CostlyOstrich36 any updates on this?

the setup we have is that each ML person has an account on the ClearML server (you probably recall https://clearml.slack.com/archives/CTK20V944/p1650551097924099 ) and its own set of credentials. On the machine each ML person has a clearml.conf file in which the cache directory is set to be the same of everybody else

is https://clear.ml/docs/latest/docs/faq/#clearml-api the only place to see how to use the backend_api?

Hey UnevenDolphin73 I'm also interested in this feature. Currently trying it out

how do I restart just the apiserver? If I’m not wrong this command
sudo docker-compose -f /opt/clearml/docker-compose.yml downwill stop all the services

yes, I’m following these instructions https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_config#web-login-authentication

oh, I found it 🙂
cd /opt/clearml sudo docker-compose restart apiserver

I can’t see any useful line in the log. I’ll try the function. Thanks

It would be also good if I could get all the variants for a given task_id and metric

maybe I need to call this: https://clear.ml/docs/latest/docs/references/sdk/logger#set_default_upload_destination

yes

Hey CostlyOstrich36 thank you 🙂 I haven’t yet started using the agent to run a task. I’m in the phase of tracking the experiments. The setup that I have is that there’s an yml conda env file (generated via conda env export ) that includes conda and pip packages. I’d like to log the content of that file in ClearML. I’ve tried to use force_requirements_env_freeze but it doesn’t do what I’d hope for. Any suggestion?

Here it is https://github.com/allegroai/clearml/issues/671 🙂

give it few seconds an should work. That’s what I’ve experienced 🙂

FYI: I’ve added a possible fix to the issue https://github.com/allegroai/clearml/issues/671#issuecomment-1146640498

I’d like to programmatically (e.g. jupyter notebook) retrieve files/info related to a task from the server

this is a snippet of code:
task = Task.get_task('xxx') session = Session() res = session.send(events.GetDebugImageSampleRequest( task=task.id, metric="xxx", variant="xxx") ) print(res.response_data["event"]["url"])

just to clarify, you mean clearml api credentials (not the aws ones)?

SuccessfulKoala55 is this fixed in this release https://clearml.slack.com/archives/C03E7MNDG3C/p1651763847469039 ?

CostlyOstrich36 thank you 🙂 I suppose the issues isn’t necessary if someone is already working on a fix?

In [6]: type(conf["a"]["b"]["c"]["d"]) Out [6]: tuple

getting the same behaviour

Your suggestion CostlyOstrich36 works: I’ve added the line in the clearml.conf file.

However, now when I go in the Results -> Debug Samples tab, the s3 credential window pops up. Every time that I refresh the page 😞

however, I’m not able to get the audio excerpts uploaded as debug samples. Your suggestion above requires to use REST api via clearml.backend_api but I’m not very much familiar with it

some actions can be done using clearml.Task , for instance get a local copy of a model, pull down the scalars etc.

Is there a vital reason why you want to keep the two accounts separate when they run on the same machine?

I’ve already implicitly answered this, but to be more precise, having multiple users allows to know who ran the experiment 🙂

storage { cache { # Defaults to system temp folder / cache default_base_dir: "/scratch/clearml-cache" }this is it. I’ve only changed this bit

so, yes each user has its own credentials

yes, sure 🙂 thank you

I’ll put together some more info and raise an issue