Gang Liao

Hey there! Thanks for visiting. 👋

I’m a Research Scientist on the Ads ML Ranking Infra team at Meta (Facebook), with a primary focus on advancing inference optimizations and model/infra codesign.

Our work directly aims for step-function gains in ML Revenue by enabling larger, more complex inference models with exceptional cost efficiency. Our efforts span critical domains including inference-time scalability, user tower scaling, and sequence scaling, leveraging advanced AI hardware including Nvidia, MTIA, AMD, and next-generation CPUs. We meticulously optimize performance across the entire stack, from the PyTorch framework, operators/kernels, and data preprocessing to distributed inference, service, and model/system codesign. Our mission is to refine the foundational infrastructure for serving large-scale distributed ML models, tackling intricate technical challenges that directly impact Meta’s top line and process over 98% of the company’s annual revenue.

My academic foundation includes a Ph.D. in DB & Systems from the University of Maryland College Park, completed under the guidance of Daniel Abadi, with my dissertation focusing on the evolution of cloud data architectures. I also hold an M.S. in High-Performance Computing (HPC) from KAUST, where I conducted research with David Keyes.

My ongoing passion and interests are centered around the understanding, design, and development of data infrastructure and ML/LLM systems.