so that it takes the auto logging but still able to run the with the previous specifications

ShinyPuppy47 do you have a small example we could take a look at?

unsure ya, I mean when I run the task logs do not get outputted into the clearml platform

hm SmugDolphin23 so I added this into the Task but I still do not get auto logging for a pytorch training task

got it will try

Facing similar issue
I need to use task.create but auto_connect_framworks= False
how can I do it?

Are you sure you passed add_task_init_call=True to task create?

Hi SoggyHamster83 ! Any reason you can't use Task.init?

been using this https://clearml.slack.com/archives/CTK20V944/p1671207870939899?thread_ts=1671206153.300929&cid=CTK20V944

Hi ShinyPuppy47 ,
Yes that is correct. Use Task.init for automagic logging

no longer using the create.

so curious if this is just a me misuderstanding and not using the .init/.create properly?

ShinyPuppy47 does add_task_init_call help your case? https://clear.ml/docs/latest/docs/references/sdk/task/#taskcreate

got it sweet, so I need to use Task.create to run on custom containers... so would the workaround here running Task.create and then running Task.init(continue_last_task=task.id) ?

but now I am just getting into weird situations were it seems like .create has the docker stuff I need but not logs and .init has the logs but not docker

ShinyPuppy47 the code that is being launched, does it call task.init?

got it ya that makes sense. I was using .init which worked great but
needed to be able to pass docker_args into the container running the task. Only way I saw that possible was with .create would like to execute it remotely would like the auto logs

Basically you should not use Task.create to log the current execution. It is used to create a Task externally and then enqueue it for remote execution. Make sense?

ShinyPuppy47 Try this: use task = Task.init(...) (no create ) then call task.set_base_docker

SmugDolphin23 so I ran that above like
` task = Task.init(project_name=project, task_name='data_upload')
task.set_base_docker(
docker_image="nvidia/cuda:10.2-cudnn7-runtime-ubuntu18.04",
docker_arguments=docker_args,
)

task.execute_remotely(queue) `and it works good, except there still does not seem to be autologging or even capturing print statements on the remote execution

any ideas?

So this should work, what is missing?