deploy_to_roboflow() reports success but model upload fails server-side ("Model Upload Failed" in UI) #1116

Search before asking

• I have searched the RF-DETR issues and found no similar bug report.

Bug

Calling model.deploy_to_roboflow(...) from a trained RFDETRLarge checkpoint completes successfully in the notebook — it loads the workspace/project and prints a deployment URL — but the model fails to process on the Roboflow side. The corresponding dataset version shows a Model Upload Failed banner in the Roboflow UI, and the version never receives a usable trained-model badge.

Environment

• rfdetr1.7.1 (installed with rfdetr[train,loggers,onnx]==1.7.1)
• roboflow 1.2.16
• Model: RFDETRLarge, resolution=800
• Runtime: Google Colab G4 GPU
• OS: Linux (Colab)

Minimal Reproducible Example

Train an RFDETRLarge model and save checkpoint_best_total.pth.
Load the checkpoint for deployment:

from rfdetr import RFDETRLarge

   model = RFDETRLarge(
       pretrain_weights="/path/to/checkpoint_best_total.pth",
       resolution=800,
   )

   model.deploy_to_roboflow(
       workspace="<WORKSPACE>",
       project_id="<PROJECT_ID>",
       api_key="<REDACTED>",
       version=<VERSION_ID>,
   )

Observe the notebook output reports success.

Open the dataset version in the Roboflow UI and observe Model Upload Failed.

Additional

It worked with previous versions...

Are you willing to submit a PR?

• Yes, I'd like to help by submitting a PR!