Lingua Custodia, Unbabel, Tilde, Textgain Win EC’s Large AI Grand Challenge

Lingua Custodia、Unbabel、Tilde、Textgain赢得EC大型人工智能挑战赛

2024-07-17 08:10 slator

本文共346个字,阅读需4分钟

阅读模式 切换至中文

The European Commission (EC) announced on June 26, 2024, the winners of the Large AI Grand Challenge contest, launched in 2023 as part of the European Union (EU) AI-BOOST project. AI-innovative startups and small and medium-sized enterprises (SMEs) were invited to submit proposals from November 16, 2023, to February 13, 2024. Participating companies were competing for the development of a foundation model using one of the EU High-Performance Computing (HPC) sites, part of the EuroHPC Joint Undertaking (EuroHPC JU). EuroHPC JU is a collaborative effort between the EU and participating Member States created to foster the EU’s advancement in AI and supercomputing (as part of the GenAI4EU initiative). The winners are France-based Lingua Custodia, which specializes in AI and natural language processing (NLP) for the finance sector; Unbabel, a language technology and services company based in Portugal; Latvia’s Tilde, a language technology, MT, and Balto-Slavic languages company; and Textgain, an AI start-up based in Belgium focused on predictive text and speech analysis. The four winners competed with 90 other companies to win EUR 250k and access to 2 million compute hours each at one of the two EU HPC sites. These language AI technology companies will also have the opportunity to become part of a network that collaborates with European Commission DG Connect – CNECT AI and Robotics and EuroHPC-JU. The main requirements for the foundation model are that it must be trained from scratch using state-of-the-art methods and have 30 billion parameters at a minimum. In their proposals, the companies had to offer a rationale for the model’s relevance and the use of HPC. They also had to demonstrate internal expertise in training foundation models using HPC systems along with a plan for using said EuroHPC JU systems. The EU supercomputers are available through “access calls,” which are invitations to apply for time-to-use supercomputing resources. Eligible parties include researchers from academia and other institutions, public authorities, and companies established or located in an EU Member State or a third country associated with the Digital Europe Programme or Horizon Europe.
欧盟委员会(EC)于2024年6月26日宣布了大型人工智能大挑战赛的获胜者,该大赛于2023年启动,是欧盟(EU)AI-BOOST项目的一部分。人工智能创新创业公司和中小企业(SME)受邀于2023年11月16日至2024年2月13日提交提案。 参与的公司正在竞争使用欧盟高性能计算(HPC)站点之一开发基础模型,该站点是EuroHPC联合项目(EuroHPC JU)的一部分。EuroHPC JU是欧盟和参与成员国之间的合作努力,旨在促进欧盟在人工智能和超级计算方面的进步(作为GenAI4EU计划的一部分)。 获奖者是总部位于法国的Lingua Custodia,该公司专门从事金融领域的人工智能和自然语言处理(NLP);Unbabel,一家总部位于葡萄牙的语言技术和服务公司;拉脱维亚的Tilde,一家语言技术公司,MT和波罗的海斯拉夫语言公司;Textgain是一家总部位于比利时的人工智能初创公司,专注于预测文本和语音分析。 四家获胜者与其他90家公司竞争,赢得25万欧元的奖金,并在两个欧盟HPC站点之一获得200万小时的计算时间。这些语言AI技术公司也将有机会成为与欧盟委员会DG Connect-Cnect AI and Robotics和EuroHPC-JU合作的网络的一部分。 对基础模型的主要要求是,它必须使用最先进的方法从头开始训练,并且至少有300亿个参数。在他们的提案中,这些公司必须为模型的相关性和HPC的使用提供理由。他们还必须展示使用HPC系统训练基础模型的内部专业知识,以及使用所述EuroHPC JU系统的计划。 欧盟超级计算机可以通过“访问电话”获得,这是申请使用时间超级计算资源的邀请。合格方包括来自学术界和其他机构、公共当局以及在欧盟成员国或与数字欧洲计划或地平线欧洲相关的第三国建立或位于的公司的研究人员。

以上中文文本为机器翻译,存在不同程度偏差和错误,请理解并参考英文原文阅读。

阅读原文