<![CDATA[fjfdg666的博客]]>

<![CDATA[fjfdg666的博客]]>https://blog.youkuaiyun.com/fjfdg666zh-cnhttps://blog.youkuaiyun.com/<![CDATA[从混合 Batch 到 GPU Kernel：读懂 vLLM 与 SGLang 的 Attention 执行]]>https://blog.youkuaiyun.com/fjfdg666/article/details/170245235https://blog.youkuaiyun.com/fjfdg666/article/details/170245235fjfdg666Sat, 07 Nov 2026 17:49:48 +0800<![CDATA[NVIDIA Dynamo Frontend 深度解析：从请求入口到推理集群调度内核]]>https://blog.youkuaiyun.com/fjfdg666/article/details/165121649https://blog.youkuaiyun.com/fjfdg666/article/details/165121649fjfdg666Sat, 12 Sep 2026 16:43:25 +0800<![CDATA[SGLang HiCache 与分离式推理：技术原理及部署实践]]>https://blog.youkuaiyun.com/fjfdg666/article/details/164405182https://blog.youkuaiyun.com/fjfdg666/article/details/164405182fjfdg666Sun, 06 Sep 2026 00:02:11 +0800<![CDATA[vLLM 模型编译：从 torch.compile 到分段 CUDA Graph]]>https://blog.youkuaiyun.com/fjfdg666/article/details/164149539https://blog.youkuaiyun.com/fjfdg666/article/details/164149539fjfdg666Fri, 28 Aug 2026 16:52:42 +0800<![CDATA[vLLM 里的 CUDA Graph：动态推理如何塞进一张“静态图”]]>https://blog.youkuaiyun.com/fjfdg666/article/details/163782670https://blog.youkuaiyun.com/fjfdg666/article/details/163782670fjfdg666Sat, 15 Aug 2026 19:10:28 +0800<![CDATA[Megatron Core 并行训练主线：不看代码版]]>https://blog.youkuaiyun.com/fjfdg666/article/details/161367823https://blog.youkuaiyun.com/fjfdg666/article/details/161367823fjfdg666Sun, 24 May 2026 18:25:33 +0800<![CDATA[从一次 Hermes Agent 会话看 System Prompt、Tools 和 Skills]]>https://blog.youkuaiyun.com/fjfdg666/article/details/161129544https://blog.youkuaiyun.com/fjfdg666/article/details/161129544fjfdg666Sat, 16 May 2026 05:10:04 +0800<![CDATA[介绍下近期的开源大模型]]>https://blog.youkuaiyun.com/fjfdg666/article/details/157840282https://blog.youkuaiyun.com/fjfdg666/article/details/157840282fjfdg666Sat, 07 Feb 2026 12:41:36 +0800<![CDATA[Agent记忆管理方法概述]]>https://blog.youkuaiyun.com/fjfdg666/article/details/157695638https://blog.youkuaiyun.com/fjfdg666/article/details/157695638fjfdg666Tue, 03 Feb 2026 17:35:20 +0800<![CDATA[理一下OpenRLHF的Ray资源调度和PPO训练流程]]>https://blog.youkuaiyun.com/fjfdg666/article/details/157060999https://blog.youkuaiyun.com/fjfdg666/article/details/157060999fjfdg666Sat, 17 Jan 2026 11:08:34 +0800<![CDATA[Online RLHF-PPO、GRPO、DAPO等方法]]>https://blog.youkuaiyun.com/fjfdg666/article/details/156947935https://blog.youkuaiyun.com/fjfdg666/article/details/156947935fjfdg666Wed, 14 Jan 2026 16:09:53 +0800<![CDATA[Offline RLHF-DPO/KTO/ORPO/SPO等偏好学习方法]]>https://blog.youkuaiyun.com/fjfdg666/article/details/156910190https://blog.youkuaiyun.com/fjfdg666/article/details/156910190fjfdg666Tue, 13 Jan 2026 18:00:50 +0800<![CDATA[从Diffusion到VLA pi0(π0)]]>https://blog.youkuaiyun.com/fjfdg666/article/details/156239801https://blog.youkuaiyun.com/fjfdg666/article/details/156239801fjfdg666Wed, 24 Dec 2025 18:59:05 +0800<![CDATA[语音大模型-Audio Tokenizer的进展]]>https://blog.youkuaiyun.com/fjfdg666/article/details/156055816https://blog.youkuaiyun.com/fjfdg666/article/details/156055816fjfdg666Thu, 18 Dec 2025 15:03:06 +0800<![CDATA[Qwen3-Omni技术详解]]>https://blog.youkuaiyun.com/fjfdg666/article/details/155137540https://blog.youkuaiyun.com/fjfdg666/article/details/155137540fjfdg666Sat, 22 Nov 2025 18:46:17 +0800<![CDATA[PTX指令集基础以及warp级矩阵乘累加指令介绍]]>https://blog.youkuaiyun.com/fjfdg666/article/details/149722392https://blog.youkuaiyun.com/fjfdg666/article/details/149722392fjfdg666Mon, 28 Jul 2025 17:58:31 +0800<![CDATA[CUDA C++核心库（CCCL）]]>https://blog.youkuaiyun.com/fjfdg666/article/details/149544366https://blog.youkuaiyun.com/fjfdg666/article/details/149544366fjfdg666Tue, 22 Jul 2025 18:44:45 +0800<![CDATA[全网最详llm.c源码解读--forward、backward、AdamW等CUDA Kernel详解]]>https://blog.youkuaiyun.com/fjfdg666/article/details/149516797https://blog.youkuaiyun.com/fjfdg666/article/details/149516797fjfdg666Mon, 21 Jul 2025 22:16:39 +0800<![CDATA[NCCL的基本使用和常用通信算法源码分析]]>https://blog.youkuaiyun.com/fjfdg666/article/details/149092802https://blog.youkuaiyun.com/fjfdg666/article/details/149092802fjfdg666Thu, 03 Jul 2025 14:07:51 +0800<![CDATA[Liger Kernel中的部分优化kernel详解]]>https://blog.youkuaiyun.com/fjfdg666/article/details/149045899https://blog.youkuaiyun.com/fjfdg666/article/details/149045899fjfdg666Tue, 01 Jul 2025 14:18:33 +0800