Xiaodong Wang
Research Scientist at Facebook
Menlo Park, California, United States
Summary
🤩
Rockstar🎓
Top SchoolXiaodong Wang is a research scientist at Facebook's AI Infrastructure Foundation with eight years of experience optimizing hardware performance for cutting-edge machine learning workloads. He blends deep academic training (Ph.D. in Computer Engineering from Cornell) with hands-on engineering—previously defining next-generation datacenter compute as a performance engineer and contributing system-level cache optimizations at Cavium. At Facebook he works on open-source software frameworks and production hardware available through the Open Compute Project, with notable contributions to the PyTorch backend improving CUDA compatibility, performance, and stability for distributed training. Based in Menlo Park, he focuses on squeezing hardware efficiency from both software and architecture angles, and he has a track record of fixing subtle race conditions and deprecated API issues that improve long-term maintainability.
7 years of coding experience
6 years of employment as a software developer
ECE, ECE at Georgia Institute of TechnologyDoctor of Philosophy (Ph.D.), Computer Engineering, Doctor of Philosophy (Ph.D.), Computer Engineering at Cornell UniversityB.S., Electronic Science and Technology, B.S., Electronic Science and Technology at Shanghai Jiao Tong University
Chinese, English
Github Skills (15)
cuda10
pytorch10
machine-learning10
c-language10
tensor10
deep-learning10
cprogramming-language10
gpu10
performance-optimization10
python10
autograd9
distributed-computing9
neural-network9
ccl9
roc8
Programming languages (7)
Github contributions (5)
pytorch/pytorch
Jun 2018 - Jan 2023
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
Back-end Developer
Back-end DeveloperContributions:145 reviews, 5 commits, 123 PRs in 4 years 8 months
Contributions summary:Xiaodong primarily focused on improving the PyTorch backend, particularly for CUDA-related functionalities. Their work included fixing namespace ambiguities and cleaning up deprecated CUDA APIs, indicating a focus on code correctness and compatibility with newer CUDA versions. They also addressed performance bottlenecks and race conditions related to caching and logging within the PyTorch Inductor and distributed training components. Furthermore, they contributed to the performance and stability of PyTorch's core tensor operations.
pythongpu-accelerationdeep-learninggpunumpy
xw285cornell/pytorch
Jun 2018 - Mar 2025
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Contributions:221 pushes, 115 branches in 6 years 10 months
pythongpu-accelerationdeep-learninggpuacceleration
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.