Thanks! I installed CUDA/CuDNN on the image and now the GPU is being utilized.

yall thought of everything. this fixed it! Having another issue now, but will post seperately

ok, i suppose that will have to do for now. thank you!

is there a way to get the instance's external IP address from clearml? i wouldve thought it would be in the info tab, but its not

This is what the instance state looks like, as logged by clearml:

yes, on my windows machine I am running:
cloned_task = Task.clone(source_task=base_task, name="Auto generated cloned task") Task.enqueue(cloned_task.id, queue_name='test_queue')I see the task successfully start in the clearml server. In the installed packages section it includes pywin32 == 303 even though that is not in my requirements.txt.

In the results --> console section, I see the agent is running and trying to install all packages, but then stops at pywin32. Some lines from t...

the 2nd option looks good. would everyone's credentials be displayed on the server though?

Windows 10 python 3.8 virtualenv

Also, this would change it globally. is there a way to set it for specific jobs and metrics?

We are using self-hosted clearMl w/ the following versions:

Worker CLEARML-AGENT version 1.1.2
The autoscaler instance Clearml-AGENT version: 1.2.3
ClearML WebApp: 1.2.0-153 Server: 1.2.0-153 API: 2.16
python pip package 1.3.2

this worked, thank you!

gotit. any solution where the credentials wouldnt be visible to everyone?

exactly

Awesome! any way to hear the talk w/o/ registering for the whole conference?

gotit. the instance logs showed
/var/log/syslog.1:May 5 03:25:27 ip-172-31-37-234 kernel: [53387.840425] python invoked oom-killer: gfp_mask=0x100cca(GFP_HIGHUSER_MOVABLE), order=0, oom_score_adj=0 /var/log/syslog.1:May 5 03:25:27 ip-172-31-37-234 kernel: [53387.840442] oom_kill_process+0xe6/0x120
I assume this is something I have to fix on my end (or increase instance memory). Does ClearML also happen to have solutions for this?

gotit. reg is indeed free 🙂

TimelyPenguin76 not sure what you mean by "as a service or via the apps", but we are self-hosting it. Does that answer the question?

Also, not sure what you mean by which "clearml version". How do we check this? The clearml python package is 1.1.4. Is that what you wanted?

is there a way to explicitly make it some install certain packages, or at least stick to the requirements.txt file rather than the actual environment

We were able to find an error from the autoscalaer agent:

Stuck spun instance dynamic_worker:clearml-agent-autoscale:p2.xlarge:i-015001a93e0910a09 of type clearml-agent-autoscale

2022-04-19 19:16:58,339 - clearml.auto_scaler - INFO - Spinning down stuck worker: 'dynamic_worker:clearml-agent-autoscale:p2.xlarge:i-015001a93e0910a09

Im thinking it may have something to do with
Using cached repository in "/home/ubuntu/.clearml/vcs-cache/ai_dev.git.42a0e941ddbf5c69216f37ceac2eca6b/ai_dev.git"We tried to reset the machines but the cache is still there. any idea how to clear it?

yes

no 64 bit. but do you mean the PC where I am spinning the task up or the machine where I am running the task

perfect. exactly what i was looking for!

relately, I just noticed that the GPU is not starting. This was in the logs:
2022-04-07 20:59:54.464854: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
Do we need to call a specific instance w/ CUDA preinstalled or does clearml take care of it?

the latter is an ec2 instance

yes. i think the problem is that its trying to recreate the environment the task was spun up on - which was on a windows machine - on a linux ec2 instance

Thanks this worked!

Thanks AgitatedDove14 . Does this go in the local clearml.conf file w/ each user's credentials, or in the conf file for the server?

huh. i really like how easy it is w/ the automated TB. Is there a way to still use the auto_connect but limit the amount of debug imgs?

SourLion48