About
llm-sys studies how large language model architectures evolve and how to make them efficient: attention and sparsity, model structure, and the kernels and serving stacks that run them. We publish papers with code and write up the ideas here.
People
Taosong Fang
Researcher, Dots Studio, Xiaohongshu
Taosong Fang is a researcher at Dots Studio, Xiaohongshu, working on LLM pretraining and on co-designing model backbones with the infrastructure that trains and serves them. Taosong’s research interests are broad: earlier work spans LLM post-training and machine learning systems for LLM training, inference, and agentic workflows. Taosong received a master’s degree from the Institute of Software, Chinese Academy of Sciences, in 2026.
Fanxu Meng
Researcher, Dots Studio, Xiaohongshu
Fanxu Meng is a researcher at Dots Studio, Xiaohongshu, and holds a Ph.D. from Peking University. He has authored several first-author papers at leading conferences such as NeurIPS and ICML, including PiSSA and TransMLA. His current research focuses on efficient sparse attention, with representative works including HISA, ResMISA, and DoSA.
Contact
Code is on GitHub. For questions about a post, write to the corresponding author listed on it, or to fangtaosong@xiaohongshu.com.