Hi AppetizingPelican85 , how exactly did you configure and run the agent?

So you can either create a set of credentials from your own user and give that to the agent, or be nice, and simply define a user/password for the agent 🙂

FierceHamster54 in all reports I've seen concerning a similar error, it seems the problem was when trying to copy a folder from/to a bucket without specifying the recursive option (some were with the aws cli, but I suspect this is similar). Since the setup_aws_upload() attempts to upload a file, I suspect there's some confusion and it tries somehow to upload a folder, but does that as if it was a file... the actual implementation is providing boto3 with a Key and a Body, which should b...

Cool! what permission?

ElegantCoyote26 what clearml version are you using?

What do you mean by NodeJS API?

I think we'll need JuicyFox94 ’s input on this

@<1661542579272945664:profile|SaltySpider22> as part of the local task run, ClearML SDK will pick up the repository details and record them on the task, you only need to set up the agent with the required credentials that will allow it to clone the repository

In the browser's dev tools under Application/cookies try to delete the cookies and reload using Ctrl-F5

Hi RoughTiger69 , still getting these? I did not experience it

Hi @<1600661428556009472:profile|HighCoyote66> , I think you're right. However, this is an automated call and cannot involve an interactive user. You can test it by simply using curl <address> - that response should be a json paylpad

Updating that the helm chart has been updated - thanks JuicyFox94 !

Hi UnevenDolphin73 , basically the configuration loading is deferred, but once it's loaded it's not re-loaded. As long as you won't call any ClearML SDK method before setting the env vars, I think you should be fine

Hi WackyRabbit7 and ReassuredTiger98
While what Tim said was true, it only applies to the Python binary used to create the venv (i.e. the venv will still be created).
WackyRabbit7 , to specify an pre-existing venv you can use the CLEARML_AGENT_SKIP_PIP_VENV_INSTALL=<path-to-venv-binary> env var

Hi ThickChicken87 , if you mean while the server is unavailable (for any reason), I guess you are correct.

You mean the migration from Trains Server 0.15 to 0.16?

A step is changed while the pipeline is running?

If you ssh into that machine and into the venv, can you see if it inherits the system packages?

WackyRabbit7 what format are you getting, exactly?

When initializing your task, simply make sure to turn off the auto-connect for pytorch, using Task.init(..., auto_connect_frameworks={'pytorch': False})

SubstantialElk6 here you go 🙂 - https://clear.ml/docs/latest/docs/references/api/index

But are you trying to run the task remotely using the agent?

What would be the use case for triggering the auto scaler programmatically? I mean, I'd imagine you'd only have one (or a few) of those running at any given time, right?

Obviously, you can put whatever you want in the startup_bash_script

When you have experiments, you can access the reported metrics per experiment - that should allow you to decide which experiment is the best, and than you can take that experiment's model, WDYT?

auth.user

@<1546303293918023680:profile|MiniatureRobin9> which task exactly are you referencing?

Hi @<1523703012214706176:profile|GorgeousMole24> , I'm not sure I understand exactly what happened - can you perhaps attach some logs of the crashed experiment?

Hi ElegantCoyote26 ,
When a task is first executed (locally), ClearML will record the python version used to run it. When a ClearML Agent will run the task remotely, it will first try to find the exact same python version to run the task with. If the version can't be found, the Agent will try to run the task using the python version specified in it's python_binary configuration value (if defined, see https://github.com/allegroai/clearml-agent/blob/822984301889327ae1a703ffdc56470ad006a...

Hi @<1636537829553082368:profile|DistraughtSeahorse47> , ClearML Server 1.13 is still in the process of being released (not officially out yet), and the repository will be updated within the next few 🙂 (this is as part of our standard release process)