https://zhiyuan1i.github.io/en/ Recent content on Zhiyuan Li https://zhiyuan1i.github.io/images/papermod-cover.png https://zhiyuan1i.github.io/images/papermod-cover.png Hugo -- 0.155.3 en-US Sat, 21 Feb 2026 10:44:23 +0000 https://zhiyuan1i.github.io/en/posts/dplr-mathematics/ Sat, 21 Feb 2026 10:44:23 +0000 https://zhiyuan1i.github.io/en/posts/dplr-mathematics/ A deep dive into the chunk-wise parallel algorithm for DPLR, understanding the WY representation of explicit diagonal-plus-low-rank transition matrices, and exploring the unified framework with KDA/IPLR https://zhiyuan1i.github.io/en/posts/kda-mathematics/ Tue, 17 Feb 2026 03:00:00 +0000 https://zhiyuan1i.github.io/en/posts/kda-mathematics/ A deep dive into the chunk-wise parallel algorithm of KDA, establishing the theoretical framework of Affine transformations from basic matrix multiplication lemmas https://zhiyuan1i.github.io/en/about/ Mon, 16 Feb 2026 00:00:00 +0000 https://zhiyuan1i.github.io/en/about/ <h2 id="zhiyuan-li">Zhiyuan Li</h2> <p>AI Infra Engineer at <a href="https://www.moonshot.cn/">Moonshot AI</a>.</p> <p>Focusing on efficient implementation and optimization of <strong>Linear Attention</strong>. Honored to have contributed to the development of <a href="https://github.com/MoonshotAI/Kimi-Linear">Kimi Linear</a> and <strong>Kimi Delta Attention (KDA)</strong>, learning a lot from the excellent colleagues on the team.</p> <hr> <h3 id="-research-interests">🔬 Research Interests</h3> <ul> <li><strong>Linear Attention</strong>: Exploring sub-quadratic sequence modeling methods for more efficient long sequences</li> <li><strong>Efficient Inference Optimization</strong>: CUDA kernel optimization, memory bandwidth optimization, Tensor Core acceleration</li> <li><strong>Model Architectures</strong>: RWKV-6/7, Gated DeltaNet, and other novel attention mechanisms</li> </ul> <h3 id="-open-source-contributions">🚀 Open Source Contributions</h3> <ul> <li>Contributed to <a href="https://github.com/fla-org/flash-linear-attention"><strong>flash-linear-attention</strong></a> community project - Efficient implementations of state-of-the-art linear attention models</li> </ul> <h3 id="-articles">📝 Articles</h3> <ul> <li><a href="https://zhuanlan.zhihu.com/p/1989809041849988324">Learning KDA from Scratch - Part 1</a> - Understanding KDA parallelization from an Infra perspective (Chinese)</li> </ul> <hr> <h3 id="-about-this-site">💬 About This Site</h3> <p>This site documents my learning notes, technical articles, and some immature thoughts in the AI Infra field. I&rsquo;m still learning, so please feel free to point out any mistakes. Looking forward to exchanging ideas with you.</p> https://zhiyuan1i.github.io/en/posts/tech-stack/ Mon, 16 Feb 2026 00:00:00 +0000 https://zhiyuan1i.github.io/en/posts/tech-stack/ Introduction to the tech stack used to build this site