Okay now let's try: EDIT
docker run -t --rm nvidia/cuda:10.1-base-ubuntu18.04 bash -c "echo 'Binary::apt::APT::Keep-Downloaded-Packages \"true\";' > /etc/apt/apt.conf.d/docker-clean && apt-get update && apt-get install -y git python3-pip && python3 -m pip install trains-agent && python3 -m trains-agent --help"

Thanks @<1550289509273309184:profile|CooperativeBeetle24> !
Is this an error with the CLI not working with a certain version of numpy ?
Any chance you can PR the fix ?

None

But I believe it would be harder for our team to detect and respond to failures in the event handler functions if they were placed there because it seems unclear how we could use our existing systems and practices to do that.

Okay I think this is the issue, handler functions are not "supposed" to fail, they are supposed to trigger Tasks, these can fail.
e.g.:
Model Tag Trigger -> handler function creates a Task -> Task does something, like build container, trigger CI/CD etc -> Task...

Yes that's the part that is supposed to only pull the GPU usage for your process (and sub processes) instead of globally on the entire system

CrookedWalrus33 I found the issue, this is only failing with Python 3.6.
Let me check something

Hi FunnyTurkey96
Which pip are you using, basically pip changed the dependency resolver after 20.1
Change: https://github.com/allegroai/clearml-agent/blob/aede6f4bac71c8fc56e7cf982318a48527953a3c/docs/clearml.conf#L57
pip_version: "<20.2"See if that helps

SmarmySeaurchin8 regarding the original question:
task.set_project(project_id)Task.get_projects() to get all the project names/ids

Hi @<1523704157695905792:profile|VivaciousBadger56>
You should replace

task.mark_completed()

with:

task.close()

To your point

parameters = task.connect(parameters)

Will be retrieved with:

task.get_parameters()

fyi:
connect_configuration -> get_configuration_objects

Hi FierceFly22
You called execute_remotely a bit too soon. If you have any manual configuration, they have to be called before, so they are stored in the Task. This includes task.connect and task.connct_configuration.

Oh that makes sense.
So now you can just get the models as dict as well (basically clearml allows you to access them both as a list, so it is easy to get the last created, and as dict so you can match the filenames)
This one will get the list of models
print(task.models["output"].keys())Now you can just pick the best one
model = task.models["output"]["epoch13-..."] my_model_file = model.get_local_copy()

This means that if something happens with the k8s node the pod runs on,

Actually if the pod crashed (the pod not the Task) k8s should re spin it, no?

I also experience that if a worker pod running a task is terminated, clearml does not fail/abort the task.

From the k8s perspective, if the task ended (failed/completed) it always return with exit code 0, i.e. success. Because the agent was able to spin the Task. We do not want Tasks with exception to litter the k8s with endless r...

JitteryCoyote63 Hmmm in theory, yes.
In practice you need to change this line:
https://github.com/allegroai/clearml/blob/fbbae0b8bc933fbbb9811faeabb9b6d9a0ea8d97/clearml/automation/aws_auto_scaler.py#L78
` python -m clearml_agent --config-file '/root/clearml.conf' daemon --queue '{queue}' {docker} --gpus 0 --detached

python -m clearml_agent --config-file '/root/clearml.conf' daemon --queue '{queue}' {docker} --gpus 1 --detached

python -m clearml_agent --config-file '/root/clearml.conf' d...

Still figuring out, what is the best orchestration tool,which can run this end-2-end.

DeliciousBluewhale87 / PleasantGiraffe85 based on the scenario above what is the missing step that you need to cover? Is it the UI presenting the entire workflow? Or maybe the a start trigger that can be configured ?

Oh I see, that kind of make sense
I think this is the section you should use:
None
But instead of the clearml-services container you should use the regular container (or just have it installed as part of the entry-point on any ubuntu based container)
Notice the important parts here are:
[None](https://github.com/allegroai/clearml-server/blob/6a1fc04d1e8b112fb334c8743d...

JitteryCoyote63 what's the clearml version ?
Are you always seeing the "model uploaded completed" message ?
What's the python version you are using?

@<1523701079223570432:profile|ReassuredOwl55>

Hey, here’s a quickie – is it possible to specify different “types” of input parameters (“Args/…“) such that they are handled nicely on the front end?

You me cast / checked in the UI ?

Anyhow from your response is it safe to assume that mixing in 

 code with the core ML task code has not occurred to you as something problematic to start with?

Correct 🙂 Actually we believe it makes it easier, as worst case scenario you can always run clearml in "offline" without the need for the backend, and later if needed you can import that run.
That said, regrading (3), the "mid" interaction is always the challenge, clearml will do the auto tracking/upload of the mod...

Hi GreasyLeopard35

I try to resume a stopped or aborted parameter optimization experiment,

How are you continuing the HPO? are you runing everything locally? is this with an agent? are you seeing the '[0, 0]' value on the configuration when launching the HPO or when continuing it ?

RoughTiger69 I think you need the latest version (+1.3.0 with UI support)
If you are using an older version, you need to specify that you are continuing an execution (Change the "Configuration/Args/continue_pipeline" to True)
EDIT: clearml 1.3.x will work with clearml-server 1.2

Hi StraightCoral86

When I run an experiment using 

Task.create()

 ,

Use Task.init 🙂
Task.create is meant to create an extranl Task (i.e. Job) ins the system, Not to auto-gernerate a job from the running code. Make sense ?

For now we've monkey-patched it to our usecase:

LOL, that's a cool hack

That gives us the benefit of creating "local datasets" (confined to the scope of the project, do not appear in

Datasets

tabs, but appear as normal tasks within the project)

So what would be a "perfect" solution here?
I think I'm missing the point on why it became an issue in the first place.
Notice that in new versions Dataset will be registered on the Tasks that use them (they are already...

(But in venv mode is also hangs the same way)

Hmm this is strange, could it be you are running out of storage ?

Yey!

agent.cuda_driver_version = ...
agent.cuda_runtime_version = ...

Interesting idea! (I assume for reporting only, not configuration)

... The agent mentionned used output from nvcc (2) ...

The dependencies I shared are not how the agent works, but how Nvidia CUDA works 🙂
regrading the cuda check with nvcc , I'm not saying this is a perfect solution, I just mentioned that this is how this is currently done.
I'm actually not sure if there is an easy way to get it from nvid...

Do you accidentally know if there are any plans for an implementation with the logger variable, so that in case of something it would be possible to write to different tables?

CheerfulGorilla72 what do you mean "an implementation with the logger variable" ? pytorch-lighting defaults to the TB logger, which clearml will automatically catch and log into the clearml-server, you can always add additional logs with clearml interface Logger.current_logger().report_???
What am I mis...

Hi JitteryCoyote63
The NVIDIA_VISIBLE_DEVICES is set automatically for the process the trains-agent spins, so from your code, it is transparent, you can only "see" GPU 0.
(Obviously not using docker you can forcefully change the OS environment in runtime, but you should avoid that ;))

Correct, (if this is running on k8s it is most likely be passed via env variables , CLEARML_WEB_HOST etc,)

SubstantialElk6 try to add -e CLEARML_AGENT_EXTRA_PYTHON_PATH=/code/app/flair
It should add it to the runtime pythonpath
(to the BASE DOCKER IMAGE on the Task itself)

What I'm trying to do is to filter is between two datetimes...Is that possible?

could you expand ?

It is the folder the clearml creates and the folder we create ourself to store the predictions

I see... If that is the case, the only solution I can think of is manually uploading the files with StorageManager(...) then get the url, and register it as debug_media or artifact:
logger.report_media("image", "type a", iteration=iteration, url=" ") task.upload_artifact('a link', artifact_object=' ')