Does this device agree with the trained model?

Runs the exported model here, on this handset, and compares the numbers against what PyTorch produced for the same input. Run it once on any device before you trust a count from it.