Erjia Guan
Software Engineer at Meta
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolErjia Guan is a software engineer based in New York with five years of experience building reliable, production-grade systems at Meta and earlier AI-focused startups. He contributes to core PyTorch projects—helping improve DataLoader, distributed shuffling, and torchdata datapipes—bringing practical expertise in back-end development, MLOps, and DevOps workflows like CI/release automation. His work emphasizes robustness in distributed data loading, deterministic behavior across processes, and improved testing and serialization, reflecting a strong focus on code quality and reproducibility. A CMU-trained engineer with a B.E. from Tianjin University, he blends academic grounding with hands-on open-source impact on one of the most widely used deep learning frameworks.
5 years of coding experience
3 years of employment as a software developer
Bachelor of Engineering (B.E.), Bachelor of Engineering (B.E.) at Tianjin UniversityMaster's degree, AIS, Master's degree, AIS at Carnegie Mellon University
Github Skills (30)
pytorch10
data-pipelines10
python10
data-engineering10
python-multiprocessing10
distributed-systems10
software-quality10
cicd10
data-loading10
multiprocessing10
multi-process10
data-pipeline10
testing9
back-end-development9
rep9
Programming languages (5)
Github contributions (5)
pytorch/data
Feb 2021 - Jan 2023
A PyTorch repo for data loading and utilities to be shared by the PyTorch domain libraries.
Role in this project:
Back-end Developer & DevOps Engineer
Back-end Developer & DevOps EngineerContributions:6 releases, 998 reviews, 357 commits in 1 year 11 months
Contributions summary:Erjia primarily contributed to the development of data loading and utility features for PyTorch, evidenced by code changes in `torchdata/datapipes` and examples. They also focused on improvements in testing and configuration, particularly related to flake8 and code style, which suggests a focus on code quality. Additionally, the user implemented the continuous integration (CI) matrix and release workflow, showing involvement in DevOps and build processes.
pytorchpythondeep-learningtorchmachine-learning
pytorch/pytorch
Sep 2020 - Dec 2022
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:Back-end Developer & MLOps Engineer
Back-end Developer & MLOps EngineerContributions:839 reviews, 264 commits, 202 PRs in 2 years 2 months
Contributions summary:Erjia primarily focused on improving the `DataLoader` component of PyTorch, specifically addressing issues related to distributed and persistent data loading. They implemented deterministic behavior for `ShufflerDataPipe` in distributed environments, ensuring consistent shuffling across processes. Furthermore, the user made improvements to the sharing of the random seed via process group and ensured that the `DataLoader` code base could deal with the custom sharding data pipes. They also worked on resolving bugs related to multi-processing and serialization, and improved error handling for distributed seed sharing to enhance the reliability of the data loading process.
pythongpu-accelerationdeep-learninggpunumpy
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.