Juntang Zhuang
Member Of Technical Staff at xAI
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolJuntang Zhuang is a machine learning engineer and researcher with 7 years of experience, currently working on Grok-Next as a Member of Technical Staff at xAI in San Francisco. He was a core contributor at OpenAI, inventing the algorithm that enabled GPT-4 Turbo’s 128k long-context capability and helping author models and embeddings including GPT-4o, DALLE-3, and embedding-v3. Trained as a PhD in Biomedical/Medical Engineering with an MA in Statistics from Yale and a BE in Engineering Physics from Tsinghua, he blends rigorous academic grounding with production ML at scale. His open-source work includes contributions to the AdaBelief optimizer in PyTorch, reflecting a persistent interest in optimization and training dynamics. Colleagues describe him as someone who moves seamlessly between research and production engineering, often surfacing analytical insights that improve model training and inference.
7 years of coding experience
2 years of employment as a software developer
Doctor of Philosophy - PhD Biomedical/Medical Engineering, Doctor of Philosophy - PhD Biomedical/Medical Engineering at Yale UniversityMinor in law, Minor in law at Tsinghua University
Github Skills (9)
algorithm10
pytorch10
machine-learning10
deep-learning10
python10
optimisation10
optimizers10
optimization10
adam9
Programming languages (5)
Github contributions (5)
juntang-zhuang/Adabelief-Optimizer
Oct 2020 - Aug 2022
Repository for NeurIPS 2020 Spotlight "AdaBelief Optimizer: Adapting stepsizes by the belief in observed gradients"
Role in this project:
ML Engineer
ML EngineerContributions:383 commits, 14 PRs, 240 pushes in 1 year 10 months
Contributions summary:Juntang primarily contributes to the project by adding and updating various optimizers like AdaBelief and MSVAG within the PyTorch framework. They also modify the main training scripts to integrate and test the performance of these optimizers. The code changes include modifications to the AdaBelief optimizer, related setup files, and integration within different experiment settings, which shows an interest in optimizing model training. The user appears to be involved in evaluating and refining optimization techniques for deep learning models.
pytorchgradientsneurips-2020deep-learningneurips
juntang-zhuang/GSAM
Feb 2022 - Aug 2022
PyTorch repository for ICLR 2022 paper (GSAM) which improves generalization (e.g. +3.8% top-1 accuracy on ImageNet with ViT-B/32)
Contributions:103 commits, 87 pushes, 15 comments in 5 months
pytorchiclrimagenetgeneralizationdeep-learning
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.