Pick a sample image (or upload your own) and hit Run inference. You'll see every tensor the model produces — patch embedding, position add, each layer's Q/K/V, attention scores, softmax, residual adds, MLP — and the final ImageNet prediction.