<![CDATA[ogzhen的博客]]>https://blog.csdn.net/ogzhenzh-cnhttps://blog.csdn.net/<![CDATA[如何计算大模型占用显存大小]]>https://blog.csdn.net/ogzhen/article/details/168345558https://blog.csdn.net/ogzhen/article/details/168345558ogzhenSun, 18 Oct 2026 22:08:22 +0800<![CDATA[fp32,fp16,bf16的区别]]>https://blog.csdn.net/ogzhen/article/details/168144839https://blog.csdn.net/ogzhen/article/details/168144839ogzhenSat, 17 Oct 2026 11:21:13 +0800<![CDATA[ROPE外推：从PI、NTK-Aware到YaRN一统江湖]]>https://blog.csdn.net/ogzhen/article/details/166338813https://blog.csdn.net/ogzhen/article/details/166338813ogzhenSun, 04 Oct 2026 15:38:44 +0800<![CDATA[一文讲透ROPE]]>https://blog.csdn.net/ogzhen/article/details/164715689https://blog.csdn.net/ogzhen/article/details/164715689ogzhenSat, 12 Sep 2026 19:53:01 +0800<![CDATA[一次讲透MHA、GQA、MQA、MLA]]>https://blog.csdn.net/ogzhen/article/details/164292735https://blog.csdn.net/ogzhen/article/details/164292735ogzhenWed, 02 Sep 2026 11:31:38 +0800<![CDATA[Weight Tying共享权重]]>https://blog.csdn.net/ogzhen/article/details/164211827https://blog.csdn.net/ogzhen/article/details/164211827ogzhenMon, 31 Aug 2026 14:50:43 +0800<![CDATA[KV Cache和因果掩码]]>https://blog.csdn.net/ogzhen/article/details/163613198https://blog.csdn.net/ogzhen/article/details/163613198ogzhenSun, 09 Aug 2026 17:38:32 +0800<![CDATA[正余弦位置编码详解]]>https://blog.csdn.net/ogzhen/article/details/162735477https://blog.csdn.net/ogzhen/article/details/162735477ogzhenSun, 12 Jul 2026 19:55:04 +0800<![CDATA[目标检测分类部分损失函数：BCE → Focal Loss → VFL → MAL 的演进]]>https://blog.csdn.net/ogzhen/article/details/162497147https://blog.csdn.net/ogzhen/article/details/162497147ogzhenWed, 01 Jul 2026 22:33:22 +0800<![CDATA[DETR（detection transformer）中的匈牙利匹配算法]]>https://blog.csdn.net/ogzhen/article/details/162398191https://blog.csdn.net/ogzhen/article/details/162398191ogzhenTue, 30 Jun 2026 11:42:31 +0800<![CDATA[为什么交叉熵可以完美适配分类任务？]]>https://blog.csdn.net/ogzhen/article/details/155826386https://blog.csdn.net/ogzhen/article/details/155826386ogzhenThu, 11 Dec 2025 20:36:52 +0800<![CDATA[softmax，交叉熵对logits的梯度计算]]>https://blog.csdn.net/ogzhen/article/details/155788179https://blog.csdn.net/ogzhen/article/details/155788179ogzhenThu, 11 Dec 2025 10:22:41 +0800<![CDATA[pythorch显卡利用率过低的问题]]>https://blog.csdn.net/ogzhen/article/details/103977490https://blog.csdn.net/ogzhen/article/details/103977490ogzhenTue, 14 Jan 2020 17:58:45 +0800<![CDATA[关于信息熵的解释]]>https://blog.csdn.net/ogzhen/article/details/101020045https://blog.csdn.net/ogzhen/article/details/101020045ogzhenThu, 19 Sep 2019 13:34:53 +0800<![CDATA[感知哈希算法（pHash算法+python代码实现）]]>https://blog.csdn.net/ogzhen/article/details/98595552https://blog.csdn.net/ogzhen/article/details/98595552ogzhenTue, 06 Aug 2019 10:51:19 +0800<![CDATA[GIOU：对IOU的优化]]>https://blog.csdn.net/ogzhen/article/details/97915335https://blog.csdn.net/ogzhen/article/details/97915335ogzhenWed, 31 Jul 2019 14:27:23 +0800