Xi Lin

林熙

Hangzhou, Zhejiang · erix0-remove-[email protected]

🏫 I’m a senior student at Zhejiang University, majoring in Computer Science and Technology.

🔭 I’m interested in Efficient AI through algorithm–system co-design, focusing on hardware-friendly sparse and quantized module design as well as efficient inference strategies.

🚀 I’m currently a leader of ZJUSCT, a super computing team at Zhejiang University which has won several international super computing competitions.

News

⚡ FlashQLA v0.1.1 released

2026-06-30

FlashQLA v0.1.1 has been released, adding intra-card sequence parallelism for the backward pass and SM100 support. Check out the GitHub repository.

🎉 TriAttention accepted to ICML 2026

2026-05-10

TriAttention has been accepted to ICML 2026. It is an efficient KV cache compression method for long reasoning in large language models. By leveraging Q/K concentration in pre-RoPE space and a trigonometric series to score key importance, TriAttention matches Full Attention reasoning accuracy on AIME25 while achieving 2.5x higher throughput or 10.7x KV memory reduction. Check out the project page and code!

⚡ FlashQLA v0.1.0 released

2026-04-29

FlashQLA v0.1.0 has been released as an open-source high-performance linear attention kernel library for Qwen. Check out the GitHub repository.

Publications

A collection of my research publications and academic papers.

TriAttention: Efficient Long Reasoning with Trigonometric KV Compression

TriAttention: Efficient Long Reasoning with Trigonometric KV Compression

Weian Mao*, Xi Lin*, Wei Huang*, Yuxin Xie, Tianfu Fu, Bohan Zhuang, Song Han, Yukang Chen

* Equal contribution

ICML 2026April 2026

TriAttention proposes a novel KV cache compression approach for long reasoning in LLMs. It leverages trigonometric series based on fixed centers in pre-RoPE space to score key importance, achieving 2.5x higher throughput or 10.7x KV memory reduction while matching Full Attention reasoning accuracy.
PSA: Pyramid Sparse Attention for Efficient Video Understanding and Generation

PSA: Pyramid Sparse Attention for Efficient Video Understanding and Generation

Xiaolong Li*, Youping Gu*, Xi Lin*, Weijie Wang, Bohan Zhuang

* Equal contribution

PreprintDecember 2025

PSA introduce an efficient attention mechanism to accelerate video understanding and generation. It leverages a multi-level sparse attention strategy, enabling the model to effectively mitigates information loss while preserving computational efficiency under a low compute budget.

Projects

Selected systems and infrastructure projects around efficient AI, LLM inference, and high-performance computing.

FlashQLA: Flash Qwen Linear Attention

FlashQLA: Flash Qwen Linear Attention

ProjectApril 2026

A high-performance linear attention kernel library for Qwen, built on TileLang. FlashQLA optimizes GDN Chunked Prefill with fused kernels, gate-driven context parallelism, and hardware-friendly reformulation.

Experience

Research Intern

Qwen Infra, Alibaba Cloud

Research intern in Qwen Infra, working on efficient inference infrastructure for Qwen.

  • Contributed to FlashQLA, a high-performance TileLang-based linear attention kernel library for Qwen.
  • Focused on efficient GPU kernel design, including fused forward/backward kernels, GDN Chunked Prefill optimization, gate-driven context parallelism, and hardware-friendly reformulation.
Mar 2026 - Present

Team Leader / Member

Led and contributed to student teams in supercomputing competitions, focusing on machine learning systems (MLSys) and high-performance computing (HPC) solutions.
October 2023 - May 2026

Teaching Assistant

Zhejiang University
  • HPC101, Summer 2024 and Summer 2025
  • Computer Architecture, Spring 2025
2024 - 2025

Awards

5th Linear Solver Algorithm and Performance Optimization Competition

Third Prize

Aug 2025

Zhejiang University Scholarship

First Prize

Oct 2024

4th Linear Solver Algorithm and Performance Optimization Competition

Second Prize

Jul 2024

ISC 2024 Student Cluster Competition (online)

4th Place

May 2024

ASC24 Student Supercomputer Challenge

Second Prize

Feb 2024

Zhejiang University Scholarship

Third Prize

Oct 2023

Education

Zhejiang University

Bachelor of Science
Computer Science and Technology

2022 - Present

Blog

Visit my personal blog to read about my thoughts, experiences, and technical insights.

Nifty tech tag lists from Wouter Beeftink