Comment by contrast

You seem to be talking about a production-grade model rather than building an LLM as an exercise? Or if not, why do you disagree with the article's example of building a small LLM for $100?