Cao Gao
Software Engineer Tech Lead Manager at Meta
Ann Arbor, Michigan, United States
Summary
👤
Senior🎓
Top SchoolCao Gao is a software engineer and tech lead manager with 12 years of experience specializing in mobile systems, computer architecture, compilers, performance analysis, and ML-centric hardware/software co-design. Currently at Meta leading AI SW/HW co-design and previously at Google working on Edge TPU and Google Tensor, he bridges silicon architecture, compiler toolchains, and performance optimization to accelerate real-world ML workloads. His PhD work and internships reflect deep academic grounding in VLSI and mobile UX metrics, while his hands-on contributions to Facebook's FBGEMM (rowwise quantization for int2/int4/int8 with AVX2 optimizations) demonstrate practical impact in high-performance ML inference. Skilled at translating numerical methods into production-grade, architecture-aware implementations, he thrives on squeezing performance from both hardware and software. Based in Ann Arbor, he combines research rigor with engineering leadership to deliver efficient, scalable AI acceleration solutions.
12 years of coding experience
9 years of employment as a software developer
Master of Science (M.S.), Computer Engineering, Master of Science (M.S.), Computer Engineering at University of MichiganBachelor of Engineering, Electronic and Information Engineering, Bachelor of Engineering, Electronic and Information Engineering at Zhejiang University
English, Chinese
Github Skills (9)
avx10
quantization10
floating-point10
pytorch10
c-language10
float3210
cprogramming-language10
performance-optimization10
machine-learning9
Programming languages (3)
Github contributions (5)
pytorch/FBGEMM
Jun 2021 - Feb 2022
FB (Facebook) + GEMM (General Matrix-Matrix Multiplication) - https://code.fb.com/ml-applications/fbgemm/
Role in this project:
ML Engineer
ML EngineerContributions:9 commits, 9 PRs in 8 months
Contributions summary:Cao implemented and refined functions for quantizing and dequantizing floating-point embeddings, specifically focusing on rowwise quantization techniques for int2, int4, and int8 formats. These changes involved optimizing performance via AVX2 instructions and ensuring compatibility with different data types (float and float16). Furthermore, the user contributed to benchmark tests, and added support for INT8 and INT4 to the P2P benchmark.
matrix-multiplicationfacebookmultiplicationmatrixml-applications
3ancho/llvm-loop-fission
Dec 2013 - Dec 2013
Contributions:77 commits in 6 days
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.