Yanbo Liang
Software Engineer at The Apache Software Foundation
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolYanbo Liang is a Principal Researcher and tech lead based in San Francisco with 11 years of experience building large-scale AI infrastructure and foundation model pre-training systems. Currently at ByteDance Seed, he architects compiler-first distributed training with Megatron-style parallelism and drives cross-stack optimizations—Attention/MoE kernels, comp–comm overlap, kernel fusion, CUDA Graphs and HW–SW co-design—for production-grade massive-scale training. Previously a PyTorch core engineer at Meta, Yanbo contributed to torch.compile (TorchDynamo/TorchInductor) and efficiency work for Llama training and FlexAttention, bringing deep compiler and GPU-kernel expertise. He is an active open-source committer (Apache Spark PMC) with notable contributions across XGBoost-Spark, PyTorch, Keras and Spark that bridge research, systems and applied ML. Colleagues know him for translating low-level kernel and compiler innovations into robust, deployable training platforms that materially improve training throughput and reliability.
11 years of coding experience
8 years of employment as a software developer
Master's degree Computer Science, Master's degree Computer Science at Beihang University
English, Chinese, Japanese
Stackoverflow
94reputation
11kreached
9answers
0questions
Github Skills (41)
text-generation10
pytorch10
xgboost10
spark10
back-end-development10
python10
data-science10
machine-learning10
feature-engineering10
java10
scala10
linear-regression10
transformer-models10
javas10
keras10
Programming languages (7)
Github contributions (5)
pytorch-labs/gpt-fast
Dec 2023 - Oct 2024
Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.
Role in this project:
ML Engineer
ML EngineerContributions:8 reviews, 30 PRs, 13 pushes in 9 months
Contributions summary:Yanbo's primary contribution involved integrating the Mixtral-8x7B model into the `gpt-fast` repository, evidenced by the creation of a dedicated sub-folder and associated modifications to the generation script and README.md. These changes included adding the Mixtral-8x7B model, removing unused logic, and updating the README to reflect the integration. This demonstrates the user's focus on expanding the capabilities of the project by incorporating a new transformer model.
pytorch/pytorch
May 2019 - Jan 2023
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:1148 reviews, 63 commits, 515 PRs in 3 years 9 months
Contributions summary:Yanbo's contributions primarily focused on enhancing the PyTorch framework, specifically within the domain of deep learning and GPU acceleration, as indicated by the repository's description and topics. Their work included implementing improvements to the Inductor backend, which involves optimizing the compilation of models. The commits demonstrate involvement in addressing issues related to autograd.Function, improving code generation and testing, and ensuring functionality for users in various areas within the PyTorch ecosystem.
pythongpu-accelerationdeep-learninggpunumpy
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.