Christian Sarofeen
Director Of Engineering For Deep Learning Frameworks at NVIDIA
California, United States
Summary
🤩
Rockstar🎓
Top SchoolChristian Sarofeen is a Director of Engineering who leads development of PyTorch, JAX, and the nvFuser deep learning compiler at NVIDIA, optimizing frameworks and automatic code generation to maximize performance on NVIDIA GPUs. With nine years in industry and a background in high-performance scientific computing, he brings deep expertise in C++, CUDA, distributed parallelism, and mixed-precision training—having contributed kernel-level optimizations to torch/cunn, cutorch, and NVIDIA/apex. He combines hands-on low-level engineering (FP16 and CUDA9 compatibility, fused RNN kernels) with strategic leadership across multiple framework teams. Based in California, he is recruiting engineers who are passionate about large-scale frameworks and CUDA performance, and uniquely blends academic HPC experience with production-grade ML systems engineering.
9 years of coding experience
13 years of employment as a software developer
BS, Mechanical Engineering, 3.48, BS, Mechanical Engineering, 3.48 at Penn State University
Stackoverflow
2,232reputation
226kreached
61answers
1question
cuda
top-5%
Github Skills (28)
pytorch10
gru10
c-language10
distributed-training10
gpgpu10
rnn-model10
n10
multiprecision10
kernel10
lstm10
deep-learning10
parallel-computing10
gpu10
kernel-mode10
cuda10
Programming languages (6)
Github contributions (5)
NVIDIA/apex
Apr 2018 - Aug 2019
A PyTorch Extension: Tools for easy mixed precision and distributed training in Pytorch
Role in this project:
ML Engineer
ML EngineerContributions:2 reviews, 48 commits, 15 PRs in 1 year 4 months
Contributions summary:Christian primarily contributed to the `apex` repository, which focuses on PyTorch extensions for mixed precision and distributed training. Their work involved fixing distributed training issues, refactoring distributed components, and addressing race conditions within the distributed data parallel (DDP) module. They also made changes to the build process, including CUDA compilation and finding CUDA libraries. The contributions are focused on improving the efficiency and robustness of distributed training functionalities.
distributed-trainingpytorch
torch/cutorch
Jun 2017 - Aug 2017
A CUDA backend for Torch7
Role in this project:Back-end Developer
Back-end DeveloperContributions:5 commits, 6 PRs, 23 comments in 2 months
Contributions summary:Christian primarily focused on optimizing and extending CUDA functionality within the Torch7 CUDA backend. Their contributions include fixing half-precision floating-point (FP16) issues for CUDA 9, implementing a non-contiguous dimension reduction kernel to improve performance on small tensors, and updating code for CUDA 9 compatibility. The user also addressed issues related to larger grid sizes for THCApply and made adjustments to the grid size for batch cat tensor operations.
backendcuda
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.