Qian Yu
Backend Engineer at TikTok
Seattle, Washington, Singapore
Summary
👤
Senior🎓
Top SchoolQian Yu is a software engineer with nine years of experience specializing in deep learning and high-performance computing, currently working at TikTok in Seattle. She has led large-scale machine translation services at Trip.com, operating sub-100ms responses across hundreds of business scenarios and processing hundreds of millions of characters daily. Proficient in Python, C++, CUDA, and TensorFlow, she has contributed performance-focused improvements to prominent open-source projects like Tensor2Tensor—optimizing transformer decoding and attention visualization—and to TensorFlow Models. Her background spans production ML systems (MT, pose estimation, face recognition) and research-driven feature engineering, bridging academic rigor from Northeastern University with hands-on deployment experience. An understated strength is her knack for squeezing latency and concurrency gains from complex models, enabling real-time, high-throughput inference in production.
9 years of coding experience
1 year of employment as a software developer
Exchange Computer Science, Exchange Computer Science at University of WashingtonBachelor of Science - BS Computer Science, Bachelor of Science - BS Computer Science at National University of Singapore
Chinese, English
Github Skills (11)
attention-mechanism10
transformer-models10
caching10
machine-learning10
deep-learning10
tensorflow10
python10
machine-translation8
nlp7
tpu5
reinforcement-learning4
Programming languages (5)
Github contributions (5)
tensorflow/tensor2tensor
Dec 2018 - Jan 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:6 commits, 3 PRs, 6 comments in 29 days
Contributions summary:Qian contributed significantly to the tensor2tensor repository by implementing and refining caching mechanisms for the transformer model, particularly focusing on relative dot-product attention. This included modifying the `common_attention.py` and `transformer.py` files to support faster decoding using these mechanisms. Furthermore, the user enabled and improved the visualization of relative dot-product attention and fixed potential errors within the beam search process when using certain feed-forward network configurations.
pytorchautoencoderdeep-learningmachine-translationreinforcement-learning
aeloyq/BerkeleyRemoteLab
Sep 2017 - Jun 2021
Contributions:11 pushes, 1 branch in 3 years 9 months
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.