ASP projects

ChatIMPACT

Nowadays, the most relevant source of information on LLMs and LLM- related items, such as datasets or metrics, is the Hugging Face (HF) portal (Hugging Face – The AI community building the future, 2024). Despite its popularity, HF has several limitations: some details regarding models or datasets are not available on the platform, and some queries linking different entities cannot be performed.

The goal of the chatIMPACT project is to bridge this gap, building a platform that allows users to easily access the knowledge they need in the LLMs domain. The main outcome of our work is a proof of concept achieved in two steps:

  • Designing a comprehensive conceptual map of LLMs and LLMs related entities, namely datasets used for training, metrics to evaluate models and tasks that the models could solve.
  • Implementing the conceptual map in the form of an interactive, easy-to-navigate web interface accessible to different types of users.