One of the key industrial partners in this ambitious project is Diabolocom, an AI player and company that has recognized the importance of supporting innovations at the crossroads between academic knowledge and business requirements. Diabolocom’s involvement in the development of CroissantLLM underlines its commitment to responsible and accessible technological advances.
Diabolocom is very proud to have participated in this project through its PhD student, Nicolas Boizard, associated with CentraleSupélec’s MICS laboratory, under the supervision of Pierre Colombo, Céline Hudelot and Kevin El Haddad. Diabolocom’s contribution to the development of CroissantLLM underlines its commitment to responsible and accessible technological advances.
CroissantLLM features
CroissantLLM stands out remarkably from other generative AI models for its five key features, each reflecting a commitment to excellence, ethics and sustainability:
Sovereignty
Developed by a French research team, CroissantLLM takes advantage of the Jean Zay national supercomputer and is based on French public data sets, created for the occasion, guaranteeing total transparency and traceability. This model stands out as a reliable alternative capable of competing with larger models by requiring less computing power to operate efficiently.
Openness and accountability
The model is entirely open source, from algorithms to datasets, guaranteeing accessibility and transparency unprecedented in the AI field.
Mastery of the French language
Unlike many international models, which are mainly trained on Anglo-Saxon knowledge, CroissantLLM benefits from a balanced training approach between French and English content. This unique approach enables it to integrate and master the subtleties of the French language and culture, offering richer, more authentic comprehension and interaction.
Ethics
The research team has rigorously adhered to the ethical standards established by the recent AI Act, ensuring that the model adheres to the principles of responsible AI.
Frugality and performance
With just 1.3B of parameters, CroissantLLM runs efficiently on low-power devices like smartphones and personal computers, while being the most powerful Francophone model for its size.
This development represents the fruit of collaboration between leading academic institutions, such as CentraleSupelec, Sorbonne University, INESC-ID, Instituto Superior Técnico, Carnegie Mellon University, and the DATAIA Institute, as well as visionary industrial partners such as Diabolocom, Illuin Technology, Unbabel and EqualAI.
Professor Céline Hudelot, director of the MICS laboratory, underlines the importance of this synergy between academia and industry: “This work illustrates the importance of collaboration between academia and industry in advancing AI research. CroissantLLM is the direct result of this union of efforts, marking an important milestone in our quest for sovereign and ethical AI.”
The innovation driven by CroissantLLM and the support of partners like Diabolocom highlight a promising path for the future development of artificial intelligence, where technology not only advances, but does so in a way that respects and values our cultural heritage and ethical principles.
Read the scientific article on CroissantLLM.
Learn more about our AI with Diabolocom !
