{"id":76831,"date":"2025-12-03T16:42:18","date_gmt":"2025-12-03T08:42:18","guid":{"rendered":"https:\/\/aicats.wiki\/?p=76831"},"modified":"2025-12-03T16:42:45","modified_gmt":"2025-12-03T08:42:45","slug":"rmsnorm%e5%9c%a8transformer%e6%a8%a1%e5%9e%8b%e4%b8%ad%e7%9a%84%e4%bd%9c%e7%94%a8%e8%a7%a3%e6%9e%90%e5%8f%8a%e5%af%a6%e7%94%a8%e6%8c%87%e5%8d%97","status":"publish","type":"post","link":"https:\/\/aicats.wiki\/tw\/2025\/12\/03\/76831-html","title":{"rendered":"rmsnorm\u5728Transformer\u6a21\u578b\u4e2d\u7684\u4f5c\u7528\u89e3\u6790\u53ca\u5be6\u7528\u6307\u5357"},"content":{"rendered":"\n<p><strong>rmsnorm\uff08Root Mean Square Layer Normalization\uff09<\/strong>\u4f5c\u4e3a\u65b0\u4e00\u4ee3\u5f52\u4e00\u5316\u65b9\u6cd5\uff0c\u5df2\u5e7f\u6cdb\u5e94\u7528\u4e8e\u4e3b\u6d41Transformer\u5927\u6a21\u578b\uff08\u5982LLaMA\u3001DeepSeek-V3\u7b49\uff09\uff0c\u66ff\u4ee3\u4f20\u7edfLayerNorm\u5b9e\u73b0<a href=\"https:\/\/aicats.wiki\/2025\/11\/11\/67322.html\" title=\"AI\u6a21\u578b\u8bad\u7ec3\u5fc5\u5907\u5de5\u5177\u63a8\u8350\uff1a2025\u5e7410\u6b3e\u63d0\u5347\u6548\u7387\u7684\u5b9e\u7528\u5e73\u53f0\u5927\u76d8\u70b9\">\u9ad8\u6548\u63a8\u7406\u548c\u8bad\u7ec3<\/a>\u3002\u672c\u6587\u5c06\u901a\u8fc7\u6280\u672f\u539f\u7406\u3001\u6027\u80fd\u5bf9\u6bd4\u3001\u4e1a\u754c\u6848\u4f8b\u3001\u96c6\u6210\u4ee3\u7801\u7b49\u65b9\u9762\u7cfb\u7edf\u89e3\u6790<strong>rmsnorm\u5728Transformer\u4e2d\u7684\u4f18\u52bf\u4e0e\u5b9e\u9645\u90e8\u7f72\u65b9\u6cd5<\/strong>\uff0c\u52a9\u529b\u7814\u53d1\u8005\u63d0\u5347\u6a21\u578b\u6548\u7387\u4e0e\u5de5\u4e1a\u4ef7\u503c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img fetchpriority=\"high\" decoding=\"async\" width=\"1927\" height=\"1009\" src=\"https:\/\/aicats.wiki\/wp-content\/uploads\/2025\/12\/image-36.png\" alt=\"rmsnorm\u5728Transformer\u6a21\u578b\u4e2d\u7684\u4f5c\u7528\u89e3\u6790\u53ca\u5b9e\u7528\u6307\u5357\" class=\"wp-image-77951\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\u7406\u8bba\u57fa\u7840\u4e0e\u6280\u672f\u539f\u7406<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e3a\u4ec0\u4e48\u8981\u5f52\u4e00\u5316\uff1fTransformer\u5f52\u4e00\u5316\u9700\u6c42\u6eaf\u6e90<\/h3>\n\n\n\n<p>Transformer\u7ed3\u6784\u81ea2017\u5e74\u53d1\u5e03\u4ee5\u6765\uff0c\u5f52\u4e00\u5316\u5bf9\u6536\u655b\u901f\u5ea6\u3001\u7a33\u5b9a\u6027\u53ca\u5927\u89c4\u6a21\u53c2\u6570\u4e0b\u8868\u73b0\u8d77\u5230<strong>\u5173\u952e\u5f71\u54cd<\/strong>\u3002<strong>LayerNorm<\/strong>\u662f\u6700\u65e9\u4e14\u5e38\u7528\u7684\u65b9\u6cd5\uff0c\u4f46\u5176\u8ba1\u7b97\u91cf\u548c\u5185\u5b58\u6d88\u8017\u5728\u5927\u6a21\u578b\u6269\u5c55\u65f6\u663e\u5f97\u7a81\u51fa\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">rmsnorm\u6570\u5b66\u516c\u5f0f\u4e0e\u5de5\u4f5c\u539f\u7406<\/h3>\n\n\n\n<p><strong>rmsnorm<\/strong>\u7531Shen\u7b49\u4eba\u63d0\u51fa\uff0c\u4ec5\u5229\u7528\u8f93\u5165\u7279\u5f81\u7684\u5747\u65b9\u6839\uff08RMS\uff09\u8fdb\u884c\u7f29\u653e\uff0c\u65e0\u9700\u4e2d\u5fc3\u5316\u3002\u6838\u5fc3\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n\n\n\n<!--wp-compress-html--><!--wp-compress-html no compression-->\n<pre class=\"wp-block-preformatted\">RMSNorm(x) = \u03b3 \u2299 (x \/ sqrt(mean(x^2) + \u03b5))\n<\/pre>\n<!--wp-compress-html no compression--><!--wp-compress-html-->\n\n\n\n<p><strong>\u4ec5\u4f9d\u9760RMS\uff0c\u65e0\u9700\u51cf\u5747\u503c\uff0c\u7701\u53bb\u8ba1\u7b97\u6b65\u9aa4\uff0c\u63d0\u5347\u6267\u884c\u6548\u7387<\/strong>\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"1922\" height=\"1021\" src=\"https:\/\/aicats.wiki\/wp-content\/uploads\/2025\/12\/image-37.png\" alt=\"Qwen2 RMSNorm\u96c6\u6210\" class=\"wp-image-77961\"\/><figcaption class=\"wp-element-caption\">\u5716\uff0f<a href=\"https:\/\/github.com\/QwenLM\/Qwen2\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >Qwen2 RMSNorm\u96c6\u6210<\/a><\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>\u65b9\u6cd5\u5bf9\u6bd4<\/th><th>\u8ba1\u7b97\u5185\u5bb9<\/th><th>\u4e3b\u8981\u4f18\u70b9<\/th><th>\u4e3b\u8981\u7f3a\u70b9<\/th><th>\u9002\u7528\u573a\u666f<\/th><\/tr><tr><td>LayerNorm<\/td><td>\u51cf\u5747\u503c+\u9664\u6807\u51c6\u5dee<\/td><td>\u7a33\u5b9a\uff0c\u9a8c\u8bc1\u5e7f\u6cdb<\/td><td>\u8ba1\u7b97\u91cf\u5927<\/td><td>RNN\/\u5e8f\u5217<a class=\"external\" href=\"https:\/\/aicats.wiki\/tw\/tag\/%e5%bb%ba%e6%a8%a1\" title=\"\u67e5\u770b\u8207 \u5efa\u6a21 \u76f8\u95dc\u7684\u6587\u7ae0\" target=\"_blank\">\u5efa\u6a21<\/a><\/td><\/tr><tr><td><strong>rmsnorm<\/strong><\/td><td>\u4ec5\u9664\u5747\u65b9\u6839<\/td><td>\u8ba1\u7b97\u5feb\uff0c\u8d44\u6e90\u4f4e<\/td><td>\u6f02\u79fb\u5bb9\u5fcd\u7565\u4f4e<\/td><td>Transformer\u5927\u6a21\u578b<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">rmsnorm\u5728Transformer\u4e2d\u7684\u4ef7\u503c\u4e0e\u4f5c\u7528<\/h2>\n\n\n\n<p><strong>1. \u7b80\u5316\u8ba1\u7b97\uff0c\u5927\u5e45\u63d0\u5347\u6548\u7387\uff1a<\/strong>\u7701\u53bb\u51cf\u5747\u503c\uff0c\u53ef\u964d\u4f4e15-30%\u7684\u8fd0\u7b97\u91cf\uff0c\u663e\u5b58\u8bbf\u95ee\u51cf\u5c11\uff0c\u5c24\u5176\u5728\u9ad8\u7ef4\u4e0e\u5927\u6279\u91cf\u8bad\u7ec3\u573a\u666f\u4e0b\u4f18\u52bf\u663e\u8457\u3002<\/p>\n\n\n\n<p><strong>2. \u68af\u5ea6\u4e0e\u8bad\u7ec3\u66f2\u7ebf\u66f4\u5e73\u6ed1\uff1a<\/strong>RMS\u5f52\u4e00\u5316\u80fd\u4f7f\u68af\u5ea6\u65b9\u5dee\u8f83LayerNorm\u4f4e\u7ea620%\uff0c\u663e\u8457\u964d\u4f4e\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8\u98ce\u9669\u3002<\/p>\n\n\n\n<p><strong>3. \u53c2\u6570\u66f4\u9ad8\u6548\uff0c\u8868\u8fbe\u80fd\u529b\u5f3a\uff1a<\/strong>\u65e0\u504f\u7f6e\uff0c\u53c2\u6570\u91cf\u51cf\u534a\uff0c\u901a\u9053\u65b9\u5411\u4fe1\u606f\u4fdd\u7559\u6709\u52a9\u4e8e\u66f4\u6df1\u6a21\u578b\u8868\u73b0\u3002<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>\u5f52\u4e00\u5316\u65b9\u6cd5<\/th><th>\u53bb\u5747\u503c<\/th><th>\u5f52\u4e00\u5316\u5bf9\u8c61<\/th><th>\u7f29\u653e\u53c2\u6570<\/th><th>\u504f\u79fb\u53c2\u6570<\/th><th>\u53c2\u6570\u91cf<\/th><th>\u5178\u578b\u573a\u666f<\/th><\/tr><tr><td>LayerNorm<\/td><td>\u6709<\/td><td>\u5c42\u5185<\/td><td>\u03b3<\/td><td>\u03b2<\/td><td>2n<\/td><td>RNN\/\u4f20\u7edf\u5e8f\u5217<\/td><\/tr><tr><td><strong>rmsnorm<\/strong><\/td><td>\u65e0<\/td><td>\u5c42\u5185<\/td><td>\u03b3<\/td><td>\u65e0<\/td><td>1n<\/td><td>Transformer\/LLM<\/td><\/tr><tr><td>BatchNorm<\/td><td>\u6709<\/td><td>\u6279\u95f4<\/td><td>\u03b3<\/td><td>\u03b2<\/td><td>2n<\/td><td>CV\/\u8bed\u97f3<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\u771f\u5b9e\u6848\u4f8b\uff1armsnorm\u5728\u5de5\u4e1a\u7ea7\u5927\u6a21\u578b\u7684\u5e94\u7528<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">LLaMA\u7cfb\u5217\u5927\u6a21\u578b\uff08Meta\uff09<\/h3>\n\n\n\n<p>Meta\u7684LLaMA\u7cfb\u5217\u5168\u90e8\u539f\u751f\u91c7\u7528<strong>rmsnorm<\/strong>\uff0c\u5728WikiText-103\u7b49\u4efb\u52a1\u4e0a\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8bad\u7ec3\u65f6\u957f\u7f29\u77ed18%<\/strong><\/li>\n\n\n\n<li><strong>\u4e0eLayerNorm\u6027\u80fd\u57fa\u672c\u4e00\u81f4\uff0c\u635f\u5931\u4f4e\u4e8e0.1%<\/strong><\/li>\n\n\n\n<li><strong>\u63a8\u7406\u66f4\u5feb\uff0c\u8d44\u6e90\u6d88\u8017\u66f4\u4f18<\/strong><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"1920\" height=\"1017\" src=\"https:\/\/aicats.wiki\/wp-content\/uploads\/2025\/12\/image-36-6.jpg\" alt=\"Meta LLaMA\u5927\u6a21\u578b\u5b98\u7f51\u622a\u56fe\" class=\"wp-image-77948\"\/><figcaption class=\"wp-element-caption\">\u5716\uff0f<a href=\"https:\/\/www.llama-cn.com\/\" title=\"\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >Meta LLaMA\u5927\u6a21\u578b\u5b98\u7f51\u622a\u56fe<\/a><\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">DeepSeek-V3\u548cQwen\u7b49\u65b0\u4e00\u4ee3\u5927\u6a21\u578b<\/h3>\n\n\n\n<p>2024\u5e74\u540e\u4e3b\u6d41\u5927\u6a21\u578b\u5747\u9ed8\u8ba4\u7528rmsnorm\u3002\u5f00\u53d1\u8005\u4ec5\u9700\u66ff\u6362\u5f52\u4e00\u5316\u5c42\u5373\u53ef\u9ad8\u6548\u5347\u7ea7\uff0c\u590d\u6742\u5ea6\u548c\u98ce\u9669\u6781\u4f4e\u3002<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>\u6a21\u578b\u540d\u79f0<\/th><th>\u5f52\u4e00\u5316\u5c42<\/th><th>\u8bad\u7ec3\u901f\u5ea6\u63d0\u5347<\/th><th>PPL\u5dee\u5f02<\/th><th>\u63a8\u7406\u901f\u5ea6<\/th><th>\u8d44\u6e90\u6d88\u8017<\/th><\/tr><tr><td>GPT-2\/LN<\/td><td>LayerNorm<\/td><td>\u57fa\u7ebf<\/td><td>\u57fa\u7ebf<\/td><td>\u57fa\u7ebf<\/td><td>\u57fa\u7ebf<\/td><\/tr><tr><td>GPT-2\/rmsnorm<\/td><td>rmsnorm<\/td><td>+18%<\/td><td>&lt;0.1%<\/td><td>+15%<\/td><td>\u964d\u4f4e15%<\/td><\/tr><tr><td>LLaMA<\/td><td>rmsnorm<\/td><td>\u539f\u751f<\/td><td>\u6781\u4f18<\/td><td>\u6781\u4f18<\/td><td>\u6781\u4f18<\/td><\/tr><tr><td>DeepSeek-V2<\/td><td>rmsnorm<\/td><td>+20%<\/td><td>&lt;0.2%<\/td><td>+15%<\/td><td>\u4f18\u5316<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/aicats.wiki\/wp-content\/uploads\/2025\/12\/my_prefix_1764574163.jpg\" alt=\"DeepSeek\u4ee3\u7801\u4ed3\u5e93\u622a\u56fe\" class=\"wp-image-51824\"\/><figcaption class=\"wp-element-caption\">\u5716\uff0f<a href=\"https:\/\/github.com\/deepseek-ai\/DeepSeek-V2\" title=\"\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >DeepSeek\u4ee3\u7801\u4ed3\u5e93\u622a\u56fe<\/a><\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\u5de5\u5177\u4ea7\u54c1\u9009\u578b\u53ca\u96c6\u6210\u6307\u5357<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e3b\u6d41<a class=\"external\" href=\"https:\/\/aicats.wiki\/tw\/tag\/ai\" title=\"\u67e5\u770b\u8207 AI \u76f8\u95dc\u7684\u6587\u7ae0\" target=\"_blank\">AI<\/a>\u6846\u67b6\u4e0e\u77e5\u540d\u5de5\u5177\u4ea7\u54c1\u652f\u6301<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/huggingface.co\/docs\/transformers\/main\/en\/model_doc\/llama\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >HuggingFace Transformers<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/QwenLM\/Qwen2\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >Qwen2 RMSNorm\u96c6\u6210<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/deepseek-ai\/DeepSeek-V2\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >DeepSeek-V2\u4ee3\u7801\u53c2\u8003<\/a><\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">\u4ee3\u7801\u96c6\u6210\uff08PyTorch\u8303\u4f8b\uff09<\/h3>\n\n\n\n<!--wp-compress-html--><!--wp-compress-html no compression-->\n<pre class=\"wp-block-preformatted\">import torch\nimport torch.nn as nn\n\nclass RMSNorm(nn.Module):\n    def __init__(self, dim, eps=1e-6):\n        super().__init__()\n        self.eps = eps\n        self.weight = nn.Parameter(torch.ones(dim))\n\n    def forward(self, x):\n        norm = x.norm(2, dim=-1, keepdim=True) \/ (x.shape[-1] ** 0.5)\n        return self.weight * x \/ (norm + self.eps)\n<\/pre>\n<!--wp-compress-html no compression--><!--wp-compress-html-->\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1920\" height=\"1018\" src=\"https:\/\/aicats.wiki\/wp-content\/uploads\/2025\/12\/image-37.jpg\" alt=\"HuggingFace Transformers\" class=\"wp-image-77954\"\/><figcaption class=\"wp-element-caption\">\u5716\uff0f<a href=\"https:\/\/huggingface.co\/docs\/transformers\/main\/en\/model_doc\/llama\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >HuggingFace Transformers<\/a><\/figcaption><\/figure>\n\n\n\n<p><strong>\u66ff\u6362\u53ea\u9700\u51e0\u884c\u4ee3\u7801\uff0c\u65e0\u9700\u5927\u5e45\u91cd\u6784\u3002\u5efa\u8bae\u521d\u671f\u9002\u5f53\u8c03\u4f4e\u5b66\u4e60\u7387\u3002<\/strong><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u5178\u578b\u843d\u5730\u6b65\u9aa4<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u786e\u5b9a\u6a21\u578b\u5f52\u4e00\u5316\u5c42\u4f4d\u7f6e<\/li>\n\n\n\n<li>\u6279\u91cf\u66ff\u6362\u4e3armsnorm<\/li>\n\n\n\n<li>\u6d4b\u8bd5\u8bad\u7ec3\/\u63a8\u7406\u786e\u4fdd\u65e0bug<\/li>\n\n\n\n<li>\u5168\u9762\u8bad\u7ec3\u5bf9\u6bd4\u6536\u655b\u4e0e\u6027\u80fd<\/li>\n\n\n\n<li>\u5728\u5206\u5e03\u5f0f\/\u63a8\u7406\u73af\u5883\u4e0b\u505a\u5927\u6279\u91cf\u6d4b\u8bd5<\/li>\n<\/ol>\n\n\n\n<h3 class=\"wp-block-heading\">\u90e8\u7f72\u4e0e\u4f18\u5316\u6ce8\u610f\u4e8b\u9879<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5c0f\u6279\u91cf\u540c\u6837\u9002\u7528<\/li>\n\n\n\n<li>\u5206\u5e03\u6f02\u79fb\u654f\u611f\u573a\u666f\u53ef\u52a0\u5fae\u5c0f\u504f\u79fb\u53c2\u6570\u8c03\u4f18<\/li>\n\n\n\n<li>\u7ed3\u5408\u6700\u65b0\u786c\u4ef6\u548c\u63a8\u7406\u5e93\u91ca\u653e\u6f5c\u529b<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">\u5178\u578b\u95ee\u9898\u7b54\u7591\u4e0e\u5b9e\u7528\u6e05\u5355<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u54ea\u4e9b\u573a\u666f\u6700\u9002\u5408\u76f4\u63a5\u7528rmsnorm\uff1f<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5927\u578bTransformer\u9884\u8bad\u7ec3\u4e0e\u5fae\u8c03\u6a21\u578b<\/li>\n\n\n\n<li>\u9ad8\u541e\u5410\u63a8\u7406API\u3001\u7ebf\u4e0a\u670d\u52a1<\/li>\n\n\n\n<li>\u8d85\u5927\u6279\u91cfGPU\/TPU\u8bad\u7ec3\u4efb\u52a1<\/li>\n\n\n\n<li>\u5de5\u4e1a\u573a\u666f\u9700\u9ad8\u8fc1\u79fb\u548c\u90e8\u7f72\u6548\u7387\u8005<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e3b\u6d41\u5de5\u5177\u4e0e\u6846\u67b6\u652f\u6301\u6e05\u5355<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>\u5de5\u5177\/\u6846\u67b6<\/th><th>rmsnorm\u652f\u6301<\/th><th>\u8fc1\u79fb\u96be\u5ea6<\/th><th>\u63a8\u8350\u573a\u666f<\/th><\/tr><tr><td>PyTorch<\/td><td>\u539f\u751f\u96c6\u6210<\/td><td>\u6613<\/td><td>\u7814\u53d1\/\u5b9e\u9a8c<\/td><\/tr><tr><td>TensorFlow<\/td><td>\u793e\u533a\u5b9e\u73b0<\/td><td>\u6613<\/td><td>\u7814\u53d1\/\u90e8\u7f72<\/td><\/tr><tr><td>HuggingFace<\/td><td>\u76f4\u63a5\u9009\u62e9<\/td><td>\u6613<\/td><td>LLM\u5fae\u8c03<\/td><\/tr><tr><td>ONNXRuntime\u7b49<\/td><td>\u9ad8\u5ea6\u4f18\u5316<\/td><td>\u6613<\/td><td>\u9ad8\u6548\u63a8\u7406<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/huggingface.co\/docs\/transformers\/\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >HuggingFace Transformers<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/deepseek-ai\/DeepSeek-V2\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >DeepSeek-V2\u4ee3\u7801<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.llama-cn.com\/\" title=\"\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >Meta LLaMA3<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/QwenLM\/Qwen2\" target=\"_blank\"  rel=\"nofollow noopener\"  class=\"external\" >Qwen2\u5b98\u65b9<\/a><\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">\u7ed3\u8bed<\/h2>\n\n\n\n<p><strong>rmsnorm<\/strong>\u5df2\u6210\u4e3a\u5927\u6a21\u578b\u57fa\u7840\u8bbe\u65bd\u4f18\u5316\u7684\u6807\u5fd7\u6027\u5de5\u5177\uff0c\u4e3aAI\u4ea7\u4e1a\u5e26\u6765\u65b0\u7684\u80fd\u6548\u6781\u9650\u3002\u5efa\u8bae\u5f00\u53d1\u8005\u4f18\u5148\u9009\u7528rmsnorm\u5f52\u4e00\u5316\uff0c<strong>\u6fc0\u6d3bTransformer\u5728\u8d85\u5927\u89c4\u6a21\u4e0e\u591a\u6a21\u6001\u573a\u666f\u4e0b\u7684\u5b9e\u9645\u4ef7\u503c<\/strong>\uff0c\u4ee5\u6280\u672f\u521b\u65b0\u8d4b\u80fd\u667a\u80fd\u672a\u6765\uff01<\/p>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>rmsnorm\uff08Root Mean Square Layer Normalization\uff09\u4f5c\u4e3a\u65b0\u4e00\u4ee3\u5f52\u4e00\u5316\u65b9\u6cd5 [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_crsspst_to_aicatswiki":true,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[305],"tags":[247,268,1455],"content_visibility":[262],"class_list":["post-76831","post","type-post","status-publish","format-standard","hentry","category-ai-tools-platforms","tag-ai"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/posts\/76831","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/comments?post=76831"}],"version-history":[{"count":1,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/posts\/76831\/revisions"}],"predecessor-version":[{"id":77964,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/posts\/76831\/revisions\/77964"}],"wp:attachment":[{"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/media?parent=76831"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/categories?post=76831"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/tags?post=76831"},{"taxonomy":"content_visibility","embeddable":true,"href":"https:\/\/aicats.wiki\/tw\/wp-json\/wp\/v2\/content_visibility?post=76831"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}