{"id":121101,"date":"2026-11-02T03:17:18","date_gmt":"2026-11-01T19:17:18","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/121101.html"},"modified":"2026-11-02T03:17:18","modified_gmt":"2026-11-01T19:17:18","slug":"bert%e8%87%aa%e7%84%b6%e8%af%ad%e8%a8%80%e5%a4%84%e7%90%86%e6%a1%86%e6%9e%b6%e6%b7%b1%e5%ba%a6%e8%a7%a3%e6%9e%90%ef%bc%9a%e4%bb%8etransformer%e5%88%b0%e5%8f%8c%e5%90%91%e9%a2%84%e8%ae%ad%e7%bb%83","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/121101.html","title":{"rendered":"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b"},"content":{"rendered":"<\/p>\n<h4>\u6587\u7ae0\u76ee\u5f55<\/h4>\n<ul>\n<li>\u4e00\u3001\u4e3a\u4ec0\u4e48\u9700\u8981BERT&#xff1f;\u4f20\u7edf\u65b9\u6cd5\u7684\u5c40\u9650<\/li>\n<li>\n<ul>\n<li>1.1 \u4f20\u7edfRNN\u7f51\u7edc\u7684\u95ee\u9898<\/li>\n<li>1.2 \u4f20\u7edfword2vec\u7684\u95ee\u9898<\/li>\n<\/ul>\n<\/li>\n<li>\u4e8c\u3001BERT\u6846\u67b6\u6982\u8ff0<\/li>\n<li>\u4e09\u3001Transformer\u57fa\u7840\u67b6\u6784<\/li>\n<li>\n<ul>\n<li>3.1 Encoder-Decoder\u7f16\u7801\u89e3\u7801\u6846\u67b6<\/li>\n<li>3.2 Self-Attention\u81ea\u6ce8\u610f\u529b\u673a\u5236<\/li>\n<li>\n<ul>\n<li>\u7b2c\u4e00\u6b65&#xff1a;\u6784\u9020Q\u3001K\u3001V\u4e09\u4e2a\u77e9\u9635<\/li>\n<li>\u7b2c\u4e8c\u6b65&#xff1a;\u8ba1\u7b97\u6ce8\u610f\u529b\u5f97\u5206<\/li>\n<li>\u7b2c\u4e09\u6b65&#xff1a;\u5f52\u4e00\u5316&#xff08;Scaling&#xff09;<\/li>\n<li>\u7b2c\u56db\u6b65&#xff1a;\u52a0\u6743\u6c42\u548c<\/li>\n<\/ul>\n<\/li>\n<li>3.3 Multi-Head Attention\u591a\u5934\u6ce8\u610f\u529b<\/li>\n<li>3.4 \u591a\u5c42\u5806\u53e0<\/li>\n<li>3.5 \u4f4d\u7f6e\u7f16\u7801&#xff08;Positional Encoding&#xff09;<\/li>\n<li>\n<ul>\n<li>\u65b9\u6cd5\u4e00&#xff1a;0-1\u8303\u56f4\u5f52\u4e00\u5316\u4f4d\u7f6e<\/li>\n<li>\u65b9\u6cd5\u4e8c&#xff1a;1-n\u6b63\u6574\u6570\u7f16\u53f7<\/li>\n<li>\u65b9\u6cd5\u4e09&#xff1a;\u4e09\u89d2\u51fd\u6570\u4f4d\u7f6e\u7f16\u7801&#xff08;Transformer\u91c7\u7528&#xff09;<\/li>\n<\/ul>\n<\/li>\n<li>3.6 Add &amp; Norm \u6b8b\u5dee\u4e0e\u5f52\u4e00\u5316<\/li>\n<li>3.7 Transformer\u6574\u4f53\u6846\u67b6<\/li>\n<li>3.8 Decoder\u7684Shifted Right\u64cd\u4f5c<\/li>\n<\/ul>\n<\/li>\n<li>\u56db\u3001BERT\u9884\u8bad\u7ec3\u4efb\u52a1\u8be6\u89e3<\/li>\n<li>\n<ul>\n<li>4.1 Masked Language Model&#xff08;MLM&#xff09;\u906e\u853d\u8bed\u8a00\u6a21\u578b<\/li>\n<li>4.2 Next Sentence Prediction&#xff08;NSP&#xff09;\u4e0b\u4e00\u53e5\u9884\u6d4b<\/li>\n<\/ul>\n<\/li>\n<li>\u4e94\u3001BERT\u6a21\u578b\u53d8\u4f53\u4e0e\u53c2\u6570\u89c4\u683c<\/li>\n<li>\u516d\u3001BERT\u5fae\u8c03\u4e0e\u4e0b\u6e38\u5e94\u7528<\/li>\n<li>\n<ul>\n<li>\u5e38\u89c1\u4e0b\u6e38\u4efb\u52a1\u4e0e\u5fae\u8c03\u65b9\u5f0f<\/li>\n<\/ul>\n<\/li>\n<li>\u4e03\u3001\u603b\u7ed3<\/li>\n<li>\n<ul>\n<li>\n<ul>\n<li>BERT\u7684\u5386\u53f2\u5730\u4f4d<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr \/>\n<h2>\u4e00\u3001\u4e3a\u4ec0\u4e48\u9700\u8981BERT&#xff1f;\u4f20\u7edf\u65b9\u6cd5\u7684\u5c40\u9650<\/h2>\n<p>\u5728BERT\u51fa\u73b0\u4e4b\u524d&#xff0c;NLP\u9886\u57df\u4e3b\u8981\u9762\u4e34\u4e24\u5927\u75db\u70b9&#xff1a;\u4f20\u7edfRNN\u7684\u4e32\u884c\u8ba1\u7b97\u74f6\u9888\u4e0eword2vec\u8bcd\u5411\u91cf\u7684\u9759\u6001\u8868\u793a\u95ee\u9898\u3002<\/p>\n<h3>1.1 \u4f20\u7edfRNN\u7f51\u7edc\u7684\u95ee\u9898<\/h3>\n<p>RNN&#xff08;\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff09;\u901a\u8fc7\u9690\u85cf\u72b6\u6001 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         h<\/p>\n<p>         1<\/p>\n<p>        \u2192<\/p>\n<p>         h<\/p>\n<p>         2<\/p>\n<p>        \u2192<\/p>\n<p>        \u22ef<\/p>\n<p>        \u2192<\/p>\n<p>         h<\/p>\n<p>         m<\/p>\n<p>       h_1 \\\\rightarrow h_2 \\\\rightarrow \\\\dots \\\\rightarrow h_m<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">h<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">\u2192<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">h<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">\u2192<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.3669em\"><\/span><span class=\"minner\">\u22ef<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">\u2192<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">h<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1514em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">m<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u4f9d\u6b21\u5904\u7406\u5e8f\u5217&#xff0c;\u5b58\u5728\u4ee5\u4e0b\u7f3a\u9677&#xff1a;<\/p>\n<li>\u4e32\u884c\u4f9d\u8d56\u5bfc\u81f4\u8bad\u7ec3\u6162&#xff1a;\u5fc5\u987b\u7b49 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          h<\/p>\n<p>          1<\/p>\n<p>        h_1<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">h<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u8ba1\u7b97\u5b8c\u6210\u624d\u80fd\u8ba1\u7b97 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>          h<\/p>\n<p>          2<\/p>\n<p>        h_2<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">h<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff0c;\u65e0\u6cd5\u5e76\u884c\u5316&#xff0c;\u5e8f\u5217\u8d8a\u957f\u8bad\u7ec3\u65f6\u95f4\u8d8a\u957f\u3002<\/li>\n<li>\u5e76\u884c\u8ba1\u7b97\u6548\u679c\u5dee&#xff1a;\u96be\u4ee5\u5728\u591a\u53f0\u670d\u52a1\u5668\u4e0a\u540c\u65f6\u8bad\u7ec3\u540c\u4e00\u4e2a\u7f51\u7edc&#xff0c;\u65e0\u6cd5\u5145\u5206\u5229\u7528GPU\u96c6\u7fa4\u3002<\/li>\n<h3>1.2 \u4f20\u7edfword2vec\u7684\u95ee\u9898<\/h3>\n<li>\u8bcd\u5411\u91cf\u8bad\u7ec3\u597d\u540e\u4e0d\u518d\u6539\u53d8&#xff1a;\u65e0\u8bba\u4e0a\u4e0b\u6587\u5982\u4f55&#xff0c;\u540c\u4e00\u4e2a\u8bcd\u7684\u5411\u91cf\u6c38\u8fdc\u76f8\u540c\u3002<\/li>\n<li>\u65e0\u6cd5\u533a\u5206\u591a\u4e49\u8bcd&#xff1a;\u4e0d\u540c\u8bed\u5883\u4e0b\u7684\u540c\u4e00\u8bcd\u8bed\u542b\u4e49\u4e0d\u540c&#xff0c;\u4f46\u5171\u4eab\u540c\u4e00\u4e2a\u5411\u91cf\u3002\u4f8b\u5982&#034;\u673a\u8f66&#034;\u5728\u53f0\u6e7e\u4fda\u8bed\u4e2d\u6307&#034;\u7f57\u55e6&#034;&#xff0c;\u800c\u5728\u666e\u901a\u8bdd\u4e2d\u6307&#034;\u6469\u6258\u8f66&#034;&#xff0c;\u9759\u6001\u5411\u91cf\u65e0\u6cd5\u533a\u5206\u3002<\/li>\n<p>\u6838\u5fc3\u7ed3\u8bba&#xff1a;\u6211\u4eec\u9700\u8981\u4e00\u79cd\u80fd\u6839\u636e\u4e0a\u4e0b\u6587\u52a8\u6001\u751f\u6210\u8bcd\u5411\u91cf\u3001\u5e76\u4e14\u53ef\u4ee5\u5e76\u884c\u8ba1\u7b97\u7684\u6a21\u578b\u67b6\u6784\u2014\u2014\u8fd9\u6b63\u662fTransformer\u4e0eBERT\u8bde\u751f\u7684\u80cc\u666f\u3002<\/p>\n<hr \/>\n<h2>\u4e8c\u3001BERT\u6846\u67b6\u6982\u8ff0<\/h2>\n<p>BERT&#xff08;Bidirectional Encoder Representations from Transformers&#xff09;\u662f\u57fa\u4e8eTransformer\u7f16\u7801\u5668\u7684\u53cc\u5411\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u3002<\/p>\n<table>\n<tr>\u7279\u6027\u8bf4\u660e<\/tr>\n<tbody>\n<tr>\n<td>\u6a21\u578b\u7ed3\u6784<\/td>\n<td>\u57fa\u4e8eTransformer\u7684Encoder\u90e8\u5206&#xff0c;\u91c7\u7528\u591a\u5c42\u81ea\u6ce8\u610f\u529b\u673a\u5236\u548c\u524d\u9988\u795e\u7ecf\u7f51\u7edc<\/td>\n<\/tr>\n<tr>\n<td>\u6838\u5fc3\u80fd\u529b<\/td>\n<td>\u9884\u8bad\u7ec3\u5b66\u4e60\u4e30\u5bcc\u7684\u8bed\u8a00\u8868\u793a&#xff0c;\u53ef\u8fc1\u79fb\u5230\u5404\u7c7bNLP\u4e0b\u6e38\u4efb\u52a1<\/td>\n<\/tr>\n<tr>\n<td>\u9884\u8bad\u7ec3\u4efb\u52a1<\/td>\n<td>\u906e\u853d\u8bed\u8a00\u6a21\u578b&#xff08;MLM&#xff09;&#043; \u4e0b\u4e00\u53e5\u9884\u6d4b&#xff08;NSP&#xff09;<\/td>\n<\/tr>\n<tr>\n<td>\u53cc\u5411\u6027<\/td>\n<td>\u9884\u6d4b\u4e00\u4e2a\u8bcd\u65f6\u540c\u65f6\u8003\u8651\u5de6\u53f3\u4e24\u4fa7\u4e0a\u4e0b\u6587&#xff08;\u533a\u522b\u4e8eGPT\u7684\u5355\u5411\u81ea\u56de\u5f52&#xff09;<\/td>\n<\/tr>\n<tr>\n<td>\u5fae\u8c03\u65b9\u5f0f<\/td>\n<td>\u5728\u7279\u5b9a\u4efb\u52a1\u6570\u636e\u96c6\u4e0a\u8fdb\u4e00\u6b65\u8bad\u7ec3&#xff0c;\u9002\u914d\u4e0b\u6e38\u4efb\u52a1<\/td>\n<\/tr>\n<tr>\n<td>\u5386\u53f2\u5f71\u54cd<\/td>\n<td>\u5237\u65b011\u9879NLP\u57fa\u51c6\u6d4b\u8bd5\u8bb0\u5f55&#xff0c;\u5960\u5b9a\u9884\u8bad\u7ec3\u6a21\u578b\u8303\u5f0f&#xff0c;\u50ac\u751fRoBERTa\u3001ALBERT\u7b49\u540e\u7eed\u5de5\u4f5c<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<hr \/>\n<h2>\u4e09\u3001Transformer\u57fa\u7840\u67b6\u6784<\/h2>\n<p>BERT\u7684\u6838\u5fc3\u9aa8\u67b6\u662fTransformer\u7684Encoder\u90e8\u5206\u3002\u8981\u7406\u89e3BERT&#xff0c;\u5fc5\u987b\u5148\u7406\u89e3Transformer\u3002<\/p>\n<h3>3.1 Encoder-Decoder\u7f16\u7801\u89e3\u7801\u6846\u67b6<\/h3>\n<p>Encoder-Decoder&#xff08;\u7f16\u7801-\u89e3\u7801&#xff09;\u662f\u5927\u90e8\u5206\u6ce8\u610f\u529b\u6a21\u578b\u7684\u57fa\u7840\u6846\u67b6&#xff0c;\u4e3b\u8981\u7528\u4e8e\u5904\u7406\u5e8f\u5217\u5230\u5e8f\u5217&#xff08;Seq2Seq&#xff09; \u95ee\u9898\u2014\u2014\u8f93\u5165\u4e00\u4e2a\u5e8f\u5217&#xff0c;\u8f93\u51fa\u53e6\u4e00\u4e2a\u5e8f\u5217&#xff0c;\u4e14\u4e24\u4e2a\u5e8f\u5217\u957f\u5ea6\u53ef\u4ee5\u4e0d\u540c\u3002<\/p>\n<p>\u5178\u578b\u5e94\u7528\u573a\u666f&#xff1a;<\/p>\n<ul>\n<li>\u673a\u5668\u7ffb\u8bd1&#xff1a;\u8f93\u5165\u82f1\u6587\u5e8f\u5217 \u2192 \u8f93\u51fa\u4e2d\u6587\u5e8f\u5217<\/li>\n<li>\u6587\u672c\u6458\u8981&#xff1a;\u8f93\u5165\u6587\u7ae0\u5168\u6587 \u2192 \u8f93\u51fa\u6458\u8981<\/li>\n<li>\u95ee\u7b54\u7cfb\u7edf&#xff1a;\u8f93\u5165\u95ee\u9898 \u2192 \u8f93\u51fa\u7b54\u6848<\/li>\n<\/ul>\n<p>\u5de5\u4f5c\u6d41\u7a0b&#xff1a;<\/p>\n<p>\u8f93\u5165\u5e8f\u5217 &lt;x1, x2, &#8230;, xn&gt;  \u2192  [Encoder]  \u2192  \u8bed\u4e49\u7f16\u7801C  \u2192  [Decoder]  \u2192  \u8f93\u51fa\u5e8f\u5217 &lt;y1, y2, &#8230;, ym&gt;<\/p>\n<ul>\n<li>Encoder&#xff08;\u7f16\u7801\u5668&#xff09;&#xff1a;\u5c06\u8f93\u5165\u5e8f\u5217\u7f16\u7801\u4e3a\u4e00\u4e2a\u8bed\u4e49\u5411\u91cfC&#xff0c;C\u4e2d\u5b58\u50a8\u4e86\u6574\u4e2a\u8f93\u5165\u5e8f\u5217\u7684\u4fe1\u606f\u3002\u4f20\u7edf\u5b9e\u73b0\u53ef\u7528RNN\/LSTM\/GRU\/BiLSTM\u7b49\u3002<\/li>\n<li>Decoder&#xff08;\u89e3\u7801\u5668&#xff09;&#xff1a;\u6839\u636e\u8bed\u4e49\u7f16\u7801C&#xff0c;\u9010\u6b65\u89e3\u7801\u751f\u6210\u8f93\u51fa\u5e8f\u5217\u3002<\/li>\n<li>\u7f16\u7801\u4e0e\u89e3\u7801\u65b9\u5f0f\u53ef\u4efb\u610f\u7ec4\u5408&#xff1a;Encoder\u7528BiLSTM&#xff0c;Decoder\u7528GRU&#xff0c;\u5b8c\u5168\u53ef\u4ee5\u3002<\/li>\n<\/ul>\n<p>\u901a\u4fd7\u7406\u89e3&#xff1a;\u7ffb\u8bd1&#034;Bonjour&#034;\u65f6&#xff0c;\u5927\u8111\u5148\u628a\u8fd9\u4e2a\u8bcd\u7684\u6982\u5ff5\u63d0\u53d6\u51fa\u6765&#xff08;\u7f16\u7801&#xff09;&#xff0c;\u518d\u4ece\u82f1\u8bed\u8bcd\u5e93\u4e2d\u627e\u5230\u5bf9\u5e94\u7684&#034;Hello&#034;&#xff08;\u89e3\u7801&#xff09;\u3002\u63d0\u53d6\u51fa\u6765\u7684&#034;\u6982\u5ff5&#034;\u5c31\u662f\u8bed\u4e49\u5411\u91cfC\u3002<\/p>\n<h3>3.2 Self-Attention\u81ea\u6ce8\u610f\u529b\u673a\u5236<\/h3>\n<p>\u6838\u5fc3\u95ee\u9898&#xff1a;\u4e00\u53e5\u8bdd\u4e2d\u54ea\u4e9b\u8bcd\u662f\u91cd\u70b9&#xff1f;\u5982\u4f55\u8ba9\u8ba1\u7b97\u673a\u81ea\u52a8\u5173\u6ce8\u5230\u8bdd\u8bed\u4e2d\u7684\u5173\u952e\u8bcd&#xff1f;<\/p>\n<p>Self-Attention&#xff08;\u81ea\u6ce8\u610f\u529b&#xff09;\u901a\u8fc7\u8ba9\u6bcf\u4e2a\u8bcd\u4e0e\u5e8f\u5217\u4e2d\u6240\u6709\u8bcd\u8ba1\u7b97\u5173\u8054\u7a0b\u5ea6&#xff0c;\u52a8\u6001\u6355\u6349\u4e0a\u4e0b\u6587\u8bed\u4e49\u3002<\/p>\n<h4>\u7b2c\u4e00\u6b65&#xff1a;\u6784\u9020Q\u3001K\u3001V\u4e09\u4e2a\u77e9\u9635<\/h4>\n<p>\u8f93\u5165\u8bcd\u5411\u91cf\u901a\u8fc7\u4e09\u4e2a\u53ef\u8bad\u7ec3\u7684\u6743\u91cd\u77e9\u9635 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         W<\/p>\n<p>         Q<\/p>\n<p>       W^Q<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8413em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">Q<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u3001<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         W<\/p>\n<p>         K<\/p>\n<p>       W^K<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8413em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0715em\">K<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u3001<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         W<\/p>\n<p>         V<\/p>\n<p>       W^V<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8413em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.2222em\">V<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u5206\u522b\u6295\u5f71\u5f97\u5230&#xff1a;<\/p>\n<ul>\n<li>Q&#xff08;Query&#xff0c;\u67e5\u8be2&#xff09;&#xff1a;\u5f53\u524d\u8bcd&#034;\u4e3b\u52a8\u67e5\u8be2&#034;\u5176\u4ed6\u8bcd<\/li>\n<li>K&#xff08;Key&#xff0c;\u952e&#xff09;&#xff1a;\u5176\u4ed6\u8bcd&#034;\u7b49\u5f85\u88ab\u67e5\u8be2&#034;\u7684\u7279\u5f81<\/li>\n<li>V&#xff08;Value&#xff0c;\u503c&#xff09;&#xff1a;\u5176\u4ed6\u8bcd\u5b9e\u9645\u643a\u5e26\u7684\u4fe1\u606f\u7279\u5f81<\/li>\n<\/ul>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>          q<\/p>\n<p>          i<\/p>\n<p>         &#061;<\/p>\n<p>          x<\/p>\n<p>          i<\/p>\n<p>         \u22c5<\/p>\n<p>          W<\/p>\n<p>          Q<\/p>\n<p>         ,<\/p>\n<p>          k<\/p>\n<p>          i<\/p>\n<p>         &#061;<\/p>\n<p>          x<\/p>\n<p>          i<\/p>\n<p>         \u22c5<\/p>\n<p>          W<\/p>\n<p>          K<\/p>\n<p>         ,<\/p>\n<p>          v<\/p>\n<p>          i<\/p>\n<p>         &#061;<\/p>\n<p>          x<\/p>\n<p>          i<\/p>\n<p>         \u22c5<\/p>\n<p>          W<\/p>\n<p>          V<\/p>\n<p>        q_i &#061; x_i \\\\cdot W^Q, \\\\quad k_i &#061; x_i \\\\cdot W^K, \\\\quad v_i &#061; x_i \\\\cdot W^V<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.625em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">q<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.5945em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u22c5<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.0858em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8913em\"><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">Q<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 1em\"><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0315em\">k<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0315em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.5945em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u22c5<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.0858em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8913em\"><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0715em\">K<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 1em\"><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">v<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.5945em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u22c5<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8913em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8913em\"><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.2222em\">V<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<h4>\u7b2c\u4e8c\u6b65&#xff1a;\u8ba1\u7b97\u6ce8\u610f\u529b\u5f97\u5206<\/h4>\n<p>\u6bcf\u4e2a\u8bcd\u7684Q\u4e0e\u5e8f\u5217\u4e2d\u6240\u6709\u8bcd\u7684K\u8ba1\u7b97\u5185\u79ef&#xff0c;\u8868\u793a\u5339\u914d\u7a0b\u5ea6&#xff1a;<\/p>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         Score<\/p>\n<p>         (<\/p>\n<p>          q<\/p>\n<p>          i<\/p>\n<p>         ,<\/p>\n<p>          k<\/p>\n<p>          j<\/p>\n<p>         )<\/p>\n<p>         &#061;<\/p>\n<p>          q<\/p>\n<p>          i<\/p>\n<p>         \u22c5<\/p>\n<p>          k<\/p>\n<p>          j<\/p>\n<p>        \\\\text{Score}(q_i, k_j) &#061; q_i \\\\cdot k_j<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1.0361em;vertical-align: -0.2861em\"><\/span><span class=\"mord text\"><span class=\"mord\">Score<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">q<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0315em\">k<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0315em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6389em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">q<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u22c5<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.9805em;vertical-align: -0.2861em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0315em\">k<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0315em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<ul>\n<li><span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          q<\/p>\n<p>          1<\/p>\n<p>         \u22c5<\/p>\n<p>          k<\/p>\n<p>          1<\/p>\n<p>        q_1 \\\\cdot k_1<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6389em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">q<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u22c5<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0315em\">k<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0315em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff1a;\u7b2c1\u4e2a\u8bcd\u4e0e\u81ea\u8eab\u7684\u5339\u914d\u5ea6<\/li>\n<li><span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          q<\/p>\n<p>          1<\/p>\n<p>         \u22c5<\/p>\n<p>          k<\/p>\n<p>          2<\/p>\n<p>        q_1 \\\\cdot k_2<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6389em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">q<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u22c5<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0315em\">k<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0315em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff1a;\u7b2c1\u4e2a\u8bcd\u4e0e\u7b2c2\u4e2a\u8bcd\u7684\u5339\u914d\u5ea6<\/li>\n<\/ul>\n<h4>\u7b2c\u4e09\u6b65&#xff1a;\u5f52\u4e00\u5316&#xff08;Scaling&#xff09;<\/h4>\n<p>\u76f4\u63a5\u5185\u79ef\u5728\u7ef4\u5ea6\u8f83\u5927\u65f6\u6570\u503c\u4f1a\u5f88\u5927&#xff0c;\u5bfc\u81f4Softmax\u68af\u5ea6\u6d88\u5931\u3002\u56e0\u6b64\u9664\u4ee5 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>          d<\/p>\n<p>          k<\/p>\n<p>       \\\\sqrt{d_k}<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1.04em;vertical-align: -0.1828em\"><\/span><span class=\"mord sqrt\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8572em\"><span class=\"svg-align\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\" style=\"padding-left: 0.833em\"><span class=\"mord\"><span class=\"mord mathnormal\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -2.8172em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"hide-tail\" style=\"min-width: 0.853em;height: 1.08em\"><\/p>\n<p>           <\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1828em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u8fdb\u884c\u7f29\u653e&#xff1a;<\/p>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         Attention<\/p>\n<p>         (<\/p>\n<p>         Q<\/p>\n<p>         ,<\/p>\n<p>         K<\/p>\n<p>         ,<\/p>\n<p>         V<\/p>\n<p>         )<\/p>\n<p>         &#061;<\/p>\n<p>         Softmax<\/p>\n<p>          (<\/p>\n<p>            Q<\/p>\n<p>             K<\/p>\n<p>             T<\/p>\n<p>             d<\/p>\n<p>             k<\/p>\n<p>          )<\/p>\n<p>         V<\/p>\n<p>        \\\\text{Attention}(Q, K, V) &#061; \\\\text{Softmax}\\\\left(\\\\frac{QK^T}{\\\\sqrt{d_k}}\\\\right)V<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">Attention<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\">Q<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">K<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.2222em\">V<\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 2.4684em;vertical-align: -0.95em\"><\/span><span class=\"mord text\"><span class=\"mord\">Softmax<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\"><span class=\"mopen delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size3\">(<\/span><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.5183em\"><span class=\"\" style=\"top: -2.2528em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord sqrt\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8572em\"><span class=\"svg-align\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\" style=\"padding-left: 0.833em\"><span class=\"mord\"><span class=\"mord mathnormal\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -2.8172em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"hide-tail\" style=\"min-width: 0.853em;height: 1.08em\"><\/p>\n<p>                    <\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1828em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">Q<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">K<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.1389em\">T<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.93em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mclose delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size3\">)<\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.2222em\">V<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p>\u5176\u4e2d <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         d<\/p>\n<p>         k<\/p>\n<p>       d_k<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u662fK\u5411\u91cf\u7684\u7ef4\u5ea6\u3002\u8fd9\u4e00\u6b65\u5f52\u4e00\u5316\u7684\u4f5c\u7528\u662f&#xff1a;\u9632\u6b62\u7ef4\u5ea6\u589e\u5927\u5bfc\u81f4\u5185\u79ef\u503c\u8fc7\u5927&#xff0c;\u4fdd\u6301\u68af\u5ea6\u7a33\u5b9a\u3002<\/p>\n<h4>\u7b2c\u56db\u6b65&#xff1a;\u52a0\u6743\u6c42\u548c<\/h4>\n<p>\u5f52\u4e00\u5316\u540e\u7684\u6ce8\u610f\u529b\u6743\u91cd\u4e0eV\u76f8\u4e58\u6c42\u548c&#xff0c;\u5f97\u5230\u6bcf\u4e2a\u8bcd\u7684\u6700\u7ec8\u7279\u5f81\u8868\u793a <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         z<\/p>\n<p>         i<\/p>\n<p>       z_i<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.5806em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u3002<\/p>\n<p>\u5173\u952e\u7279\u6027&#xff1a;\u6bcf\u4e2a\u8bcd\u7684Q\u4f1a\u4e0e\u6574\u4e2a\u5e8f\u5217\u4e2d\u6bcf\u4e00\u4e2aK\u8ba1\u7b97\u5f97\u5206&#xff0c;\u518d\u57fa\u4e8e\u5f97\u5206\u5206\u914dV\u7684\u6743\u91cd\u3002\u56e0\u6b64\u540c\u6837\u7684\u8bcd\u5728\u4e0d\u540c\u53e5\u5b50\u4e2d\u4f1a\u5f97\u5230\u4e0d\u540c\u7684\u7279\u5f81\u5411\u91cf\u2014\u2014\u8fd9\u6b63\u662f\u89e3\u51b3word2vec\u9759\u6001\u8868\u793a\u95ee\u9898\u7684\u6838\u5fc3\u3002<\/p>\n<p>\u5e76\u884c\u4f18\u52bf&#xff1a;Q\u3001K\u3001V\u77e9\u9635\u53ef\u4ee5\u4e00\u6b21\u6027\u5e76\u884c\u8ba1\u7b97&#xff0c;\u5b8c\u5168\u907f\u514d\u4e86RNN\u7684\u4e32\u884c\u4f9d\u8d56&#xff0c;\u6781\u5927\u63d0\u5347\u4e86\u8bad\u7ec3\u901f\u5ea6\u3002<\/p>\n<h3>3.3 Multi-Head Attention\u591a\u5934\u6ce8\u610f\u529b<\/h3>\n<p>\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u4e2d\u7528\u591a\u7ec4\u5377\u79ef\u6838\u63d0\u53d6\u4e0d\u540c\u7279\u5f81\u56fe\u3002\u540c\u7406&#xff0c;Transformer\u4e2d\u7528\u591a\u7ec4Q\u3001K\u3001V\u53ef\u4ee5\u5f97\u5230\u591a\u79cd\u8bcd\u7684\u7279\u5f81\u8868\u8fbe\u3002<\/p>\n<p>\u591a\u5934\u673a\u5236\u7684\u5de5\u4f5c\u6d41\u7a0b&#xff1a;<\/p>\n<li>\u62c6\u5206\u591a\u5934&#xff1a;\u5c06 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          d<\/p>\n<p>           m<\/p>\n<p>           o<\/p>\n<p>           d<\/p>\n<p>           e<\/p>\n<p>           l<\/p>\n<p>        d_{model}<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">m<\/span><span class=\"mord mathnormal mtight\">o<\/span><span class=\"mord mathnormal mtight\">d<\/span><span class=\"mord mathnormal mtight\">e<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u7ef4\u7684Q\u3001K\u3001V\u62c6\u5206\u4e3a <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         h<\/p>\n<p>        h<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6944em\"><\/span><span class=\"mord mathnormal\">h<\/span><\/span><\/span><\/span><\/span> \u4e2a\u4f4e\u7ef4\u5b50\u7a7a\u95f4&#xff08;\u901a\u5e38 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         h<\/p>\n<p>         &#061;<\/p>\n<p>         8<\/p>\n<p>        h&#061;8<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6944em\"><\/span><span class=\"mord mathnormal\">h<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">8<\/span><\/span><\/span><\/span><\/span>&#xff09;&#xff0c;\u6bcf\u4e2a\u5934\u72ec\u7acb\u8ba1\u7b97Attention\u3002<\/li>\n<li>\u5e76\u884c\u8ba1\u7b97&#xff1a;\u6bcf\u4e2a\u5934\u72ec\u7acb\u8f93\u51fa <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          Z<\/p>\n<p>          0<\/p>\n<p>         ,<\/p>\n<p>          Z<\/p>\n<p>          1<\/p>\n<p>         ,<\/p>\n<p>         \u2026<\/p>\n<p>         ,<\/p>\n<p>          Z<\/p>\n<p>           h<\/p>\n<p>           \u2212<\/p>\n<p>           1<\/p>\n<p>        Z_0, Z_1, \\\\dots, Z_{h-1}<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8917em;vertical-align: -0.2083em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">Z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0715em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">Z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0715em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\">\u2026<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">Z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0715em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">h<\/span><span class=\"mbin mtight\">\u2212<\/span><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2083em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff0c;\u5206\u522b\u5173\u6ce8\u4e0d\u540c\u7684\u8bed\u4e49\u4fe1\u606f&#xff08;\u6709\u7684\u5934\u5173\u6ce8\u8bed\u6cd5&#xff0c;\u6709\u7684\u5934\u5173\u6ce8\u6307\u4ee3\u5173\u7cfb&#xff09;\u3002<\/li>\n<li>\u62fc\u63a5&#xff1a;\u5c06\u6240\u6709\u5934\u7684\u8f93\u51fa\u6cbf\u7279\u5f81\u7ef4\u5ea6\u62fc\u63a5&#xff1a;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>         Concat<\/p>\n<p>         (<\/p>\n<p>          Z<\/p>\n<p>          0<\/p>\n<p>         ,<\/p>\n<p>          Z<\/p>\n<p>          1<\/p>\n<p>         ,<\/p>\n<p>         \u2026<\/p>\n<p>         ,<\/p>\n<p>          Z<\/p>\n<p>           h<\/p>\n<p>           \u2212<\/p>\n<p>           1<\/p>\n<p>         )<\/p>\n<p>        \\\\text{Concat}(Z_0, Z_1, \\\\dots, Z_{h-1})<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">Concat<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">Z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0715em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">Z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0715em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\">\u2026<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">Z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0715em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">h<\/span><span class=\"mbin mtight\">\u2212<\/span><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2083em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span>\u3002<\/li>\n<li>\u7ebf\u6027\u964d\u7ef4&#xff1a;\u62fc\u63a5\u540e\u901a\u8fc7\u4e00\u4e2a\u5168\u8fde\u63a5\u5c42 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          W<\/p>\n<p>          O<\/p>\n<p>        W^O<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8413em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0278em\">O<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u964d\u7ef4&#xff0c;\u5f97\u5230\u6700\u7ec8\u8f93\u51fa\u3002<\/li>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         MultiHead<\/p>\n<p>         (<\/p>\n<p>         Q<\/p>\n<p>         ,<\/p>\n<p>         K<\/p>\n<p>         ,<\/p>\n<p>         V<\/p>\n<p>         )<\/p>\n<p>         &#061;<\/p>\n<p>         Concat<\/p>\n<p>         (<\/p>\n<p>          head<\/p>\n<p>          1<\/p>\n<p>         ,<\/p>\n<p>         \u2026<\/p>\n<p>         ,<\/p>\n<p>          head<\/p>\n<p>          h<\/p>\n<p>         )<\/p>\n<p>          W<\/p>\n<p>          O<\/p>\n<p>        \\\\text{MultiHead}(Q, K, V) &#061; \\\\text{Concat}(\\\\text{head}_1, \\\\dots, \\\\text{head}_h) W^O<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">MultiHead<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\">Q<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">K<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.2222em\">V<\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.1413em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">Concat<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord text\"><span class=\"mord\">head<\/span><\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\">\u2026<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord text\"><span class=\"mord\">head<\/span><\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">h<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8913em\"><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0278em\">O<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p>\u5176\u4e2d <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         head<\/p>\n<p>         i<\/p>\n<p>        &#061;<\/p>\n<p>        Attention<\/p>\n<p>        (<\/p>\n<p>        Q<\/p>\n<p>         W<\/p>\n<p>         i<\/p>\n<p>         Q<\/p>\n<p>        ,<\/p>\n<p>        K<\/p>\n<p>         W<\/p>\n<p>         i<\/p>\n<p>         K<\/p>\n<p>        ,<\/p>\n<p>        V<\/p>\n<p>         W<\/p>\n<p>         i<\/p>\n<p>         V<\/p>\n<p>        )<\/p>\n<p>       \\\\text{head}_i &#061; \\\\text{Attention}(QW_i^Q, KW_i^K, VW_i^V)<\/p>\n<p>    <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord text\"><span class=\"mord\">head<\/span><\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.2361em;vertical-align: -0.2769em\"><\/span><span class=\"mord text\"><span class=\"mord\">Attention<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\">Q<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.9592em\"><span class=\"\" style=\"top: -2.4231em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><span class=\"\" style=\"top: -3.1809em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">Q<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2769em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0715em\">K<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -2.4413em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0715em\">K<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2587em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.2222em\">V<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8413em\"><span class=\"\" style=\"top: -2.4413em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.2222em\">V<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2587em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span>\u3002<\/p>\n<h3>3.4 \u591a\u5c42\u5806\u53e0<\/h3>\n<p>\u5355\u5c42Self-Attention\u7684\u8868\u8fbe\u80fd\u529b\u6709\u9650&#xff0c;\u5b9e\u9645\u4f7f\u7528\u65f6\u5c06\u591a\u4e2aEncoder Layer\u5806\u53e0\u8d77\u6765&#xff08;\u7c7b\u4f3cCNN\u5806\u591a\u5c42\u5377\u79ef&#xff09;&#xff1a;<\/p>\n<p>\u8f93\u5165Embedding \u2192 [Encoder Layer 1] \u2192 [Encoder Layer 2] \u2192 &#8230; \u2192 [Encoder Layer N] \u2192 \u8f93\u51fa\u8868\u793a<\/p>\n<p>BERT-base\u5806\u53e0\u4e8612\u5c42&#xff0c;BERT-large\u5806\u53e0\u4e8624\u5c42\u3002\u5c42\u6570\u8d8a\u591a&#xff0c;\u6a21\u578b\u80fd\u6355\u6349\u7684\u8bed\u4e49\u5c42\u6b21\u8d8a\u6df1\u3002<\/p>\n<h3>3.5 \u4f4d\u7f6e\u7f16\u7801&#xff08;Positional Encoding&#xff09;<\/h3>\n<p>\u95ee\u9898&#xff1a;Self-Attention\u672c\u8eab\u662f&#034;\u65e0\u5e8f&#034;\u7684\u2014\u2014\u5b83\u53ea\u5173\u5fc3\u53e5\u5b50\u4e2d\u6709\u54ea\u4e9b\u8bcd&#xff0c;\u4e0d\u5173\u5fc3\u8bcd\u7684\u987a\u5e8f\u3002\u4f46\u8bed\u5e8f\u5bf9\u8bed\u8a00\u81f3\u5173\u91cd\u8981&#xff08;\u201c\u6211\u6253\u4f60\u201d vs \u201c\u4f60\u6253\u6211\u201d&#xff09;\u3002<\/p>\n<p>\u56e0\u6b64\u5fc5\u987b\u5f15\u5165\u4f4d\u7f6e\u4fe1\u606f\u3002<\/p>\n<h4>\u65b9\u6cd5\u4e00&#xff1a;0-1\u8303\u56f4\u5f52\u4e00\u5316\u4f4d\u7f6e<\/h4>\n<p>\u6211\u559c\u6b22\u5403\u6d0b\u8471      \u2192 [0, 0.16, 0.32, &#8230;, 1.0]<br \/>\n\u6211\u771f\u7684\u4e0d\u559c\u6b22\u5403\u6d0b\u8471 \u2192 [0, 0.125, 0.25, &#8230;, 1.0]<\/p>\n<p>\u7f3a\u9677&#xff1a;\u53e5\u5b50\u957f\u5ea6\u4e0d\u540c&#xff0c;\u76f8\u540c\u4f4d\u7f6e\u7684\u7f16\u7801\u503c\u4e0d\u540c&#xff0c;\u65e0\u6cd5\u8de8\u53e5\u5b50\u6bd4\u8f83\u4f4d\u7f6e\u3002<\/p>\n<h4>\u65b9\u6cd5\u4e8c&#xff1a;1-n\u6b63\u6574\u6570\u7f16\u53f7<\/h4>\n<p>\u6211\u559c\u6b22\u5403\u6d0b\u8471      \u2192 [1, 2, 3, 4, 5, 6]<br \/>\n\u6211\u771f\u7684\u4e0d\u559c\u6b22\u5403\u6d0b\u8471 \u2192 [1, 2, 3, 4, 5, 6, 7]<\/p>\n<p>\u7f3a\u9677&#xff1a;\u53e5\u5b50\u8d8a\u957f&#xff0c;\u540e\u9762\u7684\u6570\u5b57\u8d8a\u5927&#xff0c;\u5bfc\u81f4\u4f4d\u7f6e\u6743\u91cd\u88ab\u6570\u503c\u5927\u5c0f\u626d\u66f2\u3002<\/p>\n<h4>\u65b9\u6cd5\u4e09&#xff1a;\u4e09\u89d2\u51fd\u6570\u4f4d\u7f6e\u7f16\u7801&#xff08;Transformer\u91c7\u7528&#xff09;<\/h4>\n<p>Transformer\u4f5c\u8005\u8bbe\u8ba1\u4e86\u57fa\u4e8e\u6b63\u5f26\/\u4f59\u5f26\u7684\u4f4d\u7f6e\u7f16\u7801\u516c\u5f0f&#xff1a;<\/p>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         P<\/p>\n<p>          E<\/p>\n<p>           (<\/p>\n<p>           p<\/p>\n<p>           o<\/p>\n<p>           s<\/p>\n<p>           ,<\/p>\n<p>           2<\/p>\n<p>           i<\/p>\n<p>           )<\/p>\n<p>         &#061;<\/p>\n<p>         sin<\/p>\n<p>         \u2061<\/p>\n<p>          (<\/p>\n<p>            p<\/p>\n<p>            o<\/p>\n<p>            s<\/p>\n<p>            10000<\/p>\n<p>             2<\/p>\n<p>             i<\/p>\n<p>             \/<\/p>\n<p>              d<\/p>\n<p>               m<\/p>\n<p>               o<\/p>\n<p>               d<\/p>\n<p>               e<\/p>\n<p>               l<\/p>\n<p>          )<\/p>\n<p>        PE_{(pos, 2i)} &#061; \\\\sin\\\\left(\\\\frac{pos}{10000^{2i\/d_{model}}}\\\\right)<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1.0385em;vertical-align: -0.3552em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">P<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3448em\"><span class=\"\" style=\"top: -2.5198em;margin-left: -0.0576em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mopen mtight\">(<\/span><span class=\"mord mathnormal mtight\">p<\/span><span class=\"mord mathnormal mtight\">os<\/span><span class=\"mpunct mtight\">,<\/span><span class=\"mord mtight\">2<\/span><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mclose mtight\">)<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3552em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.854em;vertical-align: -0.704em\"><\/span><span class=\"mop\">sin<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\"><span class=\"mopen delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size2\">(<\/span><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.1076em\"><span class=\"\" style=\"top: -2.296em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\">1000<\/span><span class=\"mord\"><span class=\"mord\">0<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.814em\"><span class=\"\" style=\"top: -2.989em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">2<\/span><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mord mtight\">\/<\/span><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3448em\"><span class=\"\" style=\"top: -2.3488em;margin-left: 0em;margin-right: 0.0714em\"><span class=\"pstrut\" style=\"height: 2.5em\"><\/span><span class=\"sizing reset-size3 size1 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">m<\/span><span class=\"mord mathnormal mtight\">o<\/span><span class=\"mord mathnormal mtight\">d<\/span><span class=\"mord mathnormal mtight\">e<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1512em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">p<\/span><span class=\"mord mathnormal\">os<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.704em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mclose delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size2\">)<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         P<\/p>\n<p>          E<\/p>\n<p>           (<\/p>\n<p>           p<\/p>\n<p>           o<\/p>\n<p>           s<\/p>\n<p>           ,<\/p>\n<p>           2<\/p>\n<p>           i<\/p>\n<p>           &#043;<\/p>\n<p>           1<\/p>\n<p>           )<\/p>\n<p>         &#061;<\/p>\n<p>         cos<\/p>\n<p>         \u2061<\/p>\n<p>          (<\/p>\n<p>            p<\/p>\n<p>            o<\/p>\n<p>            s<\/p>\n<p>            10000<\/p>\n<p>             2<\/p>\n<p>             i<\/p>\n<p>             \/<\/p>\n<p>              d<\/p>\n<p>               m<\/p>\n<p>               o<\/p>\n<p>               d<\/p>\n<p>               e<\/p>\n<p>               l<\/p>\n<p>          )<\/p>\n<p>        PE_{(pos, 2i&#043;1)} &#061; \\\\cos\\\\left(\\\\frac{pos}{10000^{2i\/d_{model}}}\\\\right)<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1.0385em;vertical-align: -0.3552em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">P<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3448em\"><span class=\"\" style=\"top: -2.5198em;margin-left: -0.0576em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mopen mtight\">(<\/span><span class=\"mord mathnormal mtight\">p<\/span><span class=\"mord mathnormal mtight\">os<\/span><span class=\"mpunct mtight\">,<\/span><span class=\"mord mtight\">2<\/span><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mbin mtight\">&#043;<\/span><span class=\"mord mtight\">1<\/span><span class=\"mclose mtight\">)<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3552em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.854em;vertical-align: -0.704em\"><\/span><span class=\"mop\">cos<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\"><span class=\"mopen delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size2\">(<\/span><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.1076em\"><span class=\"\" style=\"top: -2.296em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\">1000<\/span><span class=\"mord\"><span class=\"mord\">0<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.814em\"><span class=\"\" style=\"top: -2.989em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">2<\/span><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mord mtight\">\/<\/span><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3448em\"><span class=\"\" style=\"top: -2.3488em;margin-left: 0em;margin-right: 0.0714em\"><span class=\"pstrut\" style=\"height: 2.5em\"><\/span><span class=\"sizing reset-size3 size1 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">m<\/span><span class=\"mord mathnormal mtight\">o<\/span><span class=\"mord mathnormal mtight\">d<\/span><span class=\"mord mathnormal mtight\">e<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1512em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">p<\/span><span class=\"mord mathnormal\">os<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.704em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mclose delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size2\">)<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<ul>\n<li>pos&#xff1a;\u8bcd\u5728\u53e5\u5b50\u4e2d\u7684\u4f4d\u7f6e&#xff08;\u4ece0\u5f00\u59cb&#xff09;<\/li>\n<li><span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>           d<\/p>\n<p>            m<\/p>\n<p>            o<\/p>\n<p>            d<\/p>\n<p>            e<\/p>\n<p>            l<\/p>\n<p>         d_{model}<\/p>\n<p>      <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">m<\/span><span class=\"mord mathnormal mtight\">o<\/span><span class=\"mord mathnormal mtight\">d<\/span><span class=\"mord mathnormal mtight\">e<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff1a;\u8bcd\u5411\u91cf\u7684\u7ef4\u5ea6<\/li>\n<li><span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          i<\/p>\n<p>         i<\/p>\n<p>      <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6595em\"><\/span><span class=\"mord mathnormal\">i<\/span><\/span><\/span><\/span><\/span>&#xff1a;\u7ef4\u5ea6\u7d22\u5f15&#xff08;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         i<\/p>\n<p>         &#061;<\/p>\n<p>         0<\/p>\n<p>         ,<\/p>\n<p>         1<\/p>\n<p>         ,<\/p>\n<p>         \u2026<\/p>\n<p>         ,<\/p>\n<p>          d<\/p>\n<p>           m<\/p>\n<p>           o<\/p>\n<p>           d<\/p>\n<p>           e<\/p>\n<p>           l<\/p>\n<p>         \/<\/p>\n<p>         2<\/p>\n<p>         \u2212<\/p>\n<p>         1<\/p>\n<p>        i &#061; 0, 1, \\\\dots, d_{model}\/2 &#8211; 1<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6595em\"><\/span><span class=\"mord mathnormal\">i<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord\">0<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\">1<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\">\u2026<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">d<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">m<\/span><span class=\"mord mathnormal mtight\">o<\/span><span class=\"mord mathnormal mtight\">d<\/span><span class=\"mord mathnormal mtight\">e<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord\">\/2<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">1<\/span><\/span><\/span><\/span><\/span>&#xff09;<\/li>\n<li>\u5076\u6570\u7ef4\u7528sin&#xff0c;\u5947\u6570\u7ef4\u7528cos<\/li>\n<\/ul>\n<p>\u6700\u7ec8\u8f93\u5165&#xff1a;\u5c06\u8bcd\u5411\u91cf&#xff08;Word Embedding&#xff09;\u4e0e\u4f4d\u7f6e\u7f16\u7801&#xff08;Positional Encoding&#xff09;\u9010\u70b9\u76f8\u52a0&#xff1a;<\/p>\n<p><span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         Input<\/p>\n<p>         &#061;<\/p>\n<p>         WordEmbedding<\/p>\n<p>         &#043;<\/p>\n<p>         PositionalEncoding<\/p>\n<p>        \\\\text{Input} &#061; \\\\text{WordEmbedding} &#043; \\\\text{PositionalEncoding}<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8778em;vertical-align: -0.1944em\"><\/span><span class=\"mord text\"><span class=\"mord\">Input<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8889em;vertical-align: -0.1944em\"><\/span><span class=\"mord text\"><span class=\"mord\">WordEmbedding<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8889em;vertical-align: -0.1944em\"><\/span><span class=\"mord text\"><span class=\"mord\">PositionalEncoding<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p>\u4f18\u70b9&#xff1a;<\/p>\n<li>PE\u503c\u5206\u5e03\u5728 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>         [<\/p>\n<p>         \u2212<\/p>\n<p>         1<\/p>\n<p>         ,<\/p>\n<p>         1<\/p>\n<p>         ]<\/p>\n<p>        [-1, 1]<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mopen\">[<\/span><span class=\"mord\">\u2212<\/span><span class=\"mord\">1<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\">1<\/span><span class=\"mclose\">]<\/span><\/span><\/span><\/span><\/span> \u533a\u95f4&#xff0c;\u4e0d\u4f1a\u968f\u5e8f\u5217\u957f\u5ea6\u589e\u957f\u800c\u653e\u5927\u3002<\/li>\n<li>\u4e0d\u540c\u53e5\u5b50\u4e2d\u76f8\u540c\u4f4d\u7f6e\u7684PE\u503c\u76f8\u540c&#xff08;\u5982pos&#061;0\u65f6PE&#061;0&#xff09;&#xff0c;\u5177\u6709\u8de8\u53e5\u5b50\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/li>\n<li>\u76f8\u5bf9\u4f4d\u7f6e\u53ef\u88ab\u5b66\u4e60&#xff1a;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>         P<\/p>\n<p>          E<\/p>\n<p>           p<\/p>\n<p>           o<\/p>\n<p>           s<\/p>\n<p>           &#043;<\/p>\n<p>           k<\/p>\n<p>        PE_{pos&#043;k}<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.9694em;vertical-align: -0.2861em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">P<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0576em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">p<\/span><span class=\"mord mathnormal mtight\">os<\/span><span class=\"mbin mtight\">&#043;<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u53ef\u4ee5\u8868\u793a\u4e3a <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         P<\/p>\n<p>          E<\/p>\n<p>           p<\/p>\n<p>           o<\/p>\n<p>           s<\/p>\n<p>        PE_{pos}<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.9694em;vertical-align: -0.2861em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">P<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1514em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0576em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">p<\/span><span class=\"mord mathnormal mtight\">os<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u7684\u7ebf\u6027\u51fd\u6570\u3002<\/li>\n<h3>3.6 Add &amp; Norm \u6b8b\u5dee\u4e0e\u5f52\u4e00\u5316<\/h3>\n<p>\u6bcf\u4e2aEncoder Layer\u5185\u90e8\u5305\u542b\u4e24\u4e2a\u5173\u952e\u64cd\u4f5c&#xff1a;<\/p>\n<ul>\n<li>\n<p>Add&#xff08;\u6b8b\u5dee\u8fde\u63a5&#xff09;&#xff1a;\u5c06Attention\u7684\u8f93\u5165\u76f4\u63a5\u52a0\u5230\u8f93\u51fa\u4e0a&#xff08;\u7c7b\u4f3cResNet\u7684shortcut&#xff09;&#xff0c;\u7f13\u89e3\u6df1\u5c42\u7f51\u7edc\u68af\u5ea6\u6d88\u5931\u95ee\u9898\u3002 <span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>           Output<\/p>\n<p>           &#061;<\/p>\n<p>           Sublayer<\/p>\n<p>           (<\/p>\n<p>           x<\/p>\n<p>           )<\/p>\n<p>           &#043;<\/p>\n<p>           x<\/p>\n<p>          \\\\text{Output} &#061; \\\\text{Sublayer}(x) &#043; x<\/p>\n<p>       <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8778em;vertical-align: -0.1944em\"><\/span><span class=\"mord text\"><span class=\"mord\">Output<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">Sublayer<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\">x<\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.4306em\"><\/span><span class=\"mord mathnormal\">x<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<\/li>\n<li>\n<p>Norm&#xff08;Layer Normalization&#xff09;&#xff1a;\u5bf9\u6bcf\u4e2a\u6837\u672c\u7684\u7279\u5f81\u505a\u5f52\u4e00\u5316&#xff0c;\u52a0\u901f\u6536\u655b\u3001\u7a33\u5b9a\u8bad\u7ec3\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u5b8c\u6574\u5b50\u5c42\u7ed3\u6784&#xff1a;<\/p>\n<p>x \u2192 Multi-Head Attention \u2192 Add(x &#043; attn_output) \u2192 Norm \u2192 Feed Forward \u2192 Add \u2192 Norm \u2192 \u8f93\u51fa<\/p>\n<h3>3.7 Transformer\u6574\u4f53\u6846\u67b6<\/h3>\n<p>Transformer\u6574\u4f53\u7531Encoder\u5806\u6808\u548cDecoder\u5806\u6808\u7ec4\u6210&#xff1a;<\/p>\n<ul>\n<li>Encoder\u4fa7&#xff1a;N\u5c42Encoder Layer&#xff0c;\u6bcf\u5c42\u5305\u542b Multi-Head Attention &#043; Feed Forward&#xff08;\u4e24\u5c42\u5168\u8fde\u63a5&#xff09;\u3002<\/li>\n<li>Decoder\u4fa7&#xff1a;N\u5c42Decoder Layer&#xff0c;\u5305\u542b Masked Multi-Head Attention &#043; Encoder-Decoder Attention &#043; Feed Forward\u3002<\/li>\n<li>Feed Forward\u5c42&#xff1a;\u4e00\u4e2a\u4e24\u5c42\u7684\u5168\u8fde\u63a5\u795e\u7ecf\u7f51\u7edc&#xff0c;\u4e2d\u95f4\u6fc0\u6d3b\u51fd\u6570\u4e3aReLU\/GELU\u3002<\/li>\n<\/ul>\n<p>\u91cd\u8981\u533a\u5206&#xff1a;BERT\u53ea\u7528\u4e86Encoder\u90e8\u5206&#xff0c;\u6ca1\u6709Decoder\u3002\u8fd9\u662fBERT\u4e0eGPT&#xff08;\u53ea\u7528Decoder&#xff09;\u3001\u539f\u59cbTransformer&#xff08;\u4e24\u8005\u90fd\u7528&#xff09;\u7684\u6838\u5fc3\u7ed3\u6784\u533a\u522b\u3002<\/p>\n<h3>3.8 Decoder\u7684Shifted Right\u64cd\u4f5c<\/h3>\n<p>Decoder\u5728\u8bad\u7ec3\u65f6\u4f7f\u7528&#034;Shifted Right&#034;\u6280\u5de7&#xff1a;<\/p>\n<ul>\n<li>\u5c06\u76ee\u6807\u8f93\u51fa\u5e8f\u5217\u53f3\u79fb\u4e00\u4f4d&#xff0c;\u6700\u5de6\u4fa7\u6dfb\u52a0\u8d77\u59cb\u7b26 &lt;sos&gt;&#xff08;Start Of Sentence&#xff09;\u3002<\/li>\n<li>\u8bad\u7ec3\u65f6\u89e3\u7801\u5668\u4ee5 &lt;sos&gt; \u5f00\u5934&#xff0c;\u9010\u6b65\u9884\u6d4b\u4e0b\u4e00\u4e2a\u8bcd\u3002<\/li>\n<\/ul>\n<p>\u793a\u4f8b&#xff08;\u4e2d\u8bd1\u82f1\u4efb\u52a1&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u8f93\u5165&#xff1a;\u6211\u7231\u4e2d\u56fd<\/li>\n<li>\u76ee\u6807\u8f93\u51fa&#xff1a;I love China<\/li>\n<li>Decoder\u8f93\u5165&#xff08;shifted right&#xff09;&#xff1a;&lt;sos&gt; I love China<\/li>\n<li>Decoder\u8f93\u51fa\u9884\u6d4b&#xff1a;I love China &lt;eos&gt;<\/li>\n<\/ul>\n<p>\u8fd9\u6837\u89e3\u7801\u5668\u5728\u751f\u6210\u6bcf\u4e2a\u8bcd\u65f6&#xff0c;\u90fd\u80fd\u57fa\u4e8e\u5df2\u7ecf\u751f\u6210\u7684\u524d\u7f00\u8fdb\u884c\u9884\u6d4b&#xff0c;\u5b9e\u73b0\u81ea\u56de\u5f52\u751f\u6210\u3002<\/p>\n<hr \/>\n<h2>\u56db\u3001BERT\u9884\u8bad\u7ec3\u4efb\u52a1\u8be6\u89e3<\/h2>\n<p>BERT\u5728\u5927\u89c4\u6a21\u65e0\u6807\u6ce8\u8bed\u6599\u4e0a\u8fdb\u884c\u9884\u8bad\u7ec3&#xff0c;\u4f7f\u7528\u4e24\u4e2a\u81ea\u76d1\u7763\u4efb\u52a1&#xff1a;<\/p>\n<h3>4.1 Masked Language Model&#xff08;MLM&#xff09;\u906e\u853d\u8bed\u8a00\u6a21\u578b<\/h3>\n<p>\u6838\u5fc3\u601d\u60f3&#xff1a;\u968f\u673a\u906e\u853d\u8f93\u5165\u4e2d15%\u7684\u8bcd&#xff0c;\u8ba9\u6a21\u578b\u6839\u636e\u4e0a\u4e0b\u6587\u9884\u6d4b\u88ab\u906e\u853d\u7684\u8bcd\u3002\u8fd9\u5c31\u662f&#034;\u5b8c\u5f62\u586b\u7a7a&#034;\u4efb\u52a1\u3002<\/p>\n<p>15%\u7684Masking\u7b56\u7565\u7ec6\u8282&#xff08;BERT\u8bba\u6587\u7684\u91cd\u8981\u8bbe\u8ba1&#xff09;&#xff1a;<\/p>\n<p>\u88ab\u9009\u4e2d\u768415%\u7684\u8bcd\u4e2d&#xff1a;<\/p>\n<ul>\n<li>80% \u66ff\u6362\u4e3a [MASK] \u6807\u8bb0&#xff0c;\u5982&#034;\u6211\u4eca\u5929\u53bb\u9762\u8bd5&#034; \u2192 \u201c\u6211[MASK]\u5929\u53bb[MASK]\u8bd5\u201d<\/li>\n<li>10% \u968f\u673a\u66ff\u6362\u4e3a\u53e6\u4e00\u4e2a\u8bcd&#xff08;\u9632\u6b62\u6a21\u578b\u8fc7\u5ea6\u4f9d\u8d56[MASK]\u6807\u8bb0&#xff09;<\/li>\n<li>10% \u4fdd\u6301\u539f\u8bcd\u4e0d\u53d8&#xff08;\u964d\u4f4e\u9884\u8bad\u7ec3\u4e0e\u5fae\u8c03\u7684\u5dee\u8ddd&#xff09;<\/li>\n<\/ul>\n<p>\u4e3a\u4ec0\u4e48\u6309\u5b57Mask\u800c\u4e0d\u662f\u6309\u8bcdMask&#xff1f; \u56e0\u4e3a\u8bcd\u7684\u7ec4\u5408\u592a\u591a&#xff08;\u4eca\u5929\u3001\u660e\u5929\u3001\u540e\u5929\u3001\u4e0a\u5348\u3001\u4e0b\u5348\u2026\u2026&#xff09;&#xff0c;\u6309\u5b57Mask\u7684\u9884\u6d4b\u7a7a\u95f4\u66f4\u5c0f&#xff0c;\u6a21\u578b\u66f4\u5bb9\u6613\u5b66\u3002<\/p>\n<p>\u53cc\u5411\u6027\u7684\u4f53\u73b0&#xff1a;\u9884\u6d4b\u88abMask\u7684\u8bcd\u65f6&#xff0c;\u6a21\u578b\u540c\u65f6\u770b\u5230\u5de6\u53f3\u4e24\u4fa7\u7684\u4e0a\u4e0b\u6587\u2014\u2014\u8fd9\u6b63\u662fBERT&#034;\u53cc\u5411&#034;\u7684\u542b\u4e49\u3002<\/p>\n<h3>4.2 Next Sentence Prediction&#xff08;NSP&#xff09;\u4e0b\u4e00\u53e5\u9884\u6d4b<\/h3>\n<p>\u6838\u5fc3\u601d\u60f3&#xff1a;\u5224\u65ad\u4e24\u4e2a\u53e5\u5b50A\u548cB\u5728\u539f\u6587\u4e2d\u662f\u5426\u662f\u8fde\u7eed\u7684\u4e0a\u4e0b\u53e5\u3002<\/p>\n<p>\u8f93\u5165\u6784\u9020&#xff1a;<\/p>\n<p>[CLS] \u53e5\u5b50A [SEP] \u53e5\u5b50B [SEP]<\/p>\n<p>\u6b63\u6837\u672c&#xff08;Label &#061; IsNext&#xff09;&#xff1a;<\/p>\n<p>[CLS] \u6211\u4eca\u5929\u53bb\u9762\u8bd5 [SEP] \u51c6\u5907\u597d\u4e86\u7b80\u5386 [SEP]   \u2192  Label: yes<\/p>\n<p>\u8d1f\u6837\u672c&#xff08;Label &#061; NotNext&#xff09;&#xff1a;<\/p>\n<p>[CLS] \u6211\u4eca\u5929\u53bb\u9762\u8bd5 [SEP] \u4e2d\u5348\u5403\u4e86\u5305\u5b50 [SEP]   \u2192  Label: no<\/p>\n<p>\u7279\u6b8a\u6807\u8bb0\u8bf4\u660e&#xff1a;<\/p>\n<table>\n<tr>\u6807\u8bb0\u5168\u79f0\u4f5c\u7528<\/tr>\n<tbody>\n<tr>\n<td>[CLS]<\/td>\n<td>Classification Token<\/td>\n<td>\u653e\u5728\u5e8f\u5217\u5f00\u5934&#xff0c;\u6700\u7ec8\u8f93\u51fa\u7528\u4e8e\u5206\u7c7b\u4efb\u52a1&#xff1b;\u8bad\u7ec3\u4e2d\u4e5f\u4f5c\u4e3a\u4e00\u4e2a\u666e\u901atoken\u53c2\u4e0e\u5b66\u4e60<\/td>\n<\/tr>\n<tr>\n<td>[SEP]<\/td>\n<td>Separator Token<\/td>\n<td>\u5206\u9694\u4e24\u4e2a\u53e5\u5b50&#xff0c;\u6216\u8868\u793a\u5355\u53e5\u7ed3\u675f<\/td>\n<\/tr>\n<tr>\n<td>[MASK]<\/td>\n<td>Mask Token<\/td>\n<td>MLM\u4efb\u52a1\u4e2d\u88ab\u906e\u853d\u7684\u4f4d\u7f6e<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u4e3a\u4ec0\u4e48\u9700\u8981NSP&#xff1f; \u8bb8\u591a\u4e0b\u6e38\u4efb\u52a1&#xff08;\u5982\u95ee\u7b54QA\u3001\u81ea\u7136\u8bed\u8a00\u63a8\u7406NLI&#xff09;\u9700\u8981\u7406\u89e3\u53e5\u5b50\u95f4\u7684\u5173\u7cfb&#xff0c;NSP\u8ba9\u6a21\u578b\u5b66\u5230\u53e5\u95f4\u8fde\u8d2f\u6027&#xff0c;\u800c\u4e0d\u4ec5\u4ec5\u662f\u5355\u8bcd\u8bed\u4e49\u3002<\/p>\n<hr \/>\n<h2>\u4e94\u3001BERT\u6a21\u578b\u53d8\u4f53\u4e0e\u53c2\u6570\u89c4\u683c<\/h2>\n<p>BERT\u8bba\u6587\u4e2d\u53d1\u5e03\u4e86\u4e24\u4e2a\u7248\u672c&#xff1a;<\/p>\n<table>\n<tr>\u6a21\u578b\u89c4\u683c\u5c42\u6570&#xff08;Encoder&#xff09;\u9690\u85cf\u7ef4\u5ea6Attention\u5934\u6570\u53c2\u6570\u91cf\u9002\u7528\u573a\u666f<\/tr>\n<tbody>\n<tr>\n<td>BERT-base<\/td>\n<td>12<\/td>\n<td>768<\/td>\n<td>12<\/td>\n<td>~110M<\/td>\n<td>\u65e5\u5e38\u5fae\u8c03\u3001\u8d44\u6e90\u6709\u9650\u573a\u666f<\/td>\n<\/tr>\n<tr>\n<td>BERT-large<\/td>\n<td>24<\/td>\n<td>1024<\/td>\n<td>16<\/td>\n<td>~340M<\/td>\n<td>\u8ffd\u6c42SOTA\u6548\u679c\u3001\u7b97\u529b\u5145\u8db3\u573a\u666f<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u5176\u4ed6\u5173\u952e\u8d85\u53c2\u6570&#xff1a;<\/p>\n<ul>\n<li>\u6700\u5927\u5e8f\u5217\u957f\u5ea6&#xff1a;512 tokens<\/li>\n<li>Batch Size&#xff1a;128,000 words<\/li>\n<li>\u8bad\u7ec3\u6b65\u6570&#xff1a;1,000,000\u6b65<\/li>\n<li>\u8bad\u7ec3\u6570\u636e&#xff1a;BooksCorpus&#xff08;8\u4ebf\u8bcd&#xff09;&#043; English Wikipedia&#xff08;25\u4ebf\u8bcd&#xff09;<\/li>\n<li>\u4f18\u5316\u5668&#xff1a;Adam&#xff0c;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\n<p>          \u03b2<\/p>\n<p>          1<\/p>\n<p>         &#061;<\/p>\n<p>         0.9<\/p>\n<p>        \\\\beta_1&#061;0.9<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8889em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0528em\">\u03b2<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0528em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">0.9<\/span><\/span><\/span><\/span><\/span>, <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>          \u03b2<\/p>\n<p>          2<\/p>\n<p>         &#061;<\/p>\n<p>         0.999<\/p>\n<p>        \\\\beta_2&#061;0.999<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8889em;vertical-align: -0.1944em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0528em\">\u03b2<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0528em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">0.999<\/span><\/span><\/span><\/span><\/span>, <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\"><\/p>\n<p>         LR<\/p>\n<p>         &#061;<\/p>\n<p>         1<\/p>\n<p>         e<\/p>\n<p>         \u2212<\/p>\n<p>         4<\/p>\n<p>        \\\\text{LR}&#061;1e-4<\/p>\n<p>     <\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord text\"><span class=\"mord\">LR<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.7278em;vertical-align: -0.0833em\"><\/span><span class=\"mord\">1<\/span><span class=\"mord mathnormal\">e<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">4<\/span><\/span><\/span><\/span><\/span><\/li>\n<li>Warmup Steps&#xff1a;10,000<\/li>\n<\/ul>\n<hr \/>\n<h2>\u516d\u3001BERT\u5fae\u8c03\u4e0e\u4e0b\u6e38\u5e94\u7528<\/h2>\n<p>\u9884\u8bad\u7ec3\u5b8c\u6210\u540e&#xff0c;BERT\u53ef\u4ee5\u901a\u8fc7\u5fae\u8c03&#xff08;Fine-tuning&#xff09; \u9002\u914d\u5404\u7c7b\u4e0b\u6e38NLP\u4efb\u52a1\u3002\u5fae\u8c03\u65f6\u5c06\u9884\u8bad\u7ec3\u53c2\u6570\u521d\u59cb\u5316&#xff0c;\u7528\u4e0b\u6e38\u4efb\u52a1\u6570\u636e\u7ee7\u7eed\u8bad\u7ec3\u5373\u53ef\u3002<\/p>\n<h3>\u5e38\u89c1\u4e0b\u6e38\u4efb\u52a1\u4e0e\u5fae\u8c03\u65b9\u5f0f<\/h3>\n<table>\n<tr>\u4efb\u52a1\u7c7b\u578b\u8f93\u5165\u793a\u4f8b\u5fae\u8c03\u65b9\u5f0f<\/tr>\n<tbody>\n<tr>\n<td>\u6587\u672c\u5206\u7c7b<\/td>\n<td>\u5355\u53e5 \u2192 \u60c5\u611f\/\u4e3b\u9898\u7c7b\u522b<\/td>\n<td>\u53d6[CLS]\u8f93\u51fa\u63a5\u5206\u7c7b\u5c42<\/td>\n<\/tr>\n<tr>\n<td>\u547d\u540d\u5b9e\u4f53\u8bc6\u522b&#xff08;NER&#xff09;<\/td>\n<td>\u53e5\u5b50 \u2192 \u6bcf\u4e2a\u8bcd\u7684\u5b9e\u4f53\u6807\u7b7e<\/td>\n<td>\u5bf9\u6bcf\u4e2atoken\u8f93\u51fa\u63a5\u5206\u7c7b\u5c42<\/td>\n<\/tr>\n<tr>\n<td>\u95ee\u7b54\u7cfb\u7edf&#xff08;QA&#xff09;<\/td>\n<td>\u95ee\u9898&#043;\u6bb5\u843d \u2192 \u7b54\u6848\u8d77\u6b62\u4f4d\u7f6e<\/td>\n<td>\u9884\u6d4b\u7b54\u6848\u7684start\/end span<\/td>\n<\/tr>\n<tr>\n<td>\u81ea\u7136\u8bed\u8a00\u63a8\u7406&#xff08;NLI&#xff09;<\/td>\n<td>\u524d\u63d0&#043;\u5047\u8bbe \u2192 \u8574\u542b\/\u77db\u76fe\/\u4e2d\u7acb<\/td>\n<td>\u7528[CLS]\u505a\u53e5\u5bf9\u5206\u7c7b<\/td>\n<\/tr>\n<tr>\n<td>\u6587\u672c\u76f8\u4f3c\u5ea6<\/td>\n<td>\u53e5\u5b50A&#043;\u53e5\u5b50B \u2192 \u76f8\u4f3c\u5ea6\u5206\u6570<\/td>\n<td>\u53e5\u5bf9\u8868\u793a\u63a5\u56de\u5f52\u5c42<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>BERT\u7684\u7075\u6d3b\u6027&#xff1a;\u53ea\u9700\u5728Encoder\u8f93\u51fa\u540e\u63a5\u4e00\u4e2a\u7b80\u5355\u7684\u4efb\u52a1\u5c42&#xff08;\u5206\u7c7b\u5c42\/span\u9884\u6d4b\u5c42\u7b49&#xff09;&#xff0c;\u51e0\u4e4e\u6240\u6709NLP\u4efb\u52a1\u90fd\u80fd\u9002\u914d\u2014\u2014\u8fd9\u662fBERT\u901a\u7528\u6027\u5f3a\u7684\u6839\u672c\u539f\u56e0\u3002<\/p>\n<hr \/>\n<h2>\u4e03\u3001\u603b\u7ed3<\/h2>\n<li>BERT\u7684\u672c\u8d28&#xff1a;\u4e00\u4e2a\u57fa\u4e8eTransformer Encoder\u7684\u53cc\u5411\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b&#xff0c;\u901a\u8fc7MLM\u548cNSP\u4e24\u4e2a\u81ea\u76d1\u7763\u4efb\u52a1\u5b66\u4e60\u6df1\u5c42\u8bed\u8a00\u8868\u793a\u3002<\/li>\n<li>\u6280\u672f\u7a81\u7834&#xff1a;\n<ul>\n<li>\u53cc\u5411\u4e0a\u4e0b\u6587\u5efa\u6a21&#xff08; vs GPT\u7684\u5355\u5411&#xff09;<\/li>\n<li>Self-Attention\u5e76\u884c\u8ba1\u7b97&#xff08; vs RNN\u7684\u4e32\u884c&#xff09;<\/li>\n<li>\u9884\u8bad\u7ec3&#043;\u5fae\u8c03\u8303\u5f0f&#xff08;\u53d6\u4ee3\u4efb\u52a1\u5b9a\u5236\u6a21\u578b&#xff09;<\/li>\n<\/ul>\n<\/li>\n<li>\u5173\u952e\u7ec4\u4ef6&#xff1a;Self-Attention \u2192 Multi-Head \u2192 \u4f4d\u7f6e\u7f16\u7801 \u2192 \u6b8b\u5dee\u8fde\u63a5 \u2192 LayerNorm \u2192 \u591a\u5c42\u5806\u53e0\u3002<\/li>\n<h4>BERT\u7684\u5386\u53f2\u5730\u4f4d<\/h4>\n<p>BERT\u5f00\u542f\u4e86NLP\u7684&#034;\u9884\u8bad\u7ec3\u65f6\u4ee3&#034;&#xff0c;\u540e\u7eed\u6d8c\u73b0\u4e86\u5927\u91cf\u6539\u8fdb\u6a21\u578b&#xff1a;<\/p>\n<ul>\n<li>RoBERTa&#xff1a;\u53bb\u6389NSP\u3001\u66f4\u5927batch\u3001\u66f4\u591a\u6570\u636e&#xff0c;\u8fdb\u4e00\u6b65\u63d0\u5347\u6548\u679c<\/li>\n<li>ALBERT&#xff1a;\u7f29\u5c0f\u53c2\u6570\u91cf&#xff08;\u8de8\u5c42\u53c2\u6570\u5171\u4eab\u3001\u8bcd\u5411\u91cf\u5206\u89e3&#xff09;&#xff0c;\u63d0\u5347\u8bad\u7ec3\u6548\u7387<\/li>\n<li>DistilBERT&#xff1a;\u6a21\u578b\u84b8\u998f&#xff0c;\u53c2\u6570\u91cf\u51cf\u5c1140%&#xff0c;\u901f\u5ea6\u63d0\u534760%&#xff0c;\u4fdd\u755997%\u6548\u679c<\/li>\n<li>GPT\u7cfb\u5217&#xff1a;Decoder\u8def\u7ebf\u7684\u81ea\u56de\u5f52\u9884\u8bad\u7ec3&#xff0c;\u8d70\u5411\u751f\u6210\u5f0fAI<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u76ee\u5f55\u4e00\u3001\u4e3a\u4ec0\u4e48\u9700\u8981BERT&#xff1f;\u4f20\u7edf\u65b9\u6cd5\u7684\u5c40\u96501.1 \u4f20\u7edfRNN\u7f51\u7edc\u7684\u95ee\u98981.2 \u4f20\u7edfword2vec\u7684\u95ee\u9898\u4e8c\u3001BERT\u6846\u67b6\u6982\u8ff0\u4e09\u3001Transformer\u57fa\u7840\u67b6\u67843.1 Encoder-Decoder\u7f16\u7801\u89e3\u7801\u6846\u67b63.2 Self-Attention\u81ea\u6ce8\u610f\u529b\u673a\u5236\u7b2c\u4e00\u6b65&#xff1a;\u6784\u9020Q\u3001K\u3001V\u4e09\u4e2a\u77e9\u9635\u7b2c\u4e8c\u6b65&#xff1a;\u8ba1\u7b97\u6ce8\u610f\u529b\u5f97\u5206\u7b2c\u4e09\u6b65&#xff1a;\u5f52\u4e00\u5316&#xff08;Scaling&#xff09;\u7b2c\u56db\u6b65&#xff1a;\u52a0\u6743\u6c42\u548c3.3 Multi-H<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[2554,841,224],"topic":[],"class_list":["post-121101","post","type-post","status-publish","format-standard","hentry","category-server","tag-bert","tag-transformer","tag-224"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/121101.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u76ee\u5f55\u4e00\u3001\u4e3a\u4ec0\u4e48\u9700\u8981BERT&#xff1f;\u4f20\u7edf\u65b9\u6cd5\u7684\u5c40\u96501.1 \u4f20\u7edfRNN\u7f51\u7edc\u7684\u95ee\u98981.2 \u4f20\u7edfword2vec\u7684\u95ee\u9898\u4e8c\u3001BERT\u6846\u67b6\u6982\u8ff0\u4e09\u3001Transformer\u57fa\u7840\u67b6\u67843.1 Encoder-Decoder\u7f16\u7801\u89e3\u7801\u6846\u67b63.2 Self-Attention\u81ea\u6ce8\u610f\u529b\u673a\u5236\u7b2c\u4e00\u6b65&#xff1a;\u6784\u9020Q\u3001K\u3001V\u4e09\u4e2a\u77e9\u9635\u7b2c\u4e8c\u6b65&#xff1a;\u8ba1\u7b97\u6ce8\u610f\u529b\u5f97\u5206\u7b2c\u4e09\u6b65&#xff1a;\u5f52\u4e00\u5316&#xff08;Scaling&#xff09;\u7b2c\u56db\u6b65&#xff1a;\u52a0\u6743\u6c42\u548c3.3 Multi-H\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/121101.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-01T19:17:18+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121101.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/121101.html\",\"name\":\"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-11-01T19:17:18+00:00\",\"dateModified\":\"2026-11-01T19:17:18+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/121101.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/121101.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121101.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/121101.html","og_locale":"zh_CN","og_type":"article","og_title":"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u76ee\u5f55\u4e00\u3001\u4e3a\u4ec0\u4e48\u9700\u8981BERT&#xff1f;\u4f20\u7edf\u65b9\u6cd5\u7684\u5c40\u96501.1 \u4f20\u7edfRNN\u7f51\u7edc\u7684\u95ee\u98981.2 \u4f20\u7edfword2vec\u7684\u95ee\u9898\u4e8c\u3001BERT\u6846\u67b6\u6982\u8ff0\u4e09\u3001Transformer\u57fa\u7840\u67b6\u67843.1 Encoder-Decoder\u7f16\u7801\u89e3\u7801\u6846\u67b63.2 Self-Attention\u81ea\u6ce8\u610f\u529b\u673a\u5236\u7b2c\u4e00\u6b65&#xff1a;\u6784\u9020Q\u3001K\u3001V\u4e09\u4e2a\u77e9\u9635\u7b2c\u4e8c\u6b65&#xff1a;\u8ba1\u7b97\u6ce8\u610f\u529b\u5f97\u5206\u7b2c\u4e09\u6b65&#xff1a;\u5f52\u4e00\u5316&#xff08;Scaling&#xff09;\u7b2c\u56db\u6b65&#xff1a;\u52a0\u6743\u6c42\u548c3.3 Multi-H","og_url":"https:\/\/www.wsisp.com\/helps\/121101.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-11-01T19:17:18+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"7 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/121101.html","url":"https:\/\/www.wsisp.com\/helps\/121101.html","name":"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-11-01T19:17:18+00:00","dateModified":"2026-11-01T19:17:18+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/121101.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/121101.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/121101.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"BERT\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6846\u67b6\u6df1\u5ea6\u89e3\u6790\uff1a\u4eceTransformer\u5230\u53cc\u5411\u9884\u8bad\u7ec3\u6a21\u578b"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121101","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=121101"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121101\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=121101"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=121101"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=121101"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=121101"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}