<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[LiuYQi的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/LiuYQi</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; LiuYQi]]></copyright><item><title><![CDATA[Raft 算法深度解析：角色、选举、日志复制与分区处理优化]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/152163164</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/152163164</guid><author>LiuYQi</author><pubDate>Sat, 27 Sep 2025 11:18:28 +0800</pubDate><description><![CDATA[Raft算法是一种分布式一致性算法，通过角色分工、任期机制和多数票决策实现数据一致性。核心角色包括Leader（唯一处理写请求）、Follower（被动响应）和Candidate（选举状态）。算法通过选举超时和心跳超时机制确保单一Leader，并利用日志复制保证数据一致性。网络分区时，只有多数节点子集群能产生有效Leader；分区恢复后，通过任期和日志对比自动解决多Leader问题。Raft的核心保障机制包括任期机制、多数票决策、日志完整性优先和日志不可变原则，确保分布式系统在容错情况下的数据一致性。]]></description><category></category></item><item><title><![CDATA[AI 算力加速指南]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/152117964</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/152117964</guid><author>LiuYQi</author><pubDate>Fri, 26 Sep 2025 10:27:32 +0800</pubDate><description><![CDATA[硬件层面软件优化设计领域办公场景内容创作]]></description><category></category></item><item><title><![CDATA[分布式系统核心概念：CAP、分布式锁与事务优化解析]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/152116540</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/152116540</guid><author>LiuYQi</author><pubDate>Fri, 26 Sep 2025 09:50:16 +0800</pubDate><description><![CDATA[本文深入解析分布式系统核心概念CAP理论及其应用实践。CAP理论指出分布式系统必须在一致性(C)、可用性(A)和分区容错性(P)之间权衡，通常需在CP和AP中选择。文章对比了ZooKeeper(CP)与Eureka(AP)的服务注册中心实现差异，并详细分析了基于Redis(AP)和ZooKeeper(CP)的分布式锁实现方案及适用场景。同时阐述了从ACID到BASE理论的演进，介绍了2PC、TCC等分布式事务实现方案。]]></description><category></category></item><item><title><![CDATA[LeetCode1]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/152076229</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/152076229</guid><author>LiuYQi</author><pubDate>Thu, 25 Sep 2025 10:45:35 +0800</pubDate><description><![CDATA[本文介绍了两个算法问题的解法。组合总和问题采用递归+回溯+剪枝策略，通过排序预处理、选或不选的分支处理，以及剪枝优化来寻找所有和为target的组合。分数转小数问题则需处理符号、整数部分和小数部分，通过长除法计算小数位，并使用哈希表检测循环节，最后按格式拼接结果。两问题都强调了预处理和边界条件的处理，通过系统化的方法提高算法效率。]]></description><category></category></item><item><title><![CDATA[大模型微调实战笔记：从数据集到 Agent 集成的全流程避坑指南]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/152040299</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/152040299</guid><author>LiuYQi</author><pubDate>Wed, 24 Sep 2025 14:52:55 +0800</pubDate><description><![CDATA[《大模型微调实战避坑指南》总结了从数据集设计到全链路工具落地(训练+评估) 最后Agent集成的全流程关键点]]></description><category></category></item><item><title><![CDATA[LeetCode]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/152029322</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/152029322</guid><author>LiuYQi</author><pubDate>Wed, 24 Sep 2025 10:37:06 +0800</pubDate><description><![CDATA[LeetCode解题]]></description><category></category></item><item><title><![CDATA[递归和动态规划]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/151993030</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/151993030</guid><author>LiuYQi</author><pubDate>Tue, 23 Sep 2025 14:13:11 +0800</pubDate><description><![CDATA[递归是一种通过函数调用自身解决子问题的编程思想，将大问题拆分为结构相同的小问题，如阶乘计算。递归函数若仅依赖输入参数可视为纯函数，但存在重复计算问题。动态规划通过存储子问题解（DP数组）避免重复计算，可采用自顶向下或自底向上方法实现。递归数据结构如链表和树，其定义中包含对自身类型的引用，形成自引用结构。]]></description><category></category></item><item><title><![CDATA[LangChain框架 + 大模型]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/151992987</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/151992987</guid><author>LiuYQi</author><pubDate>Tue, 23 Sep 2025 11:03:34 +0800</pubDate><description><![CDATA[LangChain框架通过记忆组件（如对话缓存、摘要记忆）赋予大模型记忆能力，并借助Agent机制实现工具调用。快手采用&quot;垂直领域优先&quot;策略，构建了融合RAG与Agent的智能客服系统，通过多路召回和知识图谱提升回答质量。针对复杂任务，建议将流程工具分层（基础层、组合层、模板层）并封装为专用工具，以优化大模型的推理效率。此外，通过沉淀成功案例为模板工具，可复用经验，减少重复拆解任务的需求。]]></description><category></category></item><item><title><![CDATA[PPO vs GRPO：两大策略优化算法核心机制解析]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/151954824</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/151954824</guid><author>LiuYQi</author><pubDate>Mon, 22 Sep 2025 09:39:56 +0800</pubDate><description><![CDATA[本文对比分析了强化学习中PPO（近端策略优化）和GRPO（组内相对策略优化）两种算法的核心差异。]]></description><category></category></item><item><title><![CDATA[Transformer 核心组件深度解析与 Python 实现]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/151921682</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/151921682</guid><author>LiuYQi</author><pubDate>Mon, 22 Sep 2025 08:59:00 +0800</pubDate><description><![CDATA[Transformer模型通过自注意力机制彻底革新了NLP领域。本文详细剖析了各组件原理与实现，包括位置编码的数学设计、注意力分数计算、FFN结构等关键环节，为理解BERT、GPT等衍生模型奠定了基础。]]></description><category></category></item><item><title><![CDATA[Transformer代码实现思路]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/151920473</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/151920473</guid><author>LiuYQi</author><pubDate>Sun, 21 Sep 2025 09:14:04 +0800</pubDate><description><![CDATA[本文介绍了Transformer模型核心组件的实现思路：1) TokenEmbedding模块将输入序列转换为向量并添加位置编码；2) MultiHeadAttention实现多头自注意力机制，通过QKV矩阵计算Token间关系；3) LayerNorm进行残差连接和归一化；4) FFN前馈网络进行非线性变换；5) Encoder和Decoder通过堆叠上述组件完成编码解码过程。其中Decoder使用编码层输出的KV矩阵实现交叉注意力机制。整个架构通过自注意力、残差连接和归一化等机制实现序列建模。]]></description><category></category></item><item><title><![CDATA[动态规划中的背包问题：0/1 背包与完全背包的核心解析]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/151920170</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/151920170</guid><author>LiuYQi</author><pubDate>Sun, 21 Sep 2025 08:37:58 +0800</pubDate><description><![CDATA[聚焦0/1背包与完全背包的核心辨析：先拆解DP“状态与状态转移”基础逻辑，再直击两者最关键的遍历顺序差异（0/1背包内层倒序防物品重复、完全背包内层正序允重复），搭配Java代码（含错误示例）直观演示，最后用表格清晰对比核心特性，既讲透原理又贴合实际编码。]]></description><category></category></item><item><title><![CDATA[开发中遇到的JAVA小知识]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/148946613</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/148946613</guid><author>LiuYQi</author><pubDate>Fri, 27 Jun 2025 09:58:22 +0800</pubDate><description><![CDATA[java]]></description><category></category></item><item><title><![CDATA[那些在焦虑与希望里打转的日子]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/147877761</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/147877761</guid><author>LiuYQi</author><pubDate>Sun, 11 May 2025 19:45:21 +0800</pubDate><description><![CDATA[实习记录]]></description><category></category></item><item><title><![CDATA[消息队列优化指南：处理堆积与保障消息可靠性]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/147579652</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/147579652</guid><author>LiuYQi</author><pubDate>Mon, 28 Apr 2025 12:09:47 +0800</pubDate><description><![CDATA[本文针对消息队列系统中常见的消息堆积问题和消息丢失风险，提供系统性解决方案。通过分析消息堆积的成因，提出包括扩容、优化、限流等实用方法；同时，结合持久化、确认机制、高可用架构等技术手段，详细阐述如何确保消息不丢失。无论是开发人员还是运维工程师，均可通过本文快速掌握消息队列的优化策略。]]></description><category></category></item><item><title><![CDATA[消息队列堆积如何处理]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/147579559</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/147579559</guid><author>LiuYQi</author><pubDate>Mon, 28 Apr 2025 12:04:25 +0800</pubDate><description><![CDATA[消息队列堆积如何处理]]></description><category></category></item><item><title><![CDATA[虚拟机与Docker与K8s的比较]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/146568077</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/146568077</guid><author>LiuYQi</author><pubDate>Thu, 27 Mar 2025 18:36:51 +0800</pubDate><description><![CDATA[虚拟机与Docker与K8s的比较]]></description><category></category></item><item><title><![CDATA[ES分布式搜索引擎]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/146568024</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/146568024</guid><author>LiuYQi</author><pubDate>Thu, 27 Mar 2025 18:34:30 +0800</pubDate><description><![CDATA[ES分布式搜索引擎的基本架构和简单介绍]]></description><category></category></item><item><title><![CDATA[计算机视觉（CV）技术的优势和挑战]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/144527015</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/144527015</guid><author>LiuYQi</author><pubDate>Tue, 17 Dec 2024 10:16:59 +0800</pubDate><description><![CDATA[计算机视觉是一种利用计算机和算法来模拟人类视觉的技术。它可以从图像或视频中提取信息，并进行分析和理解。]]></description><category></category></item><item><title><![CDATA[JUC并发编程]]></title><link>https://blog.youkuaiyun.com/LiuYQi/article/details/144520072</link><guid>https://blog.youkuaiyun.com/LiuYQi/article/details/144520072</guid><author>LiuYQi</author><pubDate>Tue, 17 Dec 2024 03:30:00 +0800</pubDate><description><![CDATA[volatile和JMM：1. volatile修饰变量的特点、2. 内存屏障、3. volatile特性：有序性、可见性、4. happens-before之volatile变量规则、5. volatile、6. JMM]]></description><category></category></item></channel></rss>