kurakurai/Luth-LFM2-1.2B
Luth-LFM2-1.2B is a 1.2 billion parameter French fine-tuned version of the LFM2-1.2B model, developed by kurakurai in collaboration with Liquid AI. This model significantly improves French capabilities in instruction following, mathematics, and general knowledge while maintaining or slightly enhancing English performance. It is optimized for bilingual applications requiring strong French language understanding and reasoning.
Loading preview...
Overview
Luth-LFM2-1.2B is a 1.2 billion parameter language model developed by kurakurai in collaboration with Liquid AI. It is a French fine-tuned version of the LFM2-1.2B base model, trained on the Luth-SFT dataset. The fine-tuning process, detailed in a blog post and with scripts available on GitHub, focused on enhancing French instruction following, math, and general knowledge capabilities.
Key Capabilities
- Improved French Performance: Achieves higher scores on French benchmarks such as IFEval (59.95), GPQA-Diamond (28.93), MMLU (48.02), and Math500 (45.80) compared to its base model, LFM2-1.2B.
- Stable English Performance: Maintains or slightly improves English capabilities, notably in Math500 and IFEval, demonstrating effective cross-lingual transfer.
- Instruction Following: Shows enhanced instruction-following abilities in both French and English.
Good For
- Applications requiring strong French language understanding and generation.
- Bilingual (French-English) tasks where maintaining English proficiency is crucial.
- Instruction-tuned scenarios in French, including mathematical reasoning and general knowledge queries.