Edward Hu
PhD Student at University of Pennsylvania
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolEdward Hu is a PhD student and AI researcher based in San Francisco with seven years of experience at the intersection of deep learning, reinforcement learning, and robotics. His work spans high-impact academic publications (ICLR, CoRL, ICRA) and industry research, including a recent Microsoft internship that produced an ICLR'25 paper on pretraining LLMs for planning. He contributes to widely used open-source projects—applying LoRA to DeBERTa/RoBERTa and improving Microsoft’s mup library—showing practical expertise in model adaptation and optimizer parametrization. His record includes a Best Paper award (CoRL22) and multiple ICLR spotlights, reflecting both technical depth and strong experimental results. Comfortable moving between virtual and physical domains, he blends robotics intuition with large-model engineering to build scalable learning systems.
7 years of coding experience
4 years of employment as a software developer
Master's degree Computer Science, Master's degree Computer Science at University of Southern CaliforniaLynbrook High SchoolDoctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at University of Pennsylvania
English, Chinese, Spanish
Github Skills (10)
lora10
roberta10
pytorch10
machine-learning10
language-model10
deep-learning10
python10
optimization10
transformers9
unit-testing9
Programming languages (4)
Github contributions (5)
microsoft/LoRA
Jun 2021 - Nov 2022
Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"
Role in this project:
ML Engineer
ML EngineerContributions:4 releases, 19 commits, 14 PRs in 1 year 4 months
Contributions summary:Edward applied LoRA (Low-Rank Adaptation) to DeBERTa and RoBERTa models within the repository. They modified the model's architecture by integrating `loralib`, specifically targeting `DisentangledSelfAttention` and related components. Additionally, the user updated and refactored the `layers.py` file to incorporate the LoRA functionality, including weight merging during training. Furthermore, they changed transpose calls within the `layers.py` file.
debertapytorchlanguage-modelnlplarge-language-models
microsoft/mup
Mar 2022 - Jan 2023
maximal update parametrization (µP)
Role in this project:ML Engineer
ML EngineerContributions:25 commits, 4 PRs, 21 pushes in 10 months
Contributions summary:Edward primarily contributes to testing and expanding the functionality of the `mup` library, focused on Maximal Update Parametrization. Their work includes adding tests for meta tensors, which are likely used for optimization techniques within the library. The user also added an option to skip scaling weight decay for decoupled optimizers, enhancing the library's flexibility. The contributions indicate a focus on improving the core functionality and usability of the library for deep learning applications.
pytorchtransformerspythonparametrizationdeep-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.