Page 293 - AI for Good Innovate for Impact
P. 293

AI for Good Innovate for Impact



               4      Sequence Diagram                                                                              4.3 - 5G































               Figure  2  Sequence  Diagram  of  AI  Middle  Platform  Pipeline  consisting  of  the  following
               components:

               •    Data phase: build high-quality datasets, structured data, unstructured data, knowledge
                    graph.
               •    Computing power: integrate smart arithmetic, generalized arithmetic, storage, etc. for
                    training and inference.
               •    Standard training and inference stage: integrate pytorch, LangChian, TensorRT and other
                    frameworks to provide a low-code code development environment for model training
                    and inference.
               •    Model service stage: the model produced by the model, as a service provided to the
                    outside world, including large language models, multimodal models, speech large
                    models, small models,  and other components of the MaaS service.
               •    Model application construction stage: Relying on the big model Agent (intelligent body),
                    video toolchain, voice toolchain, and other application toolchains, the model's reasoning
                    ability is used to plan and disassemble the task, and external plug-in tools are used to
                    complete complex tasks[5].
               •    Industry application stage: Industry AI big model applications, such as RAG class
                    applications, ChatBI class applications, and Agent class applications, including
                    transportation big model application assistants, tourism guide application assistants,
                    and grassroots governance assistants.


               5      References

               [1]  “Nanjing University of Posts and Telecommunications,” [Online]. Available: http:// www
                    .njupt .edu .cn/  .
               [2]  Che Shaoshuai, Zhang Wei, Design of Visual Perception AI Platform Based on Cloud Edge
                    Collaboration [J]. Nanjing: Jiangsu Journal of Communications, April 2024.
               [3]  Che Shaoshuai,  Research on Multi object Detection and Tracking Technology Based on
                    O2O Learning [D]. Nanjing: Nanjing University of Posts and Telecommunications Library,
                    May 2016.
               [4]  Ministry of Industry and Information Technology (MIIT), Report on the Development
                    of Artificial Intelligence (2024), China Academy of Information and Communications



                                                                                                    257
   288   289   290   291   292   293   294   295   296   297   298