Luth-LFM2 are French fine-tuned versions of LFM2 models by Kurakura AI, trained on the Luth-SFT dataset. The models have improved their French capabilities in instruction following, math, and general knowledge. Additionally, their English capabilities have remained stable.