And adjusting the pod allocation accordingly

Yeah, we're constantly trying to improve that... 🙂

Only when you try to delete these tasks?

What are the env vars passed to ES in k8s?

@<1539780284646428672:profile|PoisedElephant79> can't you just reinstall the original docker version you had running there?

In that case since I'm pretty sure these are indexed correctly, I can only assume download time slowness

is it a self-hosted server?

Hi @<1686547380465307648:profile|StrongSeaturtle89> , you can always start a clearml-agent on any machine (local, cloud, etc.) and configure it with the proper url and credentials to connect to your clearml account (if you're using app. clear.ml )

Hi @<1686547380465307648:profile|StrongSeaturtle89> , apologies for the delay. I think the best approach would either be having the local dataset available through some network share, or seperating your use-case to ETL, than training (the second can be triggered by new data being available)

Hi @<1686547380465307648:profile|StrongSeaturtle89> , usually you'd either run all locally or all remotely. What's your specific use case?

CrabbyKoala94 I think we'll need to see the server's logs to make sure it's not a server-side issue

Hi NastyOtter17 , I'm not sure I understand - can you explain what you see in the UI after running this as opposed to what you expect to see?

Indeed 🙂

i.e. hoping there already is a tool that does this

No such tool exists, but we've love to get a PR for such a tool, if you're going to draft something yourself 😉

As a "hack", you can do this:
from clearml.storage.helper import StorageHelper StorageHelper._s3_configurations._default_use_credentials_chain = True

Roughly how many artifacts? Can you open the browsers Dev tools (f12) in the Network section, compare again and see which http call does not return?

Hi @<1612982606469533696:profile|ZealousFlamingo93> , the extra configurations are for the clearml configuration file, and not for the AWS machine spec. Btw, are you using a self-hosted server, or app.clear.ml ?

Well, this is not currently supported in the autoscaler version. Starting from the next SaaS/enterprise autoscaler version, this should be supported there. As always, you're also welcome to add a PR 🙏

Hi @<1694157594333024256:profile|DisturbedParrot38> , I'm not sure exactly what is the cause for the slow performance, but 300 experiments is fairly small, is it possible the experiments themselves (ie the metadata returned by the call) are large?

The issue with automatically converting http to https is a browser restriction, unrelated to the ClearML webapp or fileserver

might be, although ClearML Server should allow it regardless of the user

Hi @<1657918715259260928:profile|JuicyPuppy11> , see here: None

Hi AbruptElephant13 , start_locally() should be available in ClearML 1.1.3rc0 (the latest pre-release) - can you try?

Quite correct - these are stored in mogodb.

Hi @<1590514572492541952:profile|ColossalPelican54> , this looks like an issue with the cpuinfo/ultralytics packages versions

BroadSeaturtle49 this is probably due to a different python version specified for running the task (than the one installed in the image you're trying to use) - I assume you're using different versions of python for development and for remote execution?

@<1554638179657584640:profile|GlamorousParrot83> the URL to the server should start with s3:// and not http, it should also include the port number (443 is case of a secure url)

I don't think -e . will work when running from the agent context

Oh, I see

Hi @<1556450111259676672:profile|PlainSeaurchin97> , can you share the full log and an example of how the requirements file looks?