Li Jun
Principal Firmware Engineer at MaxLinear
Singapore
Summary
🤩
Rockstar🎓
Top SchoolLi Jun is a Principal Firmware Engineer based in Singapore with over 10 years of experience building high-performance firmware and systems for networking and ML infrastructure. Currently at MaxLinear after a senior engineering tenure at Intel focused on Ethernet firmware, he combines deep embedded-software expertise with practical knowledge of hardware-accelerated compute. His open-source contributions to PaddlePaddle include CUDA kernel work, TF32 optimizations, and operator improvements—showing an unusual blend of low-level firmware sensibilities and ML framework engineering. Trained at Tsinghua and NUS, he brings rigorous academic grounding in mechatronics and electrical engineering to product-grade software design. Colleagues know him for pragmatic optimizations (e.g., LogSoftmax and interpolation kernels) and for bridging static-graph compatibility with modern CUDA features. He often operates at the intersection of device firmware and ML performance, turning hardware constraints into measurable software gains.
10 years of coding experience
26 years of employment as a software developer
Bachelor of Engineering (B.Eng.), Mechatronics, Robotics, and Automation Engineering, Bachelor of Engineering (B.Eng.), Mechatronics, Robotics, and Automation Engineering at Tsinghua UniversityMaster of Engineering (M.Eng.), Electrical, Electronics and Communications Engineering, Master of Engineering (M.Eng.), Electrical, Electronics and Communications Engineering at National University of Singapore
Github Skills (16)
neural-network10
cuda10
paddlepaddle10
machine-learning10
benchmark10
benchmarking10
deep-learning10
python10
efficientnet9
testing9
apidoc9
cprogramming-language9
distributed-training9
c-language9
api9
Programming languages (3)
Github contributions (5)
PaddlePaddle/benchmark
Oct 2020 - Dec 2021
Role in this project:
ML Engineer
ML EngineerContributions:44 reviews, 11 commits, 39 PRs in 1 year 2 months
Contributions summary:Li primarily contributes to benchmarking PaddlePaddle's operators (ops) by creating and modifying Python scripts. They added numerous Python scripts for testing various 2.0 API ops, focusing on functionalities like padding, switch cases, sequence masking, and one-hot encoding. Additionally, the user modified existing code by modifying test parameters and updating tests to reflect the current API changes.
PaddlePaddle/Paddle
Sep 2020 - May 2022
PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）
Role in this project:Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:242 reviews, 36 commits, 80 PRs in 1 year 8 months
Contributions summary:Li primarily contributed to the PaddlePaddle deep learning framework by implementing shape checks and removing type error documentation for the `fill_constant` operator. They also added CUDA support for the `unique` operator, including the CUDA kernels and support for int32/int64 data types, and addressed static graph compatibility. Further contributions include adding TF32 support for A100 tensor core acceleration within cuBLAS, and adding a TF32 switch for cuDNN to enable/disable the feature. Additionally, they optimized LogSoftmax forward and backward operations, as well as bilinear and nearest interpolation within the framework.
pytorchpythonparalleldeep-learningpaddlepaddle
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.