From tokenization to inference — a plain-language walk through how large language models are structured, trained, and named.