Training an LLM with own data feels like NPM

Why all the packages involved for fine tuning a model using a few gbs of text ?

I still don't know how to do this since it apparently needs me to know all kinds of shit from other tools beforehand.

w0000t