Zeyao Du is an NLP engineer with nine years’ experience building production-grade language models and search systems, currently working at Shopee in Shanghai. He has driven LLM and NLP efforts at companies including Weibo and 阅文集团, and holds a mathematics and statistics degree from Zhejiang University. An active open-source contributor, Zeyao authored a popular 7k-star Chinese GPT-2 repo where he modernized training and generation pipelines with the transformers library and adapted tokenization for Chinese. He combines practical engineering—bug fixes, training script overhaul, and tokenization work—with research-minded model development, making him effective at taking models from experiments into production. Notably, his background spans both large social and content platforms and e-commerce, giving him a strong sense for applied NLP product constraints.
9 years of coding experience
5 years of employment as a software developer
Nanjing Foreign Language School
Bachelor's degree, MATHEMATICS AND STATISTICS, Bachelor's degree, MATHEMATICS AND STATISTICS at Zhejiang University
Chinese version of GPT2 training code, using BERT tokenizer.
Role in this project:
ML Engineer
Contributions:45 commits, 21 PRs, 305 pushes in 1 year 9 months
Contributions summary:Zeyao primarily contributed to the training and generation aspects of the GPT-2 Chinese model. Their work involved updating existing code to leverage the `transformers` library, modifying training scripts, and fixing bugs related to the generation process. The user also made changes to tokenization scripts and configuration files, suggesting a hands-on role in adapting the model and its components for the Chinese language context.
GPT2 training script for Chinese in Tensorflow 2.0
Contributions:22 commits, 21 pushes, 1 branch in 9 months
nlpchinesetensorflow-2-0trainingtensorflow
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.