Daniel Sohn
Member Of Technical Staff, AI at Microsoft AI
United States
Summary
🤩
Rockstar🎓
Top SchoolDaniel Sohn is an AI-focused software engineer with a decade of experience building and optimizing large-scale deep learning systems, currently working on LLMs at Microsoft AI after a multi-year stint at Google DeepMind. He has hands-on expertise making Transformers and PyTorch/XLA run efficiently on TPUs—contributions to high-profile open-source projects like Hugging Face Transformers and pytorch/xla improved TPU training scripts, convergence, and CI/build processes. At DeepMind he worked across Gemini, PaLM 2, Bard post-training, and YouTube-applied ML, blending research-grade model work with production engineering. His background spans core ML infrastructure, model serving, and performance tuning, informed by an earlier biomedical-engineering research foundation from Johns Hopkins that surfaces in his attention to experimental rigor.
10 years of coding experience
5 years of employment as a software developer
Johns Hopkins University
English, Korean, Spanish
Stackoverflow
881reputation
58kreached
29answers
0questions
tensorflow
top-5%
google-colaboratory
top-5%
Github Skills (23)
transformers10
pytorch10
python10
machine-learning10
conda10
tpu10
cicd10
distributed-computing10
deep-learning10
tensorflow10
natural-language-processing10
build-automation10
xla10
nlp10
google-colaboratory9
Programming languages (5)
Github contributions (5)
pytorch/xla
Nov 2018 - Sep 2021
Enabling PyTorch on XLA Devices (e.g. Google TPU)
Role in this project:
Back-end & DevOps Engineer
Back-end & DevOps EngineerContributions:1 release, 154 reviews, 284 commits in 2 years 10 months
Contributions summary:Daniel focused on improving the build and release process for the PyTorch/XLA project. They added functionality to append the git commit ID to the wheel version, and enabled continuous integration using a CI/CD pipeline, including setting up the build environment, installing dependencies, and installing the conda environment for the builds. Further, they updated the build environment with recent PyTorch versions and implemented changes to support building upstream PyTorch.
pytorchtpuxlacompilerdeep-learning
tensorflow/tpu
Jan 2019 - Oct 2019
Reference models and tools for Cloud TPUs.
Role in this project:ML Engineer
ML EngineerContributions:12 commits, 13 PRs, 6 pushes in 9 months
Contributions summary:Daniel primarily contributes to the development and maintenance of machine learning models within the TensorFlow ecosystem, specifically for the Cloud TPUs. Their work includes implementing load testing for inference servers, and integrating model serving and export functionalities. They are involved in adapting and optimizing existing models, such as Mask R-CNN and ResNet, to run efficiently on TPUs, including the addition of features such as warmup requests.
cloud
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.