fix run_name format in tutorial notebooks and add missing dataset/model properties

[CuriousDolphin]

Minor Fixes

• Fixed run_name format standardization: Updated all training examples to use f"{model.name}_{train_dataset.name}" pattern consistently
• Added missing properties: Implemented name and task properties in MapDataset class and name property in FocoosModel class
• Fixed notebook cell types: Changed raw cells to code cells in training and inference tutorials for proper execution
• Updated documentation: Improved comments for sync_to_hub parameter and training configuration examples
• Minor notebook improvements: Updated dataset listing format and removed empty markdown cells
• Updated default export runtime type from ONNX_CUDA32 to TORCHSCRIPT_32
• Reload Processor in focoos models before training start
• Sync model weights after best_checkpoint restore
• Fix osx broken installation on new sh releases

[github-actions]

Tests	Skipped	Failures	Errors	Time
221	0 💤	0 ❌	0 🔥	19.369s ⏱️

[cursor]

Bug: Sync Issue on Successful Training

The after_train method now only syncs training results and artifacts to the hub when an exception occurs. This is because the sync logic was moved inside the exception handling block (if exc_type is not None). As a result, successful training completion does not trigger the final sync.

focoos/trainer/hooks/sync_to_hub.py#L71-L104

focoos/focoos/trainer/hooks/sync_to_hub.py

Lines 71 to 104 in ed28669

	exc_type, exc_value, exc_traceback = sys.exc_info()
	if exc_type is not None:
	logger.error(
	f"Exception during training, status set to TRAINING_ERROR: {str(exc_type.__name__)} {str(exc_value)}"
	)
	status = ModelStatus.TRAINING_ERROR
	self.model_info.status = status
	if self.model_info.training_info is not None:
	self.model_info.training_info.main_status = status
	self.model_info.training_info.failure_reason = str(exc_value)
	self.model_info.training_info.end_time = datetime.now().isoformat()
	if self.model_info.training_info.status_transitions is None:
	self.model_info.training_info.status_transitions = []
	self.model_info.training_info.status_transitions.append(
	dict(
	status=status,
	timestamp=datetime.now().isoformat(),
	detail=f"{str(exc_type.__name__)}: {str(exc_value)}",
	)
	)
	self.model_info.dump_json(os.path.join(self.output_dir, ArtifactName.INFO))
	self._sync_train_job(
	sync_info=HubSyncLocalTraining(
	status=status,
	iterations=self.iteration,
	training_info=self.model_info.training_info,
	),
	upload_artifacts=[
	ArtifactName.WEIGHTS,
	ArtifactName.LOGS,
	ArtifactName.INFO,
	ArtifactName.METRICS,
	],
	)

Fix in Cursor

---

Bug: Refactoring Error: Unhandled `None` in Iteration

The refactored logic in visualizer.py introduces a TypeError: 'NoneType' object is not iterable when the classes argument is None. The new else branch attempts to iterate over classes ([str(i) for i in classes]) when it is None, whereas the original code correctly handled this case by not attempting to process classes if it was None.

focoos/utils/visualizer.py#L489-L493

focoos/focoos/utils/visualizer.py

Lines 489 to 493 in ed28669

	labels = None
	if classes is not None and class_names is not None:
	labels = [class_names[i] if i < len(class_names) else str(i) for i in classes]
	else:
	labels = [str(i) for i in classes]

Fix in Cursor

---

BugBot free trial expires on June 12, 2025
Your team has used $0.00 of the $50.00 spend limit so far. Team admins can manage spend limit in the Cursor dashboard.

Was this report helpful? Give feedback by reacting with 👍 or 👎

[giuseppeambrosio97]

LGTM!