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