Ultralitics HUB can retrieve Roboflow dataset to train

[santiago-unxdigital]

Search before asking

• I have searched the HUB issues and found no similar bug report.

HUB Component

Datasets, Training

Bug

When i try to train a model into hub and the dataset come from roboflow, the value of data parameter into the yml file created from hub is NULL.

This is a recurrent error, no matters what dataset im using, or model, no matter what device im using to train (colab, my laptop, etc)

The error is de Null into the data parameter into the yml file created by hub.

Environment

No response

Minimal Reproducible Example

No response

Additional

No response

[sergiuwaxmann]

@santiago-unxdigital Hello!
This issue occurs when the Roboflow dataset isn't exported.
When you create your model in Ultralytics HUB, we request Roboflow to export the dataset, and I noticed that sometimes the export takes a while.

[santiago-unxdigital]

yes, i know that. but the issue is into hub. the data value is NULL. roboflow already provide the url to get the dataset, as you can see in the attached image

[sergiuwaxmann]

@santiago-unxdigital I can't see the attached image.

[santiago-unxdigital]

[sergiuwaxmann]

@santiago-unxdigital Yes, the screenshot in from the Roboflow platform. Unfortunately, they have a different export for the Ultralytics HUB integration which has the issue I explained above.
Maybe you can share your model ID with me so that I can check if the export for the Ultralytics HUB integration is ready or not? You can find the model ID in the model's URL.

[santiago-unxdigital]

'https://hub.ultralytics.com/models/JoPCqWmE0vxbNABuM5pz'

[sergiuwaxmann]

@santiago-unxdigital Just checked, and indeed the issue is that the export for the Ultralytics HUB integration is not ready.
You could force Roboflow to re-trigger the export by creating a new model in Ultralytics HUB using the same dataset.

[santiago-unxdigital]

i did! a lot of times! there any way to enter maually the data value? as a custom value of data parameter?

the zip from roboflow is https://app.roboflow.com/ds/j6OrgaYiZp?key=TjjUqndDOT

[sergiuwaxmann]

@santiago-unxdigital Unfortunately, there is no way to do this now. Additionally, this is not part of our feature roadmap, but I have noted this feature request, and we might work on it in the future.
I suggest reporting this issue to Roboflow, as there is nothing wrong on our end related to Roboflow training.

[santiago-unxdigital]

i disagree with you. the issue is in hub, the NULL value for data parameter is your problem. this is a bug for hub

[sergiuwaxmann]

@santiago-unxdigital As stated above, I have noted your feature request: allowing users to specify/customize the data parameter when training. Once again, the data value is null because the Roboflow export for the Ultralytics HUB integration is not successful or slow.
I will prioritize highlighting in the UI the Roboflow dataset export status for the next Ultralytics HUB release.

[sergiuwaxmann]

@santiago-unxdigital Funny enough, I can see the export is ready now. Please try to start training again.

[santiago-unxdigital]

same thing , data=None

[sergiuwaxmann]

@santiago-unxdigital Can you try with this model: https://hub.ultralytics.com/models/JoPCqWmE0vxbNABuM5pz? This is the one I monitored and noticed that was updated by Roboflow.

[santiago-unxdigital]

4 frames
/usr/local/lib/python3.10/dist-packages/ultralytics/hub/session.py in load_model(self, model_id)
75 self.model = self.client.model(model_id)
76 if not self.model.data: # then model does not exist
---> 77 raise ValueError(emojis("❌ The specified HUB model does not exist")) # TODO: improve error handling
78
79 self.model_url = f"{HUB_WEB_ROOT}/models/{self.model.id}"

ValueError: ❌ The specified HUB model does not exist

[sergiuwaxmann]

@santiago-unxdigital Did you take the training code snippet from the model page (https://hub.ultralytics.com/models/JoPCqWmE0vxbNABuM5pz)? Asking as the error you shared indicates that the model ID in the training code is invalid.