I go by `SiriusNEO` on most websites. And I usually use `Chaos`
as a nickname (a variant of my real name). (Github | Zhihu |
Scholar)
Research
My research interest lies in building practical, scalable and efficient systems for
machine learning (MLSys) like serving system and deep learning compiler, with a special focus on
finding optimization opportunities in foundation models and using appropriate abstractions to automate them.
My long-term research goal is to make powerful AI accessible to everyone. Currently, I'm working on the following topics.
If you share the same interest with me, please don't hesitate to email me and we can schedule a discussion.
(Algorithm/Kernel Level) Efficient Sparse Attention with its applications in long context LLM inference / reasoning.
(Compiler Level) Efficient Python-Native Compilation in different hardware backends with some techniques like fast tuning.
And some of my previous projects are listed below:
ParrotServe: Serving system for LLM
applications with Semantic Variable abstraction (Pulished in OSDI'24).
[2024/6] Graduated from SJTU with Zhiyuan Outstanding Student Scholarship! My bachelor thesis
is awarded with Excellent Bachelor Thesis (Top 1%) in SJTU!
[2024/3] Parrot is accepted by OSDI'24! Code and paper are both released.
[2023/9] I will be joining IIIS, Tsinghua as a PhD student starting 24Fall.
Pokémon since I was eight years old and own a collection of its
game cards (from G1 to G9)!