Swiss startup launches milestone French language model

Any company active in content generation in French that previously mostly had access to English-trained models can now leverage the largest and publicly available French language model released to date, available in beta version on cedille.ai. Launched today, Cedille is 4 times larger than the current biggest existing French language model. With its 6 billions parameters, Cedille can generate texts in French with unprecedented quality and competes with multilingual models such as GPT-3, the current reference in the domain. Cedille is especially powerful in generating articles, summarizing long reports, rewriting texts, translating contents, brainstorming ideas, building chatbots, and summing up complex concepts.

Active in the B2B business, Coteries aims to develop applications based on the Cedille model for media agencies and companies producing a lot of content in French. Speaking of the targeted market, Sébastien Flury, Coteries cofounder explains: “The platform is a test tool to demonstrate the potential of the model, not a "ready-to-use" tool. It is intended for any type of company that wants to improve or accelerate the creation of textual content.

Striving to offer a model that is free of unsafe content, Coteries’ Machine Learning team composed of Senior Engineer's Martin Müller and Florian Laurent has taken special care to filter the data Cedille is trained on. According to the start-up, all toxic and discriminatory content has been removed, as well as low-quality material. This process was made possible by a combination of Natural Language Processing and careful manual examination of sample data.

Key support by Google

The project was launched with support from the Google TRC Program and trained for several months on Tensor Processing Units (TPUs), custom chips made by Google to speed up AI computation. Relying on this infrastructure ensured that the training process was carbon neutral. This is a major achievement as most training processes for such models require a tremendous amount of energy, drastically increasing carbon emissions.

(ES)