关于 AI Infra、Agent、机器学习实验与工程实践的技术笔记。https://jennyma11.github.io/zh-cnhttps://jennyma11.github.io/blog/speculative-decoding-methods/https://jennyma11.github.io/blog/speculative-decoding-methods/用结构图拆解 Draft Model、MTP、EAGLE、Medusa 与 N-gram：候选如何产生、如何验证，以及什么时候值得用。Wed, 23 Sep 2026 00:00:00 GMTAI InfraLLM InferenceSpeculative Decodinghttps://jennyma11.github.io/blog/attention-is-all-you-need-explained/https://jennyma11.github.io/blog/attention-is-all-you-need-explained/深入解析 Transformer 架构的开创性论文，理解 Self-Attention 如何革新了 Deep Learning。Mon, 18 May 2026 00:00:00 GMTTransformerDeep LearningNLPPaper Reviewhttps://jennyma11.github.io/blog/transformer-positional-encoding/https://jennyma11.github.io/blog/transformer-positional-encoding/理解 Transformer 如何通过 Positional Encoding 注入序列顺序信息，以及正弦函数设计的巧妙之处。Fri, 15 May 2026 00:00:00 GMTTransformerPositional EncodingDeep Learninghttps://jennyma11.github.io/blog/multi-head-attention-intuition/https://jennyma11.github.io/blog/multi-head-attention-intuition/从直觉到实现，理解为什么 Transformer 需要多个 Attention Head，以及它们各自在学习什么。Tue, 12 May 2026 00:00:00 GMTTransformerAttentionDeep Learninghttps://jennyma11.github.io/blog/building-a-durable-writing-system/https://jennyma11.github.io/blog/building-a-durable-writing-system/一种轻量级的方法，让技术笔记在第一周之后仍然有用。Fri, 01 May 2026 00:00:00 GMTWritingSystemProductivityhttps://jennyma11.github.io/blog/astro-content-collections-notes/https://jennyma11.github.io/blog/astro-content-collections-notes/为什么类型化的 Frontmatter 对于专注的个人发布工作流来说已经足够。Sat, 18 Apr 2026 00:00:00 GMTAstroMDXContenthttps://jennyma11.github.io/blog/machine-learning-experiment-logs/https://jennyma11.github.io/blog/machine-learning-experiment-logs/一种紧凑的结构，使实验更容易在以后比较。Sun, 29 Mar 2026 00:00:00 GMTMachine LearningResearchSystem