Guang Liu
BAAI Basic Language Modelling and Software Group Leader
D. in Intelligent Science and Technology from Beijing University of Posts and Telecommunications, head of NLP and Multimodal Research Center of Beijing Zhiyuan Artificial Intelligence Research Institute, responsible for FlagAI open source project of FeiZhi and Wudao-Skyhawk Aquila series of basic models (7B/34B), presided over the Science and Technology Innovation 2030 - "New Generation of Artificial Intelligence" major project, published papers in academic conferences and journals of the field of Artificial Intelligence and more than a dozen of invention patents applied for, with his main research interests in the pre-training of large models and so on.
Topic
From Theory to Practice: Analysing the Development Process and Future Prospects of the Aquila Model
In this presentation, I will explore in depth the development process of the Aquila Large Scale Language Model (LLM). I will analyse the background of the development of the Aquila model, the problems it faced, and how we dealt with them, the concrete results of the practice, and its future development direction, from theory to practice. This process covers everything from acquiring and processing the corpus, optimising the model training process, to improving the model's effectiveness and accuracy. This is a typical real-world example showing how to apply and optimise large-scale language models in real projects. My experience will provide in-depth insight and reference value for people who want to learn more about large-scale language model development, rather than just staying at the theoretical level of simple concepts. I will also provide insights into the future of the Aquila model, including how we plan to improve its performance, accuracy, and broaden its application domains, as well as how we will continue to optimise the user experience in future developments.