Model architecture

Inside the Transformer: From Prompt to Next Token

Follow a prompt end-to-end through a GPT-style Transformer — tokenization, embeddings, stacked attention + MLP blocks, and temperature/top-k/top-p sampling.