LUCIE Unveiled: Building an Open-Source LLM for Europe, by OpenLLM Europe

Go behind the technical scenes to see how pre-training was provided for LUCIE, the 100% open source LLM created by the OpenLLM Europe community. This presentation covers: 1. Project background: motivations and challenges 2. Compiling training datasets: volume, duplication, filtering, ethics and rights management 3. Training on the Jean Zay supercomputer: optimisation and feedback 4. Model evaluation: specific metrics and tools 5. Alignment for the world of education: adapting to teaching needs and EdTechs 6. Outlook: future improvements and areas for research 7. Impact of the open source approach: collaboration, innovation and response to the OSAID Find out how the Open consortium, winner of France 2030, is pushing back the boundaries of open and trusted AI, from design to evaluation.