Jianyu Huang
Research Scientist at Meta
San Francisco Bay Area United States
Summary
🤩
Rockstar🎓
Top SchoolJianyu Huang is a research scientist in the San Francisco Bay Area with 26 years of experience building high-performance ML systems, GPU kernels, and scalable LLM training and inference pipelines. His work spans academia and industry—from PhD research on practical fast matrix multiplication and Strassen’s algorithms to driving GPU/quantization performance at Meta and Facebook on projects like FBGEMM and Llama training/inference. Jianyu is an active open-source contributor to flagship projects such as PyTorch and Facebook Research xformers, where he implemented Triton kernels, flash-attention optimizations, and cross-hardware support including AMD and H100 GPUs. He combines deep numerical and systems expertise—low-precision arithmetic, vectorized GEMM kernels, and memory-layout tricks—to squeeze performance from modern accelerators. Not obviously, his background in both theoretical fast-matrix algorithms and hands-on kernel vectorization gives him a rare ability to translate algorithmic advances into production-grade GPU code. He is currently focused on AI system co-design for large models, balancing algorithmic scaling with inference deployment realities.
26 years of coding experience
12 years of employment as a software developer
PhD Computer Science, PhD Computer Science at The University of Texas at Austin
English, Chinese
Stackoverflow
333reputation
16kreached
2answers
0questions
Github Skills (47)
algorithm10
code-optimization10
algorithms10
pytorch10
c-language10
attention-mechanism10
matrix-multiplication10
gpu-programming10
machine-learning10
flash10
triton10
deep-learning10
self-attention10
gpu10
performance-optimization10
Programming languages (4)
Github contributions (5)
flame/how-to-optimize-gemm
Aug 2016 - May 2022
Role in this project:
Back-end Developer & Performance Engineer
Back-end Developer & Performance EngineerContributions:1 review, 29 commits, 4 PRs in 5 years 10 months
Contributions summary:Jianyu contributed code for matrix multiplication optimization techniques. They implemented different versions of the matrix multiplication algorithm, including 1x4 and 4x4 block sizes, with manual vectorization using SSE instructions. These optimized implementations suggest a focus on performance enhancement. Furthermore, the commits involved the creation of testing infrastructure to evaluate the performance improvements.
gemmmatrix-multiplicationcode-optimizationblisgotoblas
pytorch/FBGEMM
Nov 2018 - Jan 2023
FB (Facebook) + GEMM (General Matrix-Matrix Multiplication) - https://code.fb.com/ml-applications/fbgemm/
Role in this project:Back-end Developer
Back-end DeveloperContributions:30 reviews, 316 commits, 411 PRs in 4 years 2 months
Contributions summary:Jianyu focused on improving and optimizing the performance of the Fast and Brainy GEMM (FBGEMM) library for matrix multiplications, specifically the components related to quantized operations. Their work involved changes to improve the efficiency of floating-point to half-precision conversion, vectorizing kernel implementations, and improving the accuracy of the code. They also implemented enhancements for the use of a different storage method for weights on the GPU, and for memory access during matrix multiplication.
machine-learningmatrix-multiplication
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.