Luth：小语言模型和跨语言迁移的高效法语专业化
Luth: Efficient French Specialization for Small Language Models and Cross-Lingual Transfer
AI总结：
Luth通过定向后训练提升法语小语言模型性能，同时保留英语能力，并通过模型合并增强跨语言迁移效果。
摘要：
大型语言模型的景观仍以英语为中心，导致其他主要语言，如法语，在小语言模型（SLMs）的背景下存在显著性能差距。现有的多语言模型在法语上的表现明显低于英语，且针对法语的高效适应方法研究仍有限。为了解决这一问题，我们引入了Luth，一个法语专用的SLM家族：通过针对精选、高质量法语数据进行定向后训练，我们的模型在多个法语基准测试中优于所有同等规模的开源模型，同时保留其原有的英语能力。我们进一步表明，策略性的模型合并增强了两种语言的性能，使Luth成为法语SLM的新状态-of-the-art，并为未来的法语研究建立了稳健的基准。
英文摘要：
The landscape of Large Language Models remains predominantly English-centric, resulting in a significant performance gap for other major languages, such as French, especially in the context of Small Language Models (SLMs). Existing multilingual models demonstrate considerably lower performance in French compared to English, and research on efficient adaptation methods for French remains limited. To address this, we introduce \textbf{Luth}, a family of French-specialized SLMs: through targeted post-training on curated, high-quality French data, our models outperform all open-source counterparts of comparable size on multiple French benchmarks while retaining their original English capabilities. We further show that strategic model merging enhances performance in both languages, establishing Luth as a new state of the art for French SLMs and a robust baseline for future French-language research.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
