Sherlock Huang
Software Engineer at Meta
San Francisco Bay Area United States
Summary
🤩
Rockstar🎓
Top SchoolSherlock Huang is a software engineer with 11 years of experience building high-performance back-end systems and ML frameworks, currently contributing to PyTorch 2.0 at Meta from the San Francisco Bay Area. He has deep expertise in ML runtimes and training infrastructure, having led ONNX Runtime Training core efforts at Microsoft to accelerate large-scale model training. Sherlock’s open-source work spans influential projects like ONNX, PyTorch, functorch and ONNX Runtime, where he implemented operators, preserved autograd stack traces for better debugging, and optimized kernel and training operations. He pairs low-level C++ and systems know-how (RocksDB and table option tuning) with ML-focused engineering, enabling both inference and training performance improvements. Known for translating research-grade ideas into production-grade runtime features, he also has a track record of hackathon-winning prototypes in computer vision and accessibility. His background includes top grades in ECE (HKUST) and a consistent focus on making complex ML systems debuggable, efficient, and production-ready.
11 years of coding experience
8 years of employment as a software developer
Hong Kong University of Science and Technology (HKUST)
English, Chinese
Github Skills (31)
pytorch10
storage-engine10
c-language10
database-connection10
python10
operation10
tensorrt10
rocksdb10
graph10
databases10
machine-learning10
server-configuration10
onnx10
gradient10
deep-learning10
Programming languages (5)
Github contributions (5)
microsoft/onnxruntime
Mar 2019 - Nov 2021
ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:1082 reviews, 113 commits, 215 PRs in 2 years 8 months
Contributions summary:Sherlock contributed to the core functionalities of the ONNX Runtime, focusing on implementing and optimizing machine-learning related utilities. Their work includes introducing new mathematical operations and related kernels, specifically for row-wise and column-wise sum operations, which likely contribute to improved performance of ML models. Furthermore, the user was responsible for implementing a new gradient operation (WhereGrad), and they made various changes to training related functionalities which are crucial for the development of the training pipeline. They also made various changes to update and register training operations within the ONNX framework.
runtimetrainingtensorflowai-frameworkaccelerator
pytorch/pytorch
Mar 2022 - Jan 2023
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:719 reviews, 759 commits, 252 PRs in 10 months
Contributions summary:Sherlock primarily contributed to the development of the PyTorch framework. Their commits focused on implementing, refactoring, and testing features related to graph partitioning, fusion, and the integration of custom operators with backends like NVFuser. They were involved in enhancements for operator decompositions, particularly for enabling support for various operators and data types within the PyTorch ecosystem. The user's work included improvements to the logical schema for tensor serialization and improved the preservation of metadata during transformations.
pythongpu-accelerationdeep-learninggpunumpy
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.