摘要
AI大模型作为当前最热门的技术领域,得到了广泛的关注。以谷歌的BERT和OpenAI的GPT为代表,模型的参数规模逐步提升至千亿甚至万亿级,训练数据量级也大幅提升,随之带来了模型能力的显著提高,掀起了国内外AI大模型的研究热潮。论述了AI大模型的发展概况,剖析了具有里程碑意义的各类算法模型,介绍了其依赖的核心技术和原理,评估了其存在的挑战和局限,最后讨论了未来AI大模型的发展趋势。
As the hottest technology field in the current,AI large models received extensive attention.Taking Bert of Google and GPT of OpenAI as representatives,the parameter size of the model gradually increased to hundreds of billions or even trillions,and the training data scale also increased greatly,which brings about a significant improvement in model capability and triggers a research upsurge of AI large models at home and abroad.This paper gives an overview of the development of AI large models,analyzes various milestone algorithm models,introduces the core technologies and principles they rely on,evaluates their challenges and limitations,and finally discusses the development trends of AI large models in the future.
作者
张乾君
ZHANG Qianjun(No.10 Institute of CETC,Chengdu Sichuan 610036,China)
出处
《通信技术》
2023年第3期255-262,共8页
Communications Technology