陈序
创始人 at 零碳元宇宙与未来资产智库MetaZ
Hong Kong, China
Summary
👤
Senior🎓
Top School陈序 is a Hong Kong–based serial entrepreneur, media executive and Web3 angel investor who has spent the past two decades at the intersection of technology, journalism and product innovation. He founded MetaZ, a zero-carbon metaverse and future-assets think tank, and previously launched China’s first crowdfunding publishing platform and a content-blockchain lab that explored blockchain for media. A former executive editor for Newsweek China and senior adviser to MIT Technology Review China, he has led digital media transformations that include early experiments with AI-driven news recommendation. On GitHub he has contributed MLOps fixes to vllm, improving LLM inference stability and sampling behavior—an example of his hands-on interest in applied AI infrastructure. Holder of an EMBA from Shanghai Jiao Tong and a Silicon Valley immersion credential, he combines editorial rigor with product and technical fluency to build new media and Web3 ecosystems. Notably, he was the first mainland Chinese journalist to serve as executive editor for Newsweek’s international Chinese edition.
9 years of coding experience
7 years of employment as a software developer
中国经济研究中心第八届《财经》奖学金项目, 中国经济研究中心第八届《财经》奖学金项目 at 北京大学Silicon Valley Immersion Program, Silicon Valley Immersion Program at University of San FranciscoE.M.B.A, Business Administration and Management, General, E.M.B.A, Business Administration and Management, General at Shanghai Jiao Tong University
Chinese
Github Skills (7)
pytorch10
inference10
python10
scheduler10
llm10
cuda9
transformer9
Programming languages (7)
Github contributions (5)
vllm-project/vllm
Sep 2023 - Aug 2024
A high-throughput and memory-efficient inference and serving engine for LLMs
Role in this project:
MLOps Engineer
MLOps EngineerContributions:15 reviews, 7 PRs, 46 comments in 11 months
Contributions summary:陈序 primarily contributed to the VLLM project by identifying and resolving issues related to the inference and serving engine for LLMs. Their work included fixing bugs that caused the engine to hang due to long prompts and issues within the scheduler, demonstrating a focus on the stability and robustness of the system. They also enhanced the engine's functionality by adding stop token IDs in the sampling parameters and aligning `top_p` and `top_k` sampling with the Hugging Face implementation. Furthermore, they made a minor fix related to CUDA version.
amdcudadeepseekgpthpu
Mensu/Matrix-UWP
May 2017 - Jul 2018
Matrix client written in C#
Contributions:82 commits, 24 PRs, 51 pushes in 1 year 2 months
matrixmatrix-client
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.