Hi @<1523704157695905792:profile|VivaciousBadger56> , you can configure Task.init(..., output_uri=True) and this will save the models to the clearml file server

@<1523701070390366208:profile|CostlyOstrich36>

My training outputs a model as a zip file. The way I save and load the zip file to make up my model is custom made (no library is directly used), because we invented the entire modelling ourselves. What I did so far:

output_model = OutputModel(task=..., config_dict={...}, name=f"...")
output_model.update_weights("C:\io__path\...", is_package=True)

and I am trying to load the model in a different Python process with

mymodel = task.models['output'][0]
mymodel = mymodel.get_local_copy(extract_archive=True, raise_on_error=True)

and I get in the clearml cache a . training.pt file, which seems to be some kind of archive. Inside I have two files named data.pkl and version and a folder with the two files named 86922176 and 86934640 .

I am not sure how to proceed after trying to use pickle, zip and joblib. I am kind of at a loss. I suspect, my original zip file might be somehow inside, but I am not sure.

Sure, we could simply use the generic artifacts sdk, but I would like to use the available terminological methods and functions.

How should I proceed?

@<1523704157695905792:profile|VivaciousBadger56> I'm not sure I'm following you - is the issue not being able to upload to the ClearML server or to load the downloaded file?

@<1523701087100473344:profile|SuccessfulKoala55> I think I might have made a mistake earlier - but not in the code I posted before. Now, I have the following situation:

• In my training Python process on my notebook I train the custom made model and put it on my harddrive as a zip file. Then I run the code

output_model = OutputModel(task=task, config_dict={...}, name=f"...")
output_model.update_weights(weights_filename=r"C:\path\to\mymodel.zip", is_package=True)

2. I delete the "C:\path\to\mymodel.zip", because it would not be available on my colleagues' computers.

3. In a second process, the model-inference process, I run

mymodel = task.models['output'][-1]
mymodel = mymodel.get_local_copy(extract_archive=True, raise_on_error=True)

and get the error

ValueError: Could not retrieve a local copy of model weights 8ad4db1561474c43b0747f7e69d241a6, failed downloading

I do not have an aws S3 instance or something like that. This is why I would like to store my mymodel.zip file directly on the ClearML Hosted Service. The model is around 2MB large.

How should I proceed?

@<1523704157695905792:profile|VivaciousBadger56> It seems like whatever you pickled in the zip file relies on some additional files that are not pickled.

@<1523701083040387072:profile|UnevenDolphin73> : How do you figure? In the past, my colleagues and I just shared the .zip file via email / MS Teams and it worked. So I don't think so.

@<1523701083040387072:profile|UnevenDolphin73> : I do not see any way to download the model manually from the web app either. All I see is the link to the file on my harddrive (see shreenshot).

The second process says there is not file at all. I think, all that happened is that the update_weights only uploaded the location of the .zip file (which we denote as a .model file) on my harddrive, but not the file itself.

Yes, you're correct, I misread the exception.
Maybe it hasn't completed uploading? At least for Datasets one needs to explicitly wait IIRC

@<1523701083040387072:profile|UnevenDolphin73> : I do not get this impression, because during update_weights I get the message

2023-02-21 13:54:49,185 - clearml.model - INFO - No output storage destination defined, registering local model C:\Users..._Demodaten_FF_2023-02-21_13-53-51.624362.model

It should store it on the fileserver, perhaps you're missing a configuration option somewhere?

I am not sure if it the fact the name of the file ends with .model is an issue - but that would be somewhat crazy design...

I wouldn't put past ClearML automation (a lot of stuff depend on certain suffixes), but I don't think that's the case here hmm

missing a configuration option

Which one, where? Any idea? I did not set output_uri - do I have to do that?

I am refering to

• None
• None

@<1523701083040387072:profile|UnevenDolphin73> : If I do, what should I configure how?

Well you could start by setting the output_uri to True in Task.init .

Unbelievable! That worked.

We have the following, works fine (we also use internal zip packaging for our models):

model = OutputModel(task=self.task, name=self.job_name, tags=kwargs.get('tags', self.task.get_tags()), framework=framework)
model.connect(task=self.task, name=self.job_name)
model.update_weights(weights_filename=cc_model.save())

Heh, good @<1523704157695905792:profile|VivaciousBadger56> 😁
I was just repeating what @<1523701070390366208:profile|CostlyOstrich36> suggested, credits to him

It is documented at None ... super deep in the code. If you don't know that output_uri in TASK's (!) init is relevant, you would never know...

But, I guess @<1523701070390366208:profile|CostlyOstrich36> wrote that in a different chat, right?

@<1523701083040387072:profile|UnevenDolphin73>

Heh, well, John wrote that in the first reply in this thread 🙂
And in Task.init main documentation page (nowhere near the code), it says the following -

@<1523701083040387072:profile|UnevenDolphin73> : I see. I did not make the connection that output_uri=True is what I was missing. I thought this was the default. But the default is actually "None", which is different than "True".

Exactly 🙂

FWIW, we prefer to set it in the agent’s configuration file, then it’s all automatic

But we do use S3

Do you mean "exactly" as in "you finally got it" or in the sense of "yes, that was easy to miss"?

Either? 🙂

🙂

The documentation is messy, I’ve complained about it the in the past too 🙈