houj04
Computer Scientist at PaddlePaddle
Beijing, P.R.China
Summary
🤩
RockstarComputer Scientist with eight years of experience specializing in high-performance deep learning frameworks and MLOps, based in Beijing and affiliated with Tsinghua University's Department of Computer Science and Technology. Hands-on contributor to PaddlePaddle and PaddleHub, notably enabling multi-machine distributed training, parallel hyperparameter search, and expanding support for NPUs, XPUs (Kunlun) and GPUs to streamline model serving and auto-finetuning. Skilled in porting and optimizing operator kernels (e.g., gaussian_random, mul, cumsum) for XPU, adding device management and int64 support, and improving runtime memory behavior—work that improves both performance and cross-platform deployability. Pragmatic problem-solver who bridges algorithmic needs and production constraints, with a track record of turning research-grade models into scalable deployment pipelines. Notably comfortable working across hardware stacks, making models usable on emerging accelerators beyond conventional GPUs.
8 years of coding experience
Github Skills (30)
continuous-deployment10
modelio10
python10
modeler10
mpi10
machine-learning10
modello10
ml-deployment10
parallel-processing10
deeplearning-ai10
message-passing10
mlops10
deep-learning10
paddlepaddle10
awesomewm10
Programming languages (4)
Github contributions (5)
PaddlePaddle/Paddle
Jan 2021 - Jan 2023
PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）
Role in this project:
Back-end & ML Engineer
Back-end & ML EngineerContributions:553 reviews, 66 commits, 523 PRs in 2 years
Contributions summary:houj04 primarily contributed to the PaddlePaddle deep learning framework by implementing and modifying XPU (Kunlun) kernel functions for various operators. Their work involved porting operators such as `gaussian_random`, `mul`, `sigmoid_cross_entropy_with_logits`, and `cumsum` to XPU, alongside optimizations like adding memory copy wait commands. The user also added support for features such as the `set_device` function and for int64 data types. In addition to operator implementations, the user also fixed some bugs.
pytorchpythonparalleldeep-learningpaddlepaddle
PaddlePaddle/PaddleHub
Dec 2019 - Oct 2021
Awesome pre-trained models toolkit based on PaddlePaddle. (400+ models including Image, Text, Audio, Video and Cross-Modal with Easy Inference & Serving)【安全加固，暂停交互，请耐心等待】
Role in this project:MLOps Engineer
MLOps EngineerContributions:2 reviews, 15 commits, 9 PRs in 1 year 10 months
Contributions summary:houj04 focused on integrating multi-machine support and device compatibility (NPU, XPU, GPU) for the PaddleHub's auto-finetuning and model serving functionalities. They implemented parallel processing for hyperparameter search, enabling distributed training. Furthermore, the user expanded the hardware support for multiple pre-trained modules within the repository, increasing their usability. The commits demonstrate a strong emphasis on optimizing the performance and deployment of machine learning models.
image-textmodelvideoservingcomputer-vision
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.