Linzhang Li

I'm Linzhang Li (aka Seven Streams, Yuchuan), an undergraduate in the ACM Honors Class at Zhiyuan College, Shanghai Jiao Tong University, pursuing a B.S. in Computer Science (September 2023 – June 2027, expected), with a GPA of 3.94/4.3.

My research interests lie in machine learning systems. My current focus is on building ML compiler foundations for agents and advancing agent-driven systems development. I also work extensively on structured generation for LLMs.

News

Publications

* indicates equal contribution.

Open Source Projects

XGrammar, Co-lead

XGrammar

A fast, flexible LLM structured generation engine with 100% guaranteed format compliance.

  • Contributor since February 2025.
  • Refactored the core engine, achieving 6× faster grammar compilation than the original implementation.
  • Designed a new algorithm for token mask cache reuse, achieving a further speedup of over 6× in grammar compilation over the refactored engine.

TVM FFI, Core Contributor

An open ABI and FFI for machine learning systems.

  • Contributor since June 2026.
  • Led development of Rust stub generation for cross-language support.
  • Co-led development of the Rust structural_map and structural_walk APIs for transformation and traversal.

TIRx Harness, Core Contributor

An open compiler harness for agentic GPU programming.

  • Contributor since June 2026.
  • Optimized CPU-based checkers, reducing peak memory usage to 1/20 of the original when checking an MoE kernel.
  • Designed GPU kernel tasks for Kimi Delta Attention (KDA).
  • Integrated the harness with a kernel benchmarking server.

TVM TIRx, Core Contributor

An open ML compiler foundation for the agentic era.

  • Contributor since June 2026.
  • Developed cross-language support, enabling compiler passes written in Rust to operate directly on existing C++ IR objects.

Blogs, Talks and Tutorials

XGrammar: Flexible and Efficient Structured Generation Engine for Large Language Models

vLLM Meetup (Shanghai) · Speaker

Selected Course Projects

Guqin

A Java implementation of a compiler for Mx*, an educational C-like language. Received full marks in Compiler Design and Implementation.

  • Implemented mem2reg, function inlining, SCCP, ADCE, and local CSE, together with linear-scan register allocation.
  • Generated code achieved runtime performance comparable to Clang at -O0 on the course benchmark suite.

Viola

A processor implemented in Verilog for a Computer Architecture course.

  • Implemented Tomasulo’s algorithm for out-of-order execution.
  • Implemented the RV32I instruction set, excluding memory-ordering and system instructions.

Awards and Honors

2026🏆SenseTime Scholarship (30 students in China)
2024, 2025, 2026🏆Excellent/Good Teaching Assistant (10% of TAs)
2025🏆OPPO Scholarship (30 students at SJTU)
2025🏆Sunshine Leadership Scholarship (35 students at Zhiyuan College)
2024, 2025🏆Academic Excellence Scholarship
2023, 2024, 2025🏆Zhiyuan Honors Scholarship (2% of SJTU students)
2024🏆Foresight-Sequoia Talent Development Fund (Top 5 at Zhiyuan College)
2024🏆Merit Student (1 student per class)
2025🥈Second place of team competition in chess of SJTU

Teaching Experience

2025, 2026 SpringHead Teaching Assistant, Data Structures
2024, 2025 FallHead Teaching Assistant, Programming
2025 SummerHead Teaching Assistant, Principles and Practice of Computer Algorithms

Leadership and Service

2026.9 – presentApache TVM Committer
2024.9 – 2025.9Section Leader, SJTU Sino-Western Orchestra
2023.9 – 2024.9Class President, ACM Honors Class (2023 cohort)

Skills

ProgrammingC++, Python, Rust, CUDA, Java, Go, Verilog
LanguagesMandarin Chinese (native); English (TOEFL iBT: 103)

Miscellaneous

I like music🎵. My favorite singers/bands are 花譜 and 和楽器バンド.

I'm a drummer🥁, and I played in the SJTU Sino-western Orchestra. I also worked as an organizer of several concerts of the orchestra.

I'm a fan of ACG games🎮, and my favorite game is Arknights. I also like its music🎵, and look forward to performing its music with others!