it is in draft mode. So you could add the uncommitted changes via api or maybe even SDK.

That's the problem. ClearML has to detect the uncommitted changes somehow. This is done while the code itself is running or when running with execute_remotely() . Otherwise, someone has to do a git diff and push it into the task object(database)