Shiyan Deng
Software Engineer at Meta
United States
Summary
🤩
Rockstar🎓
Top SchoolShiyan Deng is a software engineer with nine years of experience specializing in ML infrastructure, inference enablement, and high-performance back-end systems, currently building real-time serving and model-update frameworks at Meta that handle billions of requests. He has deep hands-on expertise in PyTorch internals and compiler toolchains—contributing to torchrec, torch.fx, FBGEMM and the Glow compiler—focused on optimizer/inference pipelines, GPU/CPU embedding performance, and FP16 support. His work blends low-level performance tuning (CUDA stream and allocator fixes, efficient data transfers, multi-card correctness) with production-grade systems for recommendation models. A Georgetown CS master’s alumnus, he uniquely bridges research-grade ML tooling and large-scale production deployment, often surfacing subtle issues like pickling and metadata preservation that impact robustness in deployed models.
9 years of coding experience
1 year of employment as a software developer
Master's degree, Computer Science, Master's degree, Computer Science at Georgetown UniversityBachelor of Engineering - BE, Computer Science, Bachelor of Engineering - BE, Computer Science at Xiamen University
Github Skills (27)
pytorch10
c-language10
recommendation-system10
fba10
float3210
emm10
python10
gpu-programming10
machine-learning10
javafx10
deeplearning-ai10
deep-learning10
f10
gpu10
performance-optimization10
Programming languages (2)
Github contributions (5)
pytorch/glow
Sep 2020 - Jun 2021
Compiler for Neural Network hardware accelerators
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:35 reviews, 57 commits, 62 PRs in 9 months
Contributions summary:Shiyan's contributions primarily involved modifying and improving the Glow compiler for neural network hardware accelerators, specifically focusing on optimizations for average pooling operations and the implementation of FP16 support. They added features, such as a flag to exclude padding in average pooling and support for rotated bounding boxes. Further work included fixing casting issues related to float16, and enhancements to the BBoxTransform node, along with tests that demonstrate the user's understanding of Glow's internals and its use in machine-learning-related operations.
hardware-acceleratorscompilerneural-networkacceleratorshardware
pytorch/torchrec
Feb 2022 - Jan 2023
Pytorch domain library for recommendation systems
Role in this project:Back-end Developer / ML Engineer
Back-end Developer / ML EngineerContributions:1 review, 14 commits, 30 PRs in 11 months
Contributions summary:Shiyan contributed to the PyTorch domain library for recommendation systems. Their commits focused on improving the inference pipeline and optimizer functionalities. The user implemented changes to GPUExecutor, adding completion workers for asynchronous result processing, and also added validations for weights tensors in sparse features. Furthermore, they addressed issues related to pickling in the torchrec optimizer.
cudapytorchrecommendation-systemsdeep-learninggpu
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.