Transformers & DSPy: The Perfect Combo to Start with LLMs

Author(s): Rafael Guedes

Originally published on Towards AI.

A theoretical overview of the Transformer architecture, the novel concepts of LLaMA 3, Gemma, and Mixtral, and how to use these LLMs with DSPy

Who has never used ChatGPT? Probably every single one of us! However, we do not face one of the latest and most promising developments in artificial intelligence only when we use ChatGPT. Large Language Models (LLMs) have been implemented across different companies from different domains and we are likely exposed to them every day.

For example, customer service teams use this technology to quickly handle basic queries and let agents focus on more demanding issues. Marketing agencies use it to support their creative side when building campaigns or to understand customer sentiment in social media posts. Or, Spotify could have used this technology to create the lyrics through audio transcription.

With so many possible use cases and the level of exposure that we have, this article aims to provide a simple but detailed explanation of how the backbone architecture of LLMs works and what novel concepts companies like Meta, Mistral AI and Google introduced to this architecture with their own models, LLaMA, Mixtral and Gemma.

Finally, we provide a practical implementation in python using the library DSPy of these LLMs to tackle different use cases such as sentiment analysis, summarization, and RAG systems.

Figure 1: The world of LLMs (image generated by the author… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI