Wei Xia
Technical Expert in Code Large Models at Tencent
Master’s degree from the University of Chinese Academy of Sciences, currently responsible for the product development and research of the Codebuddy code large model. Previously worked at Huawei and Tencent on technical research and practical applications related to large language models, covering their use in recommendation, education, and code, enhancing product competitiveness and bringing users more convenient and efficient AI experiences. Has published several academic papers in international conferences such as EMNLP, AAAI, KDD, WWW, and WSDM.
Topic
AI Coding Human-Machine Collaboration: Model Exploration and Engineering Practice with CodeBuddy
This talk will focus on the AI Coding human-machine collaboration paradigm. It will begin with a review of the development background of the intelligent programming industry, analyzing the evolution of technologies and application trends. Next, the presentation will introduce the product form and core capabilities of CodeBuddy. The talk will then delve into practical experience in training code large models, including data synthesis methods and training strategies for tasks such as code completion, NES, and Craft Agent, covering the full process from pretraining and continual pretraining to post-training stages (SFT, RLHF). Additionally, we will share design principles and practical experiences in building multi-dimensional evaluation systems, along with results validated through online experiments. Finally, the session will conclude with a summary and outlook on the direction of technological evolution and future plans.
