拉脱维亚Tilde公司Tilde发射Tilde Open LLM,这是一个30B参数,多语言的人工智能模型,经过LUMI培训,符合欧盟法律,供当地使用。 Latvian firm Tilde launches TildeOpen LLM, a 30B-parameter, multilingual AI model trained on LUMI, compliant with EU laws and designed for local use.

拉脱维亚的AI公司Tilde已经推出TildeOpen LLM, 这是一个30亿参数的开源语言模型, Latvian AI firm Tilde has launched TildeOpen LLM, a 30-billion-parameter open-source language model trained on Europe’s LUMI supercomputer. 它支持34种语文,包括欧盟所有正式语文和若干其他语文,旨在提高较小的欧洲语文的绩效。 It supports 34 languages, including all EU official languages and several others, aiming to improve performance for smaller European languages. 该模型在一项关键的阅读理解测试方面优于其他模型,并且包括防止通过过滤克里姆林宫与克里姆林宫关系密切的消息来源的内容而出现虚假信息的保障措施。 The model outperforms others on a key reading comprehension test and includes safeguards against disinformation by filtering content from Kremlin-aligned sources. 它为在欧洲的当地部署设计,符合欧盟数据法和《大赦国际法》。 Designed for local deployment in Europe, it complies with EU data laws and the AI Act. 由12人小组在欧洲联盟委员会的支持下开发,根据CC-BY-4.0许可证免费获取。 Developed by a 12-person team with support from the European Commission, it is freely available under a CC-BY-4.0 license.