Jun Zhang
Jun Zhang is currently a Senior Engineer at Baidu and one of the maintainers of the PaddlePaddle open-source community. He has been a long-term contributor to open source and previously served as a member of the Technical Supervisory Committee of the OpenAtom Open Source Foundation. Zhang Jun holds over 50 invention patents in the fields of natural language processing, cloud computing, and deep learning.
Topic
Wenxin 4.5: Open-Source Large Model and Key Training Technologies
With the gradual open-sourcing of large models such as the Wenxin 4.5 series, research and applications focusing on their inference performance and deployment efficiency are rapidly growing. Based on this, ERNIEKit and FastDeploy, built on the PaddlePaddle framework, provide capabilities for fine-tuning and efficient deployment of large models. This talk will present an overview of PaddlePaddle’s open-source large models and discuss the technical challenges and solutions encountered during training and inference. Outline: * Introduction to Wenxin open-source large models * In-depth analysis of key training technologies in Wenxin 4.5 * Overview of the FastDeploy large model inference framework
