I recommend reading https://github.com/bkitano/llama-from-scratch over the article op linked.

It actually teaches you how to build llama iteratively, test, debug and interpret the training loss rather than just desribing the code.