Presentation of LLM and SLM projects that used the resources of the Jean Zay 4 supercomputer as part of the Grand Challenges !

Dates: March 12, 2026

Location: On line

Registration : https://www.eventbrite.fr/e/1983873206164?aff=oddtdtcreator

The revolution in artificial intelligence is profoundly transforming our relationship to the world and to work. Within this transformation, large language models (LLMs) occupy a unique position. They are the first AI systems to have been widely adopted by the general public. For example, ChatGPT reached more than one million users in just five days.

While American models initially captured the public’s attention, France and Europe are now seeking to develop their own tools. Several factors explain this momentum, including the growing influence of these models on how ideas are formulated in everyday contexts and the need to counter the soft power exerted by such technologies. Cultural influence amplified by AI has therefore become a strategic issue. France, currently the fifth-largest user of generative AI, is thus further encouraged to develop LLMs based on French-language data corpora.

By acquiring the Jean Zay supercomputer in 2019, France equipped itself with a powerful instrument capable of multiplying its computational capacity in the service of science. The results have been significant: in 2025, GENCI supported nearly 1,700 AI projects, with a steady increase in the number of computing hours allocated and consumed. The main research themes include health, energy, and the environment. Industrial users, startups, and academic researchers are enthusiastic and already anticipate further increases in their demand for GPU resources. The deployment, underway since the end of 2025, of the AI Factory France project will broaden and strengthen this effort.

LLMs are no exception. Enabling both academic and industrial researchers to contribute to this dynamic is among the objectives of GENCI and the national computing centers. The Grand Challenges conducted on the new Jean Zay 4 system have therefore provided startups and researchers with the opportunity to conduct work on language models, whether LLMs or SLMs (small language models).

The three projects presented today raise several crucial issues: the question of languages, that of data, and whether these models should be open source or not :

• Anastasia Stasenko (Pleias) will present Pleias 1.0, the first family of language models trained exclusively on open data under a license that allows redistribution.
• Imed Laaridh (Zaion) will introduce Vigogne SLM, a collection of high-performance models optimized for the French language.
• Finally, Jean‑Pierre Lorré, Julie Hunter, Jérôme Louradour, and Olivier Gouvert from Linagora will present Lucie 7B, a multilingual foundation language model with a strong focus on French language and culture.