Michel-Marie MAUDET
Other Social: https://www.linkedin.com/in/mmaudet/
github profile: https://github.com/mmaudet
Michel-Marie MAUDET has led LINAGORA (https://www.linagora.com/) as its Managing Director and co-founder for almost 25 years. He is a pioneer in the world of the free software in France and Europe since thirty years.
In June 2023, he started the OpenLLM France (https://www.openllm-france.fr/) community. This community brings together experts to share knowledge and work together on developing a French, sovereign, and truly Open Source LLM. For us, truly Open Source requires three conditions :
• Open Source code : source code of the code model, pre training tool released under OSI compliant Open Source Licence
• Open Model : licence and user agreement without any restrictions on who may use it and for what
• Open Training Datasets : 100% of training data must be publicly available in a format that allows for investigation of the model's biases (preferences) and for retraining
During the FOSDEM 2024, the OpenLLM France community has been changed to OpenLLM Europe in order to make possible the creation of digital commons in the field of Generative AI at European level by federating other national initiatives.
In the AI area, we need Digital Bioversity : Michel-Marie and LINAGORA work on this challenge by continuing to support open standards and aiming for an open, ethical, and responsible digital future.
Accepted Talks:
apt install lucie --from-source : LUCIE, the Fully "Libre LLM" you can build, hack and trust !
As AI becomes a core part of our digital infrastructure, the need for fully Free and auditable models is critical. LUCIE-7B is one of the first large language models trained entirely from scratch — with every component released under a libre license: model weights, training code, tokenizer, and even the full dataset.
https://huggingface.co/collections/OpenLLM-France/lucie-llm-67099ba7b992dee2c32b1f92 https://github.com/OpenLLM-France/Lucie-Training
This talk will show how LUCIE aligns perfectly with the Debian philosophy: transparency, reproducibility, community-driven development, and digital sovereignty. We’ll also discuss how LUCIE could integrate with Debian workflows, and invite the community to help build a fully libre AI stack — from inference on local hardware to distributed training setups.
Why it matters for Debian - 100% Libre: all assets released under Open Source licenses, including training data. - Reproducibility: training recipes and logs available for audit and replay. - modularity: LUCIE components can be packaged, extended, and improved by the Debian community. - sovereignty: a European-led alternative to opaque, proprietary AI systems.
We’ll explore how LUCIE fits naturally into Debian’s ethos of transparency, auditability and user freedom, and how you can help package, deploy, or even fine-tune it… the Debian way — from source to inference.