fr-gouv-coordination-ia/llm_leaderboard_fr · Open Source and Open Weight model categorisation

Pulling in my comment on @clem 's LinkedIn post just now, per @clefourrier 's suggestion:

Now we just need to add a filter for open-source (pleias, were it included), open-weight (only Chocolatine), and closed (Llama and Mistral under their proprietary and research-only licenses respectively)

Given there are several candidates for French Open Source AI models that include the dependencies required for reproducibility, and several others that provide the weights under open licenses, I'd love to work with you on starting to bifurcate Open Source and Open Weight models, and differentiate them from proprietary models. Here's some more context from the AI Action Summit this week, where we also ran an official AI Convergence challenge, albeit on Global AI Governance:

🇬🇧 Open Weight Definition Aims to Prevent Open Source AI “Open Washing” at AI Action Summit

🇫🇷 Dévoilée lors du sommet pour l'action IA, la Définition Open Weight vise à prévenir l’Open Washing de l’IA Open Source