翻译技术速递Large language models: pre-training with a twist
大型语言模型: 带有扭曲的预训练
Large language models: pre-training with a twist
The next DG CNECT technical workshop will take place on 14 June 2022 (10-16 CEST) via Zoom
In future-oriented AI, large language models (LLMs) are on the pulse of time. Combining the latest deep learning technology with heavy computing infrastructure now enables academic or commercial organisations to build language models from large amounts of text or speech data. Such models incorporate information that is useful for understanding a language, such as its vocabulary and how it expresses meaning.
Starting from such pre-trained LLMs, other organisations can train additional models for their specific applications (e.g., automated translation, summarisation, dialogue interaction, speech recognition) or domains, using much less data and computing power. That is why the process of “specialising” a pre-trained model is gaining a large impact on the field of natural language processing.
Organised by the Multilingualism sector of DG CNECT, the online workshop Large Language Models: pre-training with a twist will take place on 14 June from 10:00 to 16:00 CEST via Zoom. In this workshop, which is primarily oriented towards staff of EU institutions and European public administrations, top-level speakers will discuss various aspects of pre-trained LLMs, such as:
- how to make use of models available from repositories.
- how to specialise multilingual models, for instance for automated translation.
- how to leverage models in specific use cases, within public administrations and industry.
- how to consider legal aspects of models.
The event is part of a series of technical workshops in the ELRC project, which supports the development of eTranslation, the online machine translation service provided by the European Commission, and a wider deployment of the DG's services in terms of language resources and tools. The detailed workshop agenda can be found on the event webpage.
Interested? Then register here to participate!
大型语言模型: 带有扭曲的预训练
下一次DG CNECT技术研讨会将在2022年6月14日 (10-16 CEST) 通过Zoom举行
在面向未来的人工智能中，大型语言模型 (llm) 正处于时间的脉搏。将最新的深度学习技术与繁重的计算基础设施相结合，现在使学术或商业组织能够从大量文本或语音数据中构建语言模型。此类模型包含了对理解语言有用的信息，例如其词汇及其表达含义的方式。
从这样的预先训练的llm开始，其他组织可以使用更少的数据和计算能力为其特定应用 (例如，自动翻译、总结、对话交互、语音识别) 或领域训练额外的模型。这就是为什么 “专业化” 预先训练的模型的过程正在对自然语言处理领域产生巨大影响的原因。
由DG CNECT的多语言部门组织的在线研讨会 “大型语言模型: 扭曲的预培训” 将在通过Zoom从10:00到16:00 CEST的6月14日上进行。在本次研讨会中，主要面向欧盟机构和欧洲公共行政部门的工作人员，高层发言人将讨论经过培训的llm的各个方面，例如:
-如何利用存储库中可用的模型。
-如何专门化多语言模型，例如自动翻译。
-如何在公共行政部门和行业内的特定用例中利用模型。
-如何考虑模型的法律方面。
该活动是ELRC项目中一系列技术研讨会的一部分，该研讨会支持电子翻译的开发，欧盟委员会提供的在线机器翻译服务以及DG在语言资源和工具方面的更广泛部署。详细的研讨会议程可在活动网页上找到。
感兴趣吗？那就在这里注册参加吧!
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