Connor Holmes
Researcher at OpenAI
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolConnor Holmes is a researcher and engineer with 11 years of experience building high-performance training and inference systems for robotics and large models, currently leading data, training, and inference efforts for Sora and robotics at OpenAI. He brings deep systems and CUDA expertise from his work on DeepSpeed—optimizing half-precision kernels and memory access for distributed deep learning—and a PhD in High Performance Computing from Colorado School of Mines. Connor’s career spans research roles at Microsoft and multiple research internships, combining production-focused optimization with rigorous academic methods. Based in San Francisco, he pairs low-level performance tuning with practical system design to move cutting-edge ML research into scalable infrastructure.
11 years of coding experience
7 years of employment as a software developer
Pine Creek High SchoolDoctor of Philosophy - PhD, High Performance Computing, Doctor of Philosophy - PhD, High Performance Computing at Colorado School of Mines
Spanish
Github Skills (9)
cuda10
pytorch10
machine-learning10
deep-learning10
gpu10
inference9
compression9
c-language8
cprogramming-language8
Programming languages (2)
Github contributions (5)
deepspeedai/DeepSpeed
Aug 2022 - Jan 2023
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
Role in this project:
ML Engineer
ML EngineerContributions:78 reviews, 88 commits, 53 PRs in 5 months
Contributions summary:Connor primarily contributed to the optimization and improvement of deep learning kernels within the DeepSpeed library. Their work focused on enhancing half-precision support, including modifications to header guards and code adjustments for half-precision computations in various CUDA kernels. They implemented memory access utilities and a reduction kernel utility, indicating a focus on performance and efficiency. The user's commits demonstrate a deep understanding of CUDA programming and optimization techniques for deep learning workloads.
deep-learningdeepspeeddistributed-traininginferencepytorch
cmikeh2/grnn
Jan 2019 - Jun 2019
Contributions:15 commits, 3 pushes in 4 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.