<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Yizhou&apos;s Notes - Notes on AI infrastructure, inference systems, and engineering trade-offs.</title><description>Notes on AI infrastructure, inference systems, and engineering trade-offs.</description><link>https://yiz-liu.github.io/</link><item><title>从 vLLM Ascend 的图模式聊开去：跃出水面，高维视角下的回顾</title><link>https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-4/</link><guid isPermaLink="true">https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-4/</guid><description>把视角再抬高一层，从演进脉络与生态取舍回顾昇腾图模式这一路怎么走来。</description><pubDate>Fri, 22 May 2026 00:00:00 GMT</pubDate></item><item><title>从 vLLM Ascend 的图模式聊开去：进入深水区，回看那些“不得已”</title><link>https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-3/</link><guid isPermaLink="true">https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-3/</guid><description>跳出补丁视角，回看 Host 输入、硬件约束与这些特殊设计背后的现实原因。</description><pubDate>Fri, 15 May 2026 00:00:00 GMT</pubDate></item><item><title>从 vLLM Ascend 的图模式聊开去：理解全局后，再看设计方案</title><link>https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-2/</link><guid isPermaLink="true">https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-2/</guid><description>回到具体实现，集中看图模式里的动态性、分桶策略和不同设计取舍。</description><pubDate>Fri, 24 Apr 2026 00:00:00 GMT</pubDate></item><item><title>从 vLLM Ascend 的图模式聊开去：扩大视野，来聊聊编译</title><link>https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-1/</link><guid isPermaLink="true">https://yiz-liu.github.io/blog/graphs-in-vllm-ascend-1/</guid><description>把图模式与 torch.compile 拆开来看，集中聊编译链路、切图和融合优化。</description><pubDate>Fri, 17 Apr 2026 00:00:00 GMT</pubDate></item><item><title>从 vLLM Ascend 的图模式聊开去：有限视野下的初窥门径</title><link>https://yiz-liu.github.io/blog/graphs-in-vllm-ascend/</link><guid isPermaLink="true">https://yiz-liu.github.io/blog/graphs-in-vllm-ascend/</guid><description>从官方图模式文档出发，先看 Ascend 在图模式体验背后补了哪些工程差异。</description><pubDate>Fri, 10 Apr 2026 00:00:00 GMT</pubDate></item></channel></rss>