Peng Wang
Staff Software Engineer at Apple
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolPeng Wang is a Staff Software Engineer with 13 years of experience building and optimizing machine learning frameworks and on-device ML at major tech companies, currently based in Mountain View. With a PhD from MIT and prior roles on Google’s TensorFlow team, he combines deep research experience with production engineering to improve performance and TPU/XLA compatibility across libraries like Trax and TF-NumPy. His open-source contributions span low-level correctness work on the F* proof-oriented language to practical TensorFlow improvements—clearer candidate sampling docs, shape-inference fixes, and RNG guidance—showing attention to both formal foundations and developer experience. Comfortable moving between proof-oriented systems and high-performance ML runtimes, he brings rare expertise in verification, compiler-like extraction, and large-scale model deployment.
13 years of coding experience
12 years of employment as a software developer
Doctor of Philosophy (PhD) Computer Science, Doctor of Philosophy (PhD) Computer Science at Massachusetts Institute of TechnologyMaster of Science (MS) Computer Science, Master of Science (MS) Computer Science at Tsinghua University
Chinese, English
Github Skills (22)
programming-language10
dependent-types10
python10
machine-learning10
tpu10
numpy10
fstar10
ocaml10
fs10
deep-learning10
f10
tensorflow10
xla10
theorem-proving10
jax10
Programming languages (6)
Github contributions (5)
google/trax
Oct 2019 - Nov 2022
Trax — Deep Learning with Clear Code and Speed
Role in this project:
ML Engineer
ML EngineerContributions:56 commits, 1 push, 10 comments in 3 years 1 month
Contributions summary:Peng's commits primarily focus on enhancing the `google/trax` repository, a deep learning framework, by integrating support for XLA compilation and improving TPU compatibility. They added features for tf-numpy, a NumPy front-end for TensorFlow, to facilitate XLA-based compilation, and implemented improvements for single-core and multi-core TPU support. Additionally, the user fixed various issues and ensured compatibility with TensorFlow 2, demonstrating a focus on improving the framework's performance and usability, especially within the context of TPU acceleration.
pytorchspeeddeep-learningnumpyreinforcement-learning
tensorflow/tensor2tensor
Apr 2019 - Sep 2019
Library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research.
Role in this project:ML Engineer
ML EngineerContributions:12 commits in 5 months
Contributions summary:Peng primarily contributed to the `tensor2tensor` repository by making internal changes and improvements to the Trax library, a deep learning framework. These changes included type promotion, argument canonicalization, and improvements to the TensorFlow-NumPy integration to enable features like ResNet support and XLA. The user also added new functionalities such as `backend.eval_on_shapes` and the ability to toggle XLA within the TF-NumPy backend. The focus of the changes demonstrates an effort to optimize and extend the Trax library.
pytorchautoencoderdeep-learningmachine-translationreinforcement-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.