Autoencoder Training Loss Doesn't Decrease Despite Cloned Code and Dataset

Question

My first post on stackoverflow so thanks for any responses! I'm learning about anomaly detection using autoencoders and found a useful looking github repo notebook link here

I've cloned the repo to my directory and downloaded the same csv dataset but im getting different results despite that i seem to be identical code.

the github autoencoders training loss decreases properly and then results in a correct confusion matrix when the test data is reconstructed. However when i run the same code my training loss doesnt decrease over the 10 epochs. this then results in huge false positives and essentially the model doesnt work at all.

can anyone think why this might be? the only things i can think of is maybe the libraries have changed all the repo is only from a year ago or perhaps its my computer (mac M2 chip)

appreciate any help , maybe im just missing something obvious
thanks