<![CDATA[fuleigang的专栏]]>https://blog.csdn.net/fuleigangzh-cnhttps://blog.csdn.net/<![CDATA[别再让 AI 只会「聊天」了腾讯 WorkBuddy 注册指南（附邀请]]>https://blog.csdn.net/fuleigang/article/details/165310482https://blog.csdn.net/fuleigang/article/details/165310482fuleigangMon, 14 Sep 2026 14:01:22 +0800<![CDATA[Java 程序员第 48 阶段14：Transformer 架构总览与自注意力直觉，KV Cache：Decoder 自回归生成如何复用历史计算]]>https://blog.csdn.net/fuleigang/article/details/165279385https://blog.csdn.net/fuleigang/article/details/165279385fuleigangMon, 14 Sep 2026 08:43:42 +0800<![CDATA[Java 程序员第 48 阶段13：Transformer 架构总览与自注意力直觉，用 Java 手写一个极简 Transformer 前向推理（玩具实现）]]>https://blog.csdn.net/fuleigang/article/details/165185322https://blog.csdn.net/fuleigang/article/details/165185322fuleigangSun, 13 Sep 2026 08:47:37 +0800<![CDATA[Java 程序员第 48 阶段12：Transformer 架构总览与自注意力直觉，从 Attention 到 Transformer Block：完整前向传播串讲]]>https://blog.csdn.net/fuleigang/article/details/165078859https://blog.csdn.net/fuleigang/article/details/165078859fuleigangSat, 12 Sep 2026 08:39:22 +0800<![CDATA[Java 程序员第 48 阶段11：Transformer 架构总览与自注意力直觉，Pre-LN vs Post-LN：归一化位置如何影响训练]]>https://blog.csdn.net/fuleigang/article/details/164959678https://blog.csdn.net/fuleigang/article/details/164959678fuleigangFri, 11 Sep 2026 09:43:33 +0800<![CDATA[Java 程序员第 48 阶段10：Transformer 架构总览与自注意力直觉，Layer Normalization 与残差连接：训练稳定性的双保险]]>https://blog.csdn.net/fuleigang/article/details/164807192https://blog.csdn.net/fuleigang/article/details/164807192fuleigangThu, 10 Sep 2026 09:43:56 +0800<![CDATA[Java 程序员第 48 阶段9：Transformer 架构总览与自注意力直觉，FFN 与激活函数：Transformer 的非线性来自哪里]]>https://blog.csdn.net/fuleigang/article/details/164716896https://blog.csdn.net/fuleigang/article/details/164716896fuleigangWed, 09 Sep 2026 09:42:23 +0800<![CDATA[Java 程序员第 48 阶段8：Transformer 架构总览与自注意力直觉，注意力复杂度 O(n²) 与 Flash Attention 优化思路]]>https://blog.csdn.net/fuleigang/article/details/164571649https://blog.csdn.net/fuleigang/article/details/164571649fuleigangTue, 08 Sep 2026 09:36:52 +0800<![CDATA[Java 程序员第 48 阶段7：Transformer 架构总览与自注意力直觉，Causal Mask：Decoder 如何做到看不见未来]]>https://blog.csdn.net/fuleigang/article/details/164450819https://blog.csdn.net/fuleigang/article/details/164450819fuleigangMon, 07 Sep 2026 09:36:02 +0800<![CDATA[Java 程序员第 48 阶段6：Transformer 架构总览与自注意力直觉，Self-Attention vs Cross-Attention：信息怎么流动]]>https://blog.csdn.net/fuleigang/article/details/164422990https://blog.csdn.net/fuleigang/article/details/164422990fuleigangSun, 06 Sep 2026 09:40:23 +0800<![CDATA[Java 程序员第 48 阶段5：Transformer 架构总览与自注意力直觉，Multi-Head Attention 并行计算：用 Java 模拟多头拆分与拼接]]>https://blog.csdn.net/fuleigang/article/details/164395427https://blog.csdn.net/fuleigang/article/details/164395427fuleigangSat, 05 Sep 2026 09:36:08 +0800<![CDATA[技术实践复盘从告警驱动到预案驱动：一个跨境电商平台的稳定性治理实践]]>https://blog.csdn.net/fuleigang/article/details/164373212https://blog.csdn.net/fuleigang/article/details/164373212fuleigangFri, 04 Sep 2026 17:36:53 +0800<![CDATA[Java 程序员第 48 阶段4：Transformer 架构总览与自注意力直觉，缩放点积注意力公式拆解：√dk 缩放因子的数学直觉]]>https://blog.csdn.net/fuleigang/article/details/164362041https://blog.csdn.net/fuleigang/article/details/164362041fuleigangFri, 04 Sep 2026 10:23:03 +0800<![CDATA[Java 程序员第 48 阶段3：Transformer 架构总览与自注意力直觉，位置编码 Position Encoding：为什么 Transformer 必须知道词序]]>https://blog.csdn.net/fuleigang/article/details/164295245https://blog.csdn.net/fuleigang/article/details/164295245fuleigangFri, 04 Sep 2026 10:10:21 +0800<![CDATA[Java 程序员第 48 阶段2：Transformer 架构总览与自注意力直觉，Encoder 与 Decoder 结构对比：BERT 用 Encoder、GPT 用 Decoder 的工程启示]]>https://blog.csdn.net/fuleigang/article/details/164360476https://blog.csdn.net/fuleigang/article/details/164360476fuleigangFri, 04 Sep 2026 09:49:17 +0800<![CDATA[Java 程序员第 48 阶段1：Transformer 架构总览与自注意力直觉，从 RNN 痛点理解注意力机制]]>https://blog.csdn.net/fuleigang/article/details/164331723https://blog.csdn.net/fuleigang/article/details/164331723fuleigangThu, 03 Sep 2026 13:58:27 +0800<![CDATA[Java 程序员第 47 阶段20：多租户知识库隔离设计，适配 SaaS 系统大模型架构，可观测性与租户级 SLA 监控体系]]>https://blog.csdn.net/fuleigang/article/details/164326559https://blog.csdn.net/fuleigang/article/details/164326559fuleigangThu, 03 Sep 2026 10:27:00 +0800<![CDATA[Java 程序员第 47 阶段19：多租户知识库隔离设计，适配 SaaS 系统大模型架构，混合隔离策略（共享加独享）与 SaaS 分层定价]]>https://blog.csdn.net/fuleigang/article/details/164326049https://blog.csdn.net/fuleigang/article/details/164326049fuleigangThu, 03 Sep 2026 10:16:00 +0800<![CDATA[Java 程序员第 47 阶段18：多租户知识库隔离设计，适配 SaaS 系统大模型架构，隔离方案压测：租户密度与性能拐点]]>https://blog.csdn.net/fuleigang/article/details/164324946https://blog.csdn.net/fuleigang/article/details/164324946fuleigangThu, 03 Sep 2026 09:57:20 +0800<![CDATA[Java 程序员第 47 阶段15：多租户知识库隔离设计，适配 SaaS 系统大模型架构，租户级限流、熔断与资源配额 Resource Quota]]>https://blog.csdn.net/fuleigang/article/details/164292587https://blog.csdn.net/fuleigang/article/details/164292587fuleigangWed, 02 Sep 2026 10:48:26 +0800