Inference needs sample data — images or tensors shaped the way your model expects.
Use a bundled sample set that matches the model, or provide your own. The Hub checks that the inputs line up with the model’s contract so the run doesn’t fail for the wrong shape.
When inputs look good, continue to Run.