On December 28, the WAVE SUMMIT+Deep Learning Developers Conference 2023 hosted by the National Engineering Research Center for Deep Learning Technology and Applications was held in Beijing. Wang Haifeng, chief technology officer of Baidu and director of the National Engineering Research Center for Deep Learning Technology and Application, said that by the end of December 2023, the number of PaddlePaddle platform developers had reached 10.7 million, serving 235000 enterprises and institutions, and 860000 models had been created based on PaddlePaddle.
As a large-scale AI developer community, PaddlePaddle Galaxy Community has been fully upgraded to provide developers with rich heterogeneous computing power and common components, PaddlePaddle industrial model library and full process development tool chain, so as to realize low-cost development of AI applications.
It is reported that the Xinghe Model Community currently has all the elements of AI native application innovation, providing developers with integrated services such as development, experience, promotion, communication, and learning. The new release of the Big Model Tool Center gathers the achievements of Baidu AI over the years, including PaddlePaddle's industrial model library, Baidu Brain's AI capabilities, and supports the access of ecological tools, enabling developers to efficiently develop AI native applications.
On the scene of the conference, Wu Tian, vice president of Baidu Group and deputy director of the National Engineering Research Center for Deep Learning Technology and Application, used the development mode of "multi tool intelligent layout" of the big model tool center to demonstrate the zero code development of a "travel assistant" application that integrates multi-mode full functions such as image and text recognition, question and answer, translation, broadcasting, etc.
Ma Yanjun, general manager of Baidu AI technology ecology, released the development tool chain of PaddlePaddle open source framework version 2.6 and large model reconstruction. Among them, the open source framework version 2.6 of PaddlePaddle has achieved high scalability IR, supporting the extreme performance optimization of large models; Support unified automatic parallel programming between dynamic and static, greatly simplifying the development of mixed parallel training code.
The newly released Baidu Comate AutoWork at this meeting has achieved a new stage of evolution from single generated code to systematic processing of complex R&D tasks. Developers only need to clarify their goals and intentions, and AutoWork can deeply interpret the code repository, develop execution plans, and intelligently generate code, achieving end-to-end automation from requirement analysis to code implementation. Log in to the official website of Baidu Comate, and enterprise level users can apply for a trial.
It is understood that currently, 20% of Baidu's new code is generated by Comate Intelligence every day. Comate SaaS services have 8000 enterprise customers, and the overall code generation adoption rate exceeds 40%.
The rapid development of big model technology has put forward higher requirements for AI talents. Baidu and the Talent Exchange Center of the Ministry of Industry and Information Technology jointly launched the "Standards and Capability Evaluation of Talents in the Big Model Industry", providing clear directions for the cultivation of big model talents and scientific references for the selection and employment of enterprises.
On the PaddlePaddle platform, which gathers more than ten million developers, innovative achievements continue to emerge. For example, the Soundbridge AI Language Training Team, dedicated to providing services for hearing-impaired individuals, brings AI oral rehabilitation applications and provides written feedback guidance at any time. In the future, it will also provide multimodal feedback. (End)