Hi @<1742355077231808512:profile|DisturbedLizard6> , not sure I get that, did you use torch.save (like in here ) or some other command to save the models? When running with the clearml-agent. you have a print of all the configurations at the beginning of the log, can you verify your values are as you configure it?

Additionally, which version of clearml , clearml-agent and `...

Hi @<1577468611524562944:profile|MagnificentBear85>

Thank you for bringing this up! We’re always excited to see contributions from the community, especially around areas like hyperparameter configuration. We’d be happy to consider a PR if you’re open to working on it! Our team encourages contributions 🙂

Did you check the relevant examples from our docs?
None
[None](https://c...

Hi @<1691983266761936896:profile|AstonishingOx62> , I think it's an issue related to the schema generated by the sdk, can you try adding _allow_extra_fields_ as True with the same call?

@<1691983266761936896:profile|AstonishingOx62> agreed, will push it forward

Hi @<1742355077231808512:profile|DisturbedLizard6> ,

Currently, only argparse arguments are supported for clearml-task , click is also support, but for now, with the python sdk.

Can I suggest using the sdk? It will do both, log it to the task and will upload it to any storage wanted, like in this example

With the API you can register an artifact to a task, but the upload will be done separately with the ClearML sdk (the sdk wrap the registration and upload, with some other things inside the upload_artifact function).

Hi @<1773158059758063616:profile|PanickyParrot17> ,

You can do that with ClearML pipelines . step 1 will be pulling the data, step 2 will store the data, step 3 will create the dataset.
The pipeline controller can have the parameters of which data to pull, the name of the created dataset and more, so you can run it again from the UI and just change the data source

All the steps can run with the clearml agent, and you can also specify using o...

Hi @<1669152726245707776:profile|ManiacalParrot65> ,

Yes, you can wrap the separate function with a decorator so the function will run as a separate step in the pipeline, and even can cache the step for multi runs.

You can also add the function without a decorator, as a step to the pipeline with PipelineController.add_function_step() .

You can read about it [here](https://clear.ml/docs/latest/docs/pip...

Hi @<1644147961996775424:profile|HurtStarfish47> ,

ClearML will automatically upload you model with the original name and data, if not mistaken, best.pt is given by default from the train function.

You can rename it after the training and upload it, something like:

import shutil

# Rename best model checkpoint after training
shutil.move("runs/train/my_model/weights/best.pt", "my_model.pt")

# upload with the StorageManager
model_path = "my_model.pt"
# Define your...

The auto magical will register the original model as an artifact, so the model that will be register it the original one, you can upload the model to your task as a model (so it will get a model id, like in this example ) or as a regular artifact (like in this example ).

Hi @<1747428509627715584:profile|CumbersomeDuck6> ,

Can you check if you have such graph in the task, under ARTIFACTS -> state ?

Hi @<1747066118549278720:profile|WhoppingToad71> , can you share the use case? You want to upload the file to some storage? Or upload to a task?

Hi @<1613344994104446976:profile|FancyOtter74> , I’m getting the same, will keep you posted once a fixed version is out