Steven Li
Staff Software Engineer, ML at Meta
San Francisco Bay Area United States
Summary
🤩
Rockstar🎓
Top SchoolSteven Li is a Staff Software Engineer and ML leader based in the San Francisco Bay Area with a PhD and two decades of engineering experience spanning wireless systems, DSP, and large-scale ML systems. He currently leads LLM and multimodal model research, training, inference and optimization at Meta after architecting AI platform and model acceleration work at Microsoft. His hands-on expertise bridges low-level kernel and build engineering—evidenced by contributions to ONNX Runtime (TensorRT integration, dynamic-shape memory handling)—and high-level model tooling such as float16 conversion work in ONNXMLTools. Comfortable across cloud, GPU/ASIC, FPGA and embedded SoC stacks, he has repeatedly delivered production-ready systems and resolved critical performance bottlenecks. Known for translating academic depth into pragmatic, deployable solutions, he combines research rigor with a pragmatist’s focus on robust tooling and reproducible model deployment.
7 years of coding experience
19 years of employment as a software developer
PhD, PhD at Shanghai Jiao Tong University
Github Skills (20)
c-language10
python10
tensorrt10
machine-learning10
build-system10
cmake10
onnx10
numpy10
tensorflow10
cuda10
model-conversion10
tensor10
cprogramming-language10
model-optimization9
keras9
Programming languages (4)
Github contributions (5)
microsoft/onnxruntime
Jan 2019 - Jan 2023
ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator
Role in this project:
Back-end Developer & DevOps Engineer
Back-end Developer & DevOps EngineerContributions:2 releases, 125 reviews, 295 commits in 4 years
Contributions summary:Steven primarily focused on enhancing the build process and adding support for TensorRT within the ONNX Runtime project. This included modifying CMake files, build scripts, and test configurations to integrate TensorRT execution provider. They also contributed to the addition and refinement of unit tests related to the TensorRT integration and made improvements to accommodate dynamic shapes and memory management for TensorRT subgraphs.
runtimetrainingtensorflowai-frameworkaccelerator
onnx/onnxmltools
Jun 2018 - Apr 2019
ONNXMLTools enables conversion of models to ONNX
Role in this project:ML Engineer
ML EngineerContributions:8 commits, 15 PRs, 10 pushes in 10 months
Contributions summary:Steven's primary contribution was the development and modification of a float16 converter tool within the ONNXMLTools repository. They implemented and refined the conversion process for various data types, including tensors and model inputs/outputs, addressing compatibility issues. Furthermore, they incorporated checks for input validation and addressed build failures, indicating a focus on improving the robustness and functionality of the float16 conversion features. The user's work is central to the project goal of converting models to ONNX, particularly enabling the use of float16 data types.
python-librarypytorchdeep-learningmachine-learningonnx
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.