Sam Shleifer
Member Of Technical Staff at Thinking Machines Lab
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolSam Shleifer is a research engineer with 11 years of hands-on experience building and productionizing large-scale ML systems across NLP, vision, and time-series domains. He led engineering teams at Kensho and contributed core infrastructure at Facebook AI and Hugging Face—work that includes porting BART/Marian into transformers and developing FSDP used for training OPT-scale models. Most recently he drove efficiency and distillation research at Character.AI (serving tens of thousands QPS) before joining DeepMind and Thinking Machines Lab, blending systems engineering with applied model research. An active open-source contributor, Sam has helped evolve transformers, fairseq, and fairscale (including FSDP and optimizer sharding), showing a rare mix of backend, multi-GPU, and model-level expertise. He pairs an academic grounding (Yale BA, Stanford CS/AI master’s) with a knack for turning cutting-edge research into high-throughput production inference.
12 years of coding experience
9 years of employment as a software developer
Bachelor of Arts (B.A.) Economics, Bachelor of Arts (B.A.) Economics at Yale UniversityMilton AcademyMaster's degree CS/AI, Master's degree CS/AI at Stanford University
Portuguese
Stackoverflow
1,786reputation
174kreached
10answers
30questions
Github Skills (29)
pytorch10
artificial-intelligence10
python10
machine-learning10
deep-learning10
wandb10
natural-language-processing10
transfer-learning10
gpt10
nlp10
dialog9
testing9
distributed-training9
code-optimization9
bots9
Programming languages (14)
Github contributions (5)
huggingface/transformers
Sep 2019 - Jan 2021
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:256 reviews, 280 commits, 736 PRs in 1 year 3 months
Contributions summary:Sam appears to be working on documentation and removing unused code, with a focus on the Hugging Face Transformers library. They removed documentation for an unused kwarg, and merged a branch, indicating code maintenance and cleanup tasks. The user also made changes to the configuration and modeling files within the project, suggesting a role in the backend functionality of the library.
audioinferencemachine-learning-modelsmultimodaltransformers
facebookresearch/fairscale
Mar 2021 - Jan 2022
PyTorch extensions for high performance and large scale training.
Role in this project:ML Engineer
ML EngineerContributions:83 reviews, 11 commits, 39 PRs in 10 months
Contributions summary:Sam primarily contributed to the `fairscale` repository, which focuses on PyTorch extensions for large-scale training. The user added functionality for more fine-grained control over batchnorm buffers, and also contributed to supporting the saving and loading of optimizer states for the FullyShardedDataParallel (FSDP) module. This included fixing and improving the FSDP module, which is used for sharding model parameters and optimizer state across data parallel workers. The user also contributed to documentation related to FSDP usage.
pytorch
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.