Dinghua Li
Member Of Technical Staff at OpenAI
Bellevue, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolDinghua Li is a Member of Technical Staff at OpenAI with 12 years of experience building and productionizing large ML systems, previously focusing on TensorFlow, JAX, and GKE control planes at Google. He combines deep research training (PhD-level) with hands-on engineering, having led vehicle infrastructure and planning/control teams at TuSimple. An active open-source contributor to flagship projects like TensorFlow and JAX, he has made low-level changes for TPU/DTensor, XLA types, and export/serving compatibility that directly improve model deployment at scale. He excels at bridging research and production—optimizing kernels, shape polymorphism, and inference export paths to make advanced models operational. Based in Bellevue, WA, he brings both systems and ML expertise across cloud-native orchestration and model-serving stacks. Colleagues describe him as a pragmatic engineer who surfaces subtle runtime and compatibility issues before they reach production.
12 years of coding experience
7 years of employment as a software developer
The University of Hong Kong (HKU)Bachelor of Science (B.Sc.) Computational Mathematics, Bachelor of Science (B.Sc.) Computational Mathematics at Zhejiang University
English, Chinese
Github Skills (11)
machine-learning10
data-export10
exports10
deep-learning10
jax10
tensorflow10
python10
tpu10
exporter10
xla7
neural-network7
Programming languages (7)
Github contributions (5)
jax-ml/jax
May 2022 - Jan 2023
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Role in this project:
ML Engineer
ML EngineerContributions:2 reviews, 8 commits, 1 comment in 7 months
Contributions summary:Dinghua primarily contributed to the `jax-ml/jax` repository by implementing and improving features related to JAX and its integration with TensorFlow through `jax2tf`. They focused on enhancing the conversion process for specific JAX primitives like `ad_checkpoint.name_p`, and `dynamic_slice` and `dynamic_update_slice`. The user also worked on shape polymorphism support, specifically for `_unsafe_rbg_split`, and optimized the paged attention kernel, including introducing an "inline_seq_dim" mode.
pytorchpythonjitautomatic-differentiationgpu
tensorflow/tensorflow
Aug 2021 - Aug 2022
An Open Source Machine Learning Framework for Everyone
Role in this project:ML Engineer
ML EngineerContributions:38 commits in 1 year
Contributions summary:Dinghua contributed to the TensorFlow project by modifying TPU and saved model functionalities. They disabled autograph in the TPU initialization and shutdown functions. Additionally, they allowed ConcreteFunction aliases in SaveOptions for saved model functionality and introduced a feature to allow the setting of logical CPU devices in DTensor. Furthermore, they adjusted the code to include support for S4/U4 types in specific XLA operations.
pythondata-sciencedeep-learningmlmachine-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.