<![CDATA[一名热爱技术、专注实践的工程师，拥有一项实用新型专利及一项软件著作权。曾获国家级奖项7项、省级奖项30余项。专注网络技术与 AI Infra 领域，持续探索技术创新与工程实践，记录学习成长，分享技术经验，在探索中突破，在实践中精进。]]>https://blog.youkuaiyun.com/2201_75791683zh-cnhttps://blog.youkuaiyun.com/<![CDATA[GPU 之间如何高效搬运数据？深入理解 NCCL 与六大集合通信原语]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170503700https://blog.youkuaiyun.com/2201_75791683/article/details/1705037002201_75791683Wed, 11 Nov 2026 11:06:07 +0800<![CDATA[DeepSpeed + Megatron-LM：大模型分布式训练的协同架构]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170500961https://blog.youkuaiyun.com/2201_75791683/article/details/1705009612201_75791683Wed, 11 Nov 2026 10:53:11 +0800<![CDATA[深入理解 3D 并行——TP、PP、DP 如何协同，让上千张 GPU 高效训练大模型？]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170499719https://blog.youkuaiyun.com/2201_75791683/article/details/1704997192201_75791683Wed, 11 Nov 2026 10:32:00 +0800<![CDATA[流水线并行（PP）——把模型按层切开，串成一条高效的计算流水线]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170420910https://blog.youkuaiyun.com/2201_75791683/article/details/1704209102201_75791683Tue, 10 Nov 2026 11:24:09 +0800<![CDATA[张量并行（TP）深度解析——如何将一次矩阵运算拆分到多张 GPU 上？]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170419549https://blog.youkuaiyun.com/2201_75791683/article/details/1704195492201_75791683Tue, 10 Nov 2026 10:59:42 +0800<![CDATA[ZeRO 深度解析——从 128GB 到 16GB，如何打破大模型训练的显存墙？]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170373046https://blog.youkuaiyun.com/2201_75791683/article/details/1703730462201_75791683Tue, 10 Nov 2026 00:01:15 +0800<![CDATA[从单卡瓶颈到多卡协同，揭开大模型分布式训练的底层逻辑]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170295960https://blog.youkuaiyun.com/2201_75791683/article/details/1702959602201_75791683Mon, 09 Nov 2026 07:20:10 +0800<![CDATA[深入理解 DDP——多卡训练如何通过梯度同步实现高效并行？]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170296485https://blog.youkuaiyun.com/2201_75791683/article/details/1702964852201_75791683Mon, 09 Nov 2026 07:19:51 +0800<![CDATA[深入理解 WLAN 无线局域网原理、CAPWAP 协议与配置实现]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170247858https://blog.youkuaiyun.com/2201_75791683/article/details/1702478582201_75791683Sun, 08 Nov 2026 15:38:12 +0800<![CDATA[PPP协议原理、状态机、协商机制与认证技术]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170248032https://blog.youkuaiyun.com/2201_75791683/article/details/1702480322201_75791683Sun, 08 Nov 2026 15:37:38 +0800<![CDATA[ACL 访问控制列表原理、匹配机制与配置实践]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170197518https://blog.youkuaiyun.com/2201_75791683/article/details/1701975182201_75791683Sat, 07 Nov 2026 10:34:18 +0800<![CDATA[NAT 网络地址转换技术原理与配置实战]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170198095https://blog.youkuaiyun.com/2201_75791683/article/details/1701980952201_75791683Sat, 07 Nov 2026 10:33:53 +0800<![CDATA[深入理解 Leaf-Spine：从网络拓扑到 AI 数据中心 Fabric]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170050121https://blog.youkuaiyun.com/2201_75791683/article/details/1700501212201_75791683Fri, 06 Nov 2026 11:32:38 +0800<![CDATA[模型蒸馏、训练与推理、CANN 架构及 NVIDIA GPU 集群设计]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170189914https://blog.youkuaiyun.com/2201_75791683/article/details/1701899142201_75791683Fri, 06 Nov 2026 11:32:22 +0800<![CDATA[深入理解 DCGM，如何从 GPU 监控走向性能分析与故障定位？]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170049595https://blog.youkuaiyun.com/2201_75791683/article/details/1700495952201_75791683Thu, 05 Nov 2026 09:43:03 +0800<![CDATA[Token、模型参数量与模型量化技术详解]]>https://blog.youkuaiyun.com/2201_75791683/article/details/170049261https://blog.youkuaiyun.com/2201_75791683/article/details/1700492612201_75791683Thu, 05 Nov 2026 09:42:29 +0800<![CDATA[链路聚合原理与配置详解]]>https://blog.youkuaiyun.com/2201_75791683/article/details/169949226https://blog.youkuaiyun.com/2201_75791683/article/details/1699492262201_75791683Wed, 04 Nov 2026 00:47:19 +0800<![CDATA[GPU 监控体系与性能分析——从 nvidia-smi 到 DCGM]]>https://blog.youkuaiyun.com/2201_75791683/article/details/169949434https://blog.youkuaiyun.com/2201_75791683/article/details/1699494342201_75791683Wed, 04 Nov 2026 00:46:49 +0800<![CDATA[AI 算力基础学习笔记：GPU 多卡互联与集群通信原理]]>https://blog.youkuaiyun.com/2201_75791683/article/details/169936065https://blog.youkuaiyun.com/2201_75791683/article/details/1699360652201_75791683Tue, 03 Nov 2026 12:05:46 +0800<![CDATA[GPU 架构深度解析：显存、带宽与算力，如何决定大模型性能？]]>https://blog.youkuaiyun.com/2201_75791683/article/details/169935278https://blog.youkuaiyun.com/2201_75791683/article/details/1699352782201_75791683Tue, 03 Nov 2026 11:52:24 +0800