TildeOpen LLM boosts low-resource European languages with curriculum learning

Researchers have introduced TildeOpen LLM, a 30-billion-parameter open-weight model designed to improve performance across 34 European languages. The model addresses data imbalance by employing dataset upsampling and a curriculum-based training schedule that shifts between uniform and natural language distributions. Evaluations indicate TildeOpen outperforms existing open-weight multilingual models, especially for Baltic, Finno-Ugric, and Slavic languages, with human assessments showing a significant reduction in linguistic errors. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enhances multilingual AI capabilities, particularly for underrepresented European languages, potentially lowering barriers for non-English content generation and comprehension.

RANK_REASON This is a research paper detailing the release of a new open-weight multilingual language model.

Read on arXiv cs.CL →

• TildeOpen LLM
• European languages
• English
• Baltic languages
• Finno-Ugric languages
• Slavic languages
• Toms Bergmanis
• Hugging Face
• arXiv

• model release
• paper
• other