Xing Liu’s Home Page
刘星的个人主页
Hi, I am Xing Liu (刘星). Welcome to my cyber home!
I’m an engineer interested in LLM Serving Efficiency. Currently working at Nebius as a Principal ML Solution Architect.
What’s New
- NEW! This site now looks like it’s 1997. On purpose: everything still works, and the colophon explains why.
- New project: YOY-Labs-Website. The source of yoylabs.net.
- New post: Neuron Learning Notes. Notes on AWS Trainium and NKI: the memory hierarchy, data representation, indexing, tiling, DMA and logical NeuronCores.
- New post: Nano-vLLM Study Notes. A file-by-file tour of nano-vLLM: the scheduler, block manager, model runner, tensor parallelism and Qwen3.
- New post: From Training Paradigms to Inference-First Engineering. Why time to first token, inter-token latency and throughput, not training, now set the agenda for AI engineering.
Older news: all 12 posts · all 7 projects · RSS feed
On this site
Elsewhere
- GitHub
- RSS feed (full text)
- How this site is made