Multilingual innovation in LLMs: How open models help unlock global communication- Google Developers Blog

This post highlights the community's contributions to adapting Gemma models for multilingual LLM applications through a Kaggle competition. It showcases specific projects that fine-tuned Gemma for languages like Swahili, Traditional Chinese, Arabic, Italian, Ancient Chinese, Japanese, Hindi, Kazakh, and Old English. Techniques discussed include parameter-efficient fine-tuning (PEFT), retrieval augmentation (RAG), and instruction following. The post also mentions the upcoming Gemma 3 with pretrained support for over 140 languages.