nvflare.edge.widgets.evaluator module

class GlobalEvaluator(model_path: str | Module | Dict, eval_frequency: int = 1, torchvision_dataset: Dict | None = None, custom_dataset: Dict | None = None)

Bases: Widget

Initialize the evaluator with either a dataset path or custom dataset.

Parameters:

• model_path –

  PyTorch model to evaluate. Can be: - An nn.Module instance - A string class path (e.g., “mymodule.MyModel”) - A dict config with ‘path’ and optional ‘args’ keys

  (e.g., {“path”: “mymodule.MyModel”, “args”: {“num_classes”: 10}})

• eval_frequency – Frequency of evaluation (evaluate every N rounds)

• torchvision_dataset – Torchvision dataset (for standard datasets like CIFAR10)

• custom_dataset – Dictionary containing ‘data’ and ‘labels’ tensors

evaluate(_event_type: str, fl_ctx: FLContext)