AI Technology: Generative Language Modeling
Geo-Llama, supported by the EU's FFplus program and EuroHPC JU, is an innovative GeoAI model designed to facilitate natural language queries of geographic data, leveraging OpenStreetMap (OSM) for real-world spatial insights.
Utilizing a robust retrieval-augmented generation (RAG) system enriched with over 150,000 documents, Geo-Llama bridges the limitations of general LLMs in understanding geospatial relationships and OSM's OQL programming language.
Fine-tuning on EuroHPC’s advanced resources, the Leonardo Booster, will enable the development of 70B and 405B parameter models, addressing LLM challenges in spatial reasoning.
This approach aims to revolutionize geospatial interactions, unlocking AI-based geospatial insights across industries.