Sheng Fu
Senior Staff Software Engineer at NVIDIA
San Jose, California, United States
Summary
🤩
Rockstar🎓
Top SchoolSheng Fu is a Senior Staff Software Engineer based in San Jose with deep expertise in parallel architectures, GPU/AI accelerator programming, and performance engineering for large-scale ML systems. He has a strong track record across industry leaders—Intel, Graphcore, SambaNova, Meta, and now NVIDIA—bringing up and optimizing large language models on clusters and developing ML graph compilers with MLIR. Over seven years he has optimized open-source AI models with PyTorch, TensorFlow, and XLA, and contributed critical execution-trace fixes to PyTorch that improved profiling fidelity for tensor strides and Triton kernels. Known for practical performance modeling, workload profiling, and aggressive quantization for ultra-low latency, he combines research-level rigor from a PhD background with hands-on production impact. An uncommon strength is his ability to resolve subtle multithreading deadlocks in core frameworks, turning elusive bugs into measurable performance gains.
2 years of coding experience
12 years of employment as a software developer
Doctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at Zhejiang University
Github Skills (19)
debugging10
pytorch10
debug10
performance-analytics10
performance-monitor10
c-language10
code-profiling10
python10
performance-analysis10
profiling10
performance-monitoring10
cprogramming-language10
autograd8
tensor8
neural-network7
Programming languages (2)
Github contributions (5)
pytorch/pytorch
Feb 2024 - Mar 2025
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
Back-end Developer / Performance Engineer
Back-end Developer / Performance EngineerContributions:37 reviews, 49 PRs, 1 push in 1 year 1 month
Contributions summary:Sheng primarily contributed to optimizing and debugging the PyTorch execution trace functionality, a core component for profiling and performance analysis within the PyTorch framework. Their work focused on resolving deadlocks related to multithreaded access and improving the capture of detailed information, including tensor strides, kernel information (Triton kernels), and support for integral tensor data. These changes directly impacted the accuracy and completeness of the execution trace, ultimately aiding in performance analysis and debugging of PyTorch models.
pythongpu-accelerationdeep-learninggpunumpy
shengfukevin/pytorch
Feb 2024 - Mar 2025
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Contributions:254 pushes, 41 branches in 1 year 1 month
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.