Towards an Explorable Conceptual Map of Large Language Models

Large Language Models (LLMs) have revolutionized the current landscape of Natural Language Processing, enabling unprecedented advances in text generation, translation, summarization, and more. Currently, limited efforts have been devoted to providing a high-level and systematic description of their properties. Today's primary source of information is the Hugging Face (HF) catalog, a rich digital repository for researchers and developers. Although it hosts several models, datasets, and applications, its underlying data model supports limited exploration of linked information. In this work, we propose a conceptual map for describing the landscape of LLMs, organized by using the classical entity-relationship model. Our semantically rich data model allows end-users to answer insightful queries regarding, e.g., which metrics are most appropriate for assessing a specific LLM performance over a given downstream task. We first model the resources available in HF and then show how this map can be extended to support additional concepts and more insightful relationships. Our proposal is a first step towards developing a well-organized, high-level knowledge base supporting user-friendly interfaces for querying and discovering LLM properties.