since I've either added add_functional_step or add_step

I want to serve using Nvidia Triton for now.

after creating, I tried adding batch to it and got this error

Tagging AgitatedDove14 SuccessfulKoala55 For anyone available right now to help out.

Thanks for the help.

Also, do I have to manually keep track of dataset versions in a separate database? Or am I provided that as well in ClearML?

Agreed yeah

I'll try to see how to use the sdk method you just shared

for now installing venv fixes the problem.

1.1.5

I already have the dataset id as a hyperparameter. I get said dataset. I'm only handling one dataset right now but merging multiple ones is a simple task as well.

Also I'm not very experienced and am unsure what proposed querying is and how and if it works in ClearML here.

Any way to make it automatically install any packages it finds that it requires? Or do I have to explicitly pass them in packages?

Just to be absolutely clear.

Agent Listening on Machine A with GPU listening to Queue X.

Task enqueued onto queue X from Machine B with no GPU.

Task runs on Machine A and experiment gets published to server?

My current approach is, watch a folder, when there are sufficient data points, just move N of them into another folder and create a raw dataset and call the pipeline with this dataset.

It gets downloaded, preprocessed, and then uploaded again.

In the final step, the preprocessed dataset is downloaded and is used to train the model.

Also, since I plan to not train on the whole dataset and instead only on a subset of the data, I was thinking of making each batch of data a new dataset and then just merging the subset of data I want to train on.

Wait is it possible to do what i'm doing but with just one big Dataset object or something?

Ok since its my first time working with pipelines, I wanted to ask. Does the pipeline controller run endlessly or does it run from start to end with me telling it when to start based on a trigger?

Yeah, that would ensure its clarity.

I understand that storing data outside ClearML won't ensure its immutability. I guess this can be built in as a feature into ClearML at some future point.

And multiple agents can listen to the same queue right?

I know how to enqueue in using the UI. I'm trying to do it programatically.

So the api is something new for me. I've already seen the sdk. Am I misremembering sending python script and requirements to run on agent directly from the cli? Was there no such way?

Yeah

do I just post the issue on the main clearml repo?

I'm curious as to if this is buggy behavior or if it is expected?

then I use trigger_scheduler.start()

I'll do that.

they're also enqueued

But what's happening is, that I only publish a dataset once but every time it polls, it gets triggered and enqueues a task even though the dataset was published only once.

Thank you for the help.