Programming Large Language Models with Azure Open AI: Conversational programming and prompt engineering with LLMs

APPENDIXFRANCESCO ESPOSITO & MARTINA D’ANTONI

Inner functioning of LLMs

Unlike the rest of this book, which covers the use of LLMs, this appendix takes a step sideways, examining the internal, mathematical, and engineering aspects of recent LLMs (at least at a high level). It does not delve into the technical details of proprietary models like GPT-3.5 or 4, as they have not been released. Instead, it focuses on what is broadly known, relying on recent models like Llama2 and the open-source version of GPT-2. The intention is to take a behind-the-scenes look at these sophisticated models to dispel the veil of mystery surrounding their extraordinary performance.

Many of the concepts presented here originate from empirical observations and often ...