Sam Gross
Software Engineer at Meta
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolSam Gross is a software engineer with 14 years of experience based in New York, currently building systems at Meta and previously co-founding Jam Labs. He specializes in low-level back-end engineering for ML and deep learning frameworks, contributing substantial compatibility, performance, and CUDA optimizations to influential projects like Torch, cuDNN bindings, and PyTorch vision and examples. His work spans ML model implementations (ResNet, VGG, Inception), GPU-accelerated layers and data-parallel improvements, and interpreter-level concurrency efforts such as GIL-free Python and pybind11 thread-safety. Comfortable across C/C++, Lua, and Python, he combines production-grade systems thinking with research-adjacent model engineering—an unusual blend that enables both core runtime fixes and high-level model integrations.
14 years of coding experience
5 years of employment as a software developer
M.Eng, B.S, Electrical Engineering and Computer Science, M.Eng, B.S, Electrical Engineering and Computer Science at Massachusetts Institute of TechnologyThomas Jefferson High School for Science and Technology
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (65)
cpython10
checkpoint10
batch-normalization10
operation10
python10
multithreading10
atomics10
testing10
c1110
c1710
lua10
openmp10
deep-learning10
cudnn10
concurrency10
Programming languages (18)
Github contributions (5)
facebookarchive/fb.resnet.torch
Jan 2016 - Oct 2019
Torch implementation of ResNet from http://arxiv.org/abs/1512.03385 and training scripts
Role in this project:
ML Engineer
ML EngineerContributions:28 commits, 27 PRs, 40 pushes in 3 years 10 months
Contributions summary:Sam made several contributions focused on improving the ResNet model implementation. These changes included adding support for checkpointing to save and load model states, incorporating a pre-activation ResNet definition, and addressing issues in the data processing pipeline such as aspect ratio jitter. They also fixed errors in the code related to computing top-5 accuracy and shareGradInput usage, ensuring the model ran correctly and efficiently.
pytorcharxivabsdeep-learningtorch
colesbury/nogil
Aug 2019 - Jan 2023
Multithreaded Python without the GIL
Role in this project:Back-end Developer
Back-end DeveloperContributions:6 releases, 2 reviews, 984 commits in 3 years 5 months
Contributions summary:Sam contributed to the `nogil` project, which focuses on multithreaded Python without the Global Interpreter Lock (GIL). Their commits primarily address low-level implementation details related to atomic operations and compiler compatibility. The user implemented and refined atomic operations for various data types, added support for GCC builtin atomics, and fixed a bug in the `platform.python_compiler()` function. They also updated the bundled `pip` to work with the modified Python environment.
gilpythonpython3multithreaded
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.