Michael He
Research Scientist at Meta
Menlo Park, California, United States
Summary
🤩
Rockstar🎓
Top SchoolMichael He is a research scientist at Meta with eight years of experience building ML/AI systems, distributed training infrastructure, and compiler-aware performance optimizations. He combines a strong hardware and EDA background from Tsinghua with systems research at UT Austin to tackle model/infra co-design problems like MRS training. His open-source contributions to PyTorch—improving distributed training support, PT2 compatibility, and enabling Dynamo in core functions—showcase practical impact on widely used ML tooling. Past internships at Facebook and VMware reflect a consistent focus on performance diagnosis across virtualized and multi-process GPU environments. Colleagues know him for bridging low-level performance engineering with high-level model compilation to squeeze real-world speedups.
8 years of coding experience
The University of Texas at AustinBachelor of Engineering - BE, Electronic Engineering, Bachelor of Engineering - BE, Electronic Engineering at Tsinghua University
Github Skills (10)
pytorch10
machine-learning10
tensor10
distributed-training10
deep-learning10
gpu10
python10
autograd9
neural-network8
numpy8
Programming languages (3)
Github contributions (5)
pytorch/pytorch
Aug 2024 - Feb 2025
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
ML Engineer
ML EngineerContributions:7 PRs, 1 comment in 6 months
Contributions summary:Michael primarily contributed to improving the PyTorch framework, focusing on areas related to distributed training and PT2 (PyTorch 2.0) compatibility. Their work involved adding support for `FakeProcessGroup` to mimic multi-process tests within TorchRec, resolving compatibility issues with `torch.diff`, and implementing optimizations within the PT2 compiler to address graph breaks. They also enabled the use of Dynamo in a core function. This indicates a strong focus on model compilation, distributed training and performance.
pythongpu-accelerationdeep-learninggpunumpy
TroyGarden/fcm_docker_image
Oct 2017 - Aug 2021
Contributions:2 PRs, 111 pushes, 7 branches in 3 years 10 months
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.