Corrupted SAM3 model package: missing sam_configuration.json

Describe your question/issue here! (delete this when you post)

  • Project Type: Sam3 workflow issues
  • Operating System & Browser: Chrome and safari on MacOS
  • Project Universe Link or Workspace/Project ID: vert_ramps
  • Do you grant Roboflow Support permission to access your Workspace for troubleshooting? (Yes/No): Yes

A fine-tuned SAM3 Workflow fails in Roboflow Preview and on deployment because the model package appears incomplete.

Workspace: owl-q3a4r
Workflow ID: vertramps-vvertramps-1-sam3-large-t3-logic
Model: owl-q3a4r/vert_ramps-3-sam3-large-t3
Model package ID: 8b8da2fe824240522a39f3cde41aafae
Error type: CorruptedModelPackageError
Missing file: sam_configuration.json
Deployment: sam3-test

The exact error says:
β€œModel package is incomplete. Could not find element sam_configuration.json.”

Can you repair/regenerate this SAM3 model package, or advise whether we need to retrain/export a new SAM3 model version? Below I included the full error for additional context.

This could also be related to the model accuracy collapsing to 0 during training. I stopped it around epoch 13 but it took a while to stop training.

Workflow Error

Could not load any of model package candidate for model owl-q3a4r/vert_ramps-3-sam3-large-t3. This may be caused several issues. If you see this warning after manually specifying model package to be loaded - make sure that all required dependencies are installed. If that warning is displayed when the model package was auto-selected - there is most likely a bug in `inference-models` and you should raise an issue providing full context of the event.Here is the summary of errors for specific model packages: * model_package_id=8b8da2fe824240522a39f3cde41aafae error=Model package is incomplete. Could not find element sam_configuration.json. If you attempt to run `inference-models` locally - inspect the contents of local directory to check for completeness of model package download - lack of files may indicate network issues. Verification of connectivity may be a good first step. If you prepared the model package manually - examine the correctness of the setup. If you run on managed serving - contact support if the issue is not ephemeral. FOR FURTHER SUPPORT error_type=CorruptedModelPackageError - VISIT Model Loading - Inference Models FOR FURTHER SUPPORT

Hi @wrayjacksonr ,

Can you check your dataset masks? I suspect you have some box-only annotations in your dataset - this would explain why training job failed

Thanks!

Grzegorz