Chunwei Yan
Deep Learning Architect at NVIDIA
Haidian District, Beijing, China
Summary
🤩
Rockstar🎓
Top School
Top expert inArtificial Intelligence and Computer Vision Technologies
Chunwei Yan is a Deep Learning Architect with 15 years of experience building high-performance neural network systems, currently focused on performance architecture at NVIDIA. He led large teams at Baidu to design and ship core components of the PaddlePaddle ecosystem—including Paddle Inference, Paddle-Lite, INFRT, CINN and VisualDL—and remains an active contributor to those open-source projects. His hands-on work spans compiler backends and runtime systems, notably optimizing GPU MMA implementations in the Triton compiler and refactoring build and inference pipelines for mobile and edge. Comfortable bridging research and production, he combines a mathematics background and a CS master’s with deep expertise in MLIR-based compilers and deployment tooling. Colleagues rely on him for pragmatic architecture decisions that squeeze maximum performance from both hardware and software.
15 years of coding experience
7 years of employment as a software developer
Master of science, Computer Science, Master of science, Computer Science at 北京大学Bachelor's Degree, Mathematics and Applied Mathematics, Bachelor's Degree, Mathematics and Applied Mathematics at China Agricultural University
Github Skills (31)
markdown10
c-language10
back-end-development10
python10
llvm10
gpu-programming10
machine-learning10
inference10
cmake10
triton10
deep-learning10
markdown-it10
build-automation10
compiler-optimization10
paddlepaddle10
Programming languages (8)
Github contributions (5)
PaddlePaddle/Paddle-Lite
Aug 2019 - Sep 2021
PaddlePaddle High Performance Deep Learning Inference Engine for Mobile and Edge (飞桨高性能深度学习端侧推理引擎）
Role in this project:
Backend Developer & Automation Engineer
Backend Developer & Automation EngineerContributions:31 releases, 220 reviews, 46 commits in 2 years 1 month
Contributions summary:Chunwei primarily contributed to refactoring and enhancing the build system and compilation processes within the Paddle-Lite project. This involved moving CMake functions, refactoring the op kernel compilation system, and integrating the model optimize tool. They also modified build scripts to include options for extra builds and optimized the number of build threads. Further commits addressed third-party dependency management and enhancements to the download process.
inference-enginemobilebaidutensorflowfpga
PaddlePaddle/VisualDL
Nov 2017 - Jan 2018
Deep Learning Visualization Toolkit（『飞桨』深度学习可视化工具 ）
Role in this project:Back-end Developer
Back-end DeveloperContributions:1 release, 220 commits, 108 PRs in 2 months
Contributions summary:Chunwei contributed to the back-end development of the VisualDL toolkit by implementing and modifying the storage functionality. Their work involved updating protobuf definitions, creating and implementing modules for storage, including the core logic for adding and retrieving data, and adding methods for saving and loading data from files. The user's commits also added the initial implementation of the information maintainer.
pytorchdeep-learningpaddlepaddlecomputer-visiononnx
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.