Image Classification (CIFAR-10) on Kaggle

https://d2l.ai/chapter_computer-vision/kaggle-cifar10.html



Why are these differences?
loss nan?
Code:https://github.com/StevenJokess/d2l-en-read/blob/moreme/Ch13_CV/kaggle_cifar10%20(1).ipynb

Hi! My test dataset has 10000 images starting from 40001.png
When I run the submission file and upload it on Kaggle, it gives me an accuracy of 71%, whereas in the graph after it is run, it shows validation accuracy as 91.2%. Any help here?