Fei Wang
Research Scientist at Google
Los Angeles, California, United States
Summary
👤
Senior🎓
Top SchoolFei Wang is a research scientist specializing in robustness for natural language processing, currently based in Los Angeles with eight years of industry and research experience. Now at Google after a year-long internship there, Fei has held applied science roles at AWS and Amazon and research positions at Tencent and USC/ISI, blending production-focused ML work with deep academic training (PhD/MS from USC). Fei contributes to prominent open-source projects like Hugging Face Transformers, improving documentation, pretrained model loading examples, and fixing multi-GPU and beam-search bugs—practical fixes that improve reproducibility for large-model users. Known for bridging rigorous research and engineering, Fei excels at turning robustness insights into tools and fixes that scale in real-world training and inference pipelines. Colleagues value Fei’s attention to reproducible examples and subtle bug hunts that prevent costly failures in multi-GPU and parallelized model runs.
8 years of coding experience
2 years of employment as a software developer
Doctor of Philosophy - PhD, Doctor of Philosophy - PhD at University of Southern CaliforniaBachelor of Engineering - BE, Bachelor of Engineering - BE at Renmin University of China
English, Chinese
Github Skills (15)
transformers10
pytorch10
machine-learning10
nlp10
python10
pre-trained-model9
bert9
language-model9
xnet8
hub8
gpt8
deeplearning-ai7
deep-learning7
jax6
flax6
Programming languages (3)
Github contributions (5)
huggingface/transformers
Aug 2019 - Sep 2019
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Role in this project:
ML Engineer
ML EngineerContributions:10 commits, 5 PRs, 4 comments in 1 month
Contributions summary:Fei primarily contributed to improving the examples and documentation within the repository, specifically for loading pretrained models. They also fixed comment typos and corrected examples of loading pretrained models in the docstrings across multiple model files (Bert, XLNet, XLM, GPT2, TransfoXL). Furthermore, the user addressed a bug related to multi-GPU training during language model finetuning and corrected an issue with LLaMa beam search when using parallelization.
pythonbertspeech-recognitionstate-of-the-artflax
FeiWang96/personal_website_backup
Feb 2022 - Mar 2024
Contributions:152 pushes in 2 years
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.