{"id":36989,"date":"2025-01-22T01:01:00","date_gmt":"2025-01-21T17:01:00","guid":{"rendered":"https:\/\/17aitech.com\/?p=36989"},"modified":"2025-01-22T01:01:00","modified_gmt":"2025-01-21T17:01:00","slug":"%e8%a2%ab%e5%bf%bd%e7%95%a5%e7%9a%84%e8%b5%b7%e7%82%b9%ef%bc%9fkarpathy%e6%8f%ad%e7%a7%98%e6%9c%80%e5%88%9d%e7%9a%84%e6%b3%a8%e6%84%8f%e5%8a%9b%e8%ae%ba%e6%96%87%e8%a2%abtransformer%e5%85%89%e8%8a%92","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=36989","title":{"rendered":"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b"},"content":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b<\/a><\/p>\n

\u51e0\u4e2a\u5c0f\u65f6\u524d\uff0c\u8457\u540d AI \u7814\u7a76\u8005\u3001OpenAI \u521b\u59cb\u6210\u5458\u4e4b\u4e00 Andrej Karpathy \u53d1\u5e03\u4e86\u4e00\u7bc7\u5907\u53d7\u5173\u6ce8\u7684\u957f\u63a8\u6587\uff0c\u5176\u4e2d\u5206\u4eab\u4e86\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u80cc\u540e\u4e00\u4e9b\u6216\u8bb8\u5c11\u6709\u4eba\u77e5\u7684\u6545\u4e8b\u3002<\/p>\n

\u5176\u4e2d\u6700\u503c\u5f97\u6ce8\u610f\u7684\u4e00\u4e2a\u6545\u4e8b\u662f\u771f\u6b63\u9996\u6b21\u63d0\u51fa\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u7684\u8bba\u6587\u5176\u5b9e\u662f Dzmitry Bahdanau\u3001Kyunghyun Cho \u548c Yoshua Bengio \u7684\u300aNeural Machine Translation by Jointly Learning to Align and Translate\u300b\uff0c\u8fd9\u6bd4\u300aAttention is All you Need\u300b\u8fd8\u65e9 3 \u5e74<\/strong>\uff0c\u4f46\u5f88\u663e\u7136\uff0c\u8fd9\u7bc7\u8bba\u6587\u5e76\u6ca1\u6709\u6536\u83b7\u540e\u8005\u90a3\u822c\u7684\u5173\u6ce8\u3002\u00a0<\/p>\n

<\/a><\/section>\n
\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Karpathy \u957f\u63a8\u6587\u7684\u4e0d\u5b8c\u6574\u622a\u56fe<\/sup><\/em><\/section>\n
<\/section>\n
\u5b9e\u9645\u4e0a\uff0c\u8fd9\u4e2a\u6545\u4e8b\u6765\u81ea Dzmitry Bahdanau \u53d1\u7ed9 Karpathy \u7684\u4e00\u5c01\u90ae\u4ef6\u3002Bahdanau \u662f ServiceNow Research \u7684\u7814\u7a76\u79d1\u5b66\u5bb6\u548c\u7814\u7a76\u8d1f\u8d23\u4eba\u4ee5\u53ca\u9ea6\u5409\u5c14\u5927\u5b66\u517c\u804c\u6559\u6388\u3002<\/section>\n
<\/section>\n
\u4ed6\u5728\u53d1\u7ed9 Karpathy \u7684\u8fd9\u5c01\u90ae\u4ef6\u4e2d\u5206\u4eab\u4e86\u81ea\u5df1\u53d1\u73b0\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u7684\u65c5\u7a0b\u4ee5\u53ca Attention \u8fd9\u4e2a\u672f\u8bed\u7684\u7531\u6765\u2014\u2014\u5176\u5b9e\u6765\u81ea Yoshua Bengio\u3002\u6b64\u5916\uff0c\u4ed6\u4e5f\u63d0\u5230\u4e86 Alex Graves \u7684 NMT \u8bba\u6587\u548c Jason Weston \u7684\u8bb0\u5fc6\u7f51\u7edc\uff08Memory Networks\uff09\u8bba\u6587\u5404\u81ea\u72ec\u7acb\u53d1\u73b0\u7c7b\u4f3c\u673a\u5236\u7684\u6545\u4e8b\u3002<\/section>\n
<\/section>\n
Karpathy \u63a8\u6587\u53d1\u5e03\u540e\u53cd\u54cd\u70ed\u70c8\uff0c\u77ed\u65f6\u95f4\u5185\u5c31\u5df2\u6709\u8d85\u8fc7 20 \u4e07\u9605\u8bfb\u91cf\uff0c\u5f88\u591a\u8bfb\u8005\u90fd\u88ab\u8fd9\u4e2a\u6ce8\u610f\u529b\u80cc\u540e\u7684\u6545\u4e8b\u5438\u5f15\u3002<\/section>\n
<\/a><\/section>\n
\u6709\u8bfb\u8005\u5728\u770b\u8fc7\u8fd9\u4e2a\u6545\u4e8b\u540e\u53d1\u51fa\u611f\u53f9\uff1a2013-2017 \u5e74\u95f4\u7684\u6df1\u5ea6\u5b66\u4e60<\/mark>\u5b87\u5b99\u4e2d\u6709\u5f88\u591a\u9690\u85cf\u7684\u82f1\u96c4\u3002<\/section>\n
<\/a><\/section>\n
\u4e5f\u6709\u4eba\u5206\u4eab\u81ea\u5df1\u5bf9\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u7684\u770b\u6cd5\u3002<\/section>\n
<\/a><\/section>\n
Hyperbolic Labs \u521b\u59cb\u4eba\u548c CTO Yuchen Jin \u66f4\u662f\u6253\u8da3\u8bf4\u300aAttention Is All You Need\u300b\u7684\u53e6\u4e00\u9879\u91cd\u8981\u8d21\u732e\u662f\u5c06\u540e\u9762\u7684 AI \u8bba\u6587\u6807\u9898\u5e26\u6b6a\u4e86\uff1a\u300c\u5438\u5f15\u6ce8\u610f\u529b\u7684\u6807\u9898\u624d\u662f All You Need\u300d\u3002<\/section>\n
<\/a><\/section>\n
\u6ce8\u610f\u529b\u673a\u5236<\/mark>\u7684\u80cc\u540e\u6545\u4e8b<\/strong><\/section>\n
<\/section>\n
\u4e0b\u9762\u6211\u4eec\u5c31\u6765\u770b\u770b Karpathy \u7684\u63a8\u6587\u7a76\u7adf\u8bf4\u4e86\u4ec0\u4e48\uff1a<\/section>\n
<\/section>\n
\u300cattention\u300d\u7b97\u5b50\u2014\u2014\u4e5f\u5c31\u662f\u63d0\u51fa\u4e86 Transformer \u7684\u300aAttention is All you Need\u300b\u4e2d\u7684\u90a3\u4e2a\u6ce8\u610f\u529b\uff0c\u80cc\u540e\u7684\uff08\u771f\u5b9e\uff09\u5f00\u53d1\u548c\u7075\u611f\u6545\u4e8b\u3002\u6765\u81ea\u5927\u7ea6 2 \u5e74\u524d\u4e0e\u4f5c\u8005 @DBahdanau \u7684\u4e2a\u4eba\u7535\u5b50\u90ae\u4ef6\u901a\u4fe1\uff0c\u5728\u6b64\u53d1\u5e03\uff08\u7ecf\u8bb8\u53ef\uff09\u3002\u6b64\u524d\u51e0\u5929\uff0c\u7f51\u4e0a\u6d41\u4f20\u7740\u4e00\u4e9b\u5173\u4e8e\u5176\u5f00\u53d1\u8fc7\u7a0b\u7684\u5047\u65b0\u95fb\u3002<\/section>\n
<\/section>\n
Attention \u662f\u4e00\u79cd\u51fa\u8272\u7684\uff08\u6570\u636e\u4f9d\u8d56\u578b\uff09\u52a0\u6743\u5e73\u5747\u8fd0\u7b97\u3002\u5b83\u662f\u4e00\u79cd\u5f62\u5f0f\u7684\u5168\u5c40\u6c60\u5316<\/mark>\u3001\u5f52\u7ea6\u3001\u901a\u4fe1\u3002\u5b83\u662f\u4e00\u79cd\u4ece\u591a\u4e2a\u8282\u70b9\uff08token\u3001\u56fe\u5757\u7b49\uff09\u805a\u5408\u76f8\u5173\u4fe1\u606f\u7684\u65b9\u6cd5\u3002\u5b83\u5bcc\u6709\u8868\u73b0\u529b\u3001\u529f\u80fd\u5f3a\u5927\u3001\u5177\u6709\u8db3\u591f\u7684\u5e76\u884c\u6027\uff0c\u5e76\u4e14\u53ef\u4ee5\u9ad8\u6548\u4f18\u5316\u3002\u751a\u81f3\u591a\u5c42\u611f\u77e5\u5668\uff08MLP\uff09\u5b9e\u9645\u4e0a\u4e5f\u53ef\u4ee5\u5927\u81f4\u91cd\u5199\u4e3a\u6570\u636e\u72ec\u7acb\u578b\u6743\u91cd<\/mark>\u4e0a\u7684 Attention\uff08\u7b2c\u4e00\u5c42\u6743\u91cd<\/mark>\u662f\u67e5\u8be2<\/mark>\uff0c\u7b2c\u4e8c\u5c42\u6743\u91cd<\/mark>\u662f\u503c\uff0c\u952e\u5c31\u662f\u8f93\u5165\uff0csoftmax \u53d8\u4e3a\u5143\u7d20\u7ea7\uff0c\u5220\u9664\u4e86\u89c4\u8303\u5316<\/mark>\uff09\u3002\u7b80\u5355\u6765\u8bf4\uff0c\u6ce8\u610f\u529b\u975e\u5e38\u68d2\uff0c\u662f\u795e\u7ecf\u7f51\u7edc<\/mark>\u67b6\u6784\u8bbe\u8ba1\u4e2d\u7684\u91cd\u5927<\/strong>\u7a81\u7834\u3002<\/section>\n
<\/section>\n
\u300aAttention is All You Need\u300b\u83b7\u5f97\u7684\u2026\u2026 \u5443\u2026\u2026 \u6ce8\u610f\u529b\u5dee\u4e0d\u591a\u662f 3 \u5e74\u524d\u771f\u6b63\u63d0\u51fa Attention \u7684\u8bba\u6587\u7684 100 \u500d\uff0c\u5373 Dzmitry Bahdanau\u3001Kyunghyun Cho \u548c Yoshua Bengio \u7684\u8bba\u6587\u300aNeural Machine Translation by Jointly Learning to Align and Translate\u300b\u3002\u5728\u6211\u770b\u6765\uff0c\u8fd9\u4e00\u76f4\u6709\u70b9\u51fa\u4eba\u610f\u6599\u3002\u987e\u540d\u601d\u4e49\uff0c\u300aAttention is All You Need\u300b\u7684\u6838\u5fc3\u8d21\u732e\u662f\u63d0\u51fa\uff1aTransformer \u795e\u7ecf\u7f51\u7edc<\/mark>\u5c31\u662f\u5220\u9664\u6ce8\u610f\u529b\u4e4b\u5916\u7684\u4e00\u5207\uff0c\u7136\u540e\u57fa\u672c\u4e0a\u5c31\u662f\u5c06\u5176\u5806\u53e0<\/mark>\u5728\u5e26\u6709 MLP\uff08\u6839\u636e\u4e0a\u8ff0\u5185\u5bb9\uff0c\u8fd9\u4e5f\u53ef\u4ee5\u5927\u81f4\u88ab\u89c6\u4e3a\u6ce8\u610f\u529b\uff09\u7684 ResNet \u4e2d<\/strong>\u3002\u4f46\u6211\u786e\u5b9e\u8ba4\u4e3a\u8fd9\u7bc7 Transformer \u8bba\u6587\u6709\u81ea\u5df1\u72ec\u7279\u7684\u4ef7\u503c\uff0c\u56e0\u4e3a\u5b83\u4e00\u6b21\u6027\u6dfb\u52a0\u4e86\u5176\u5b83\u8bb8\u591a\u4ee4\u4eba\u60ca\u53f9\u7684\u60f3\u6cd5\uff0c\u5305\u62ec\u4f4d\u7f6e\u7f16\u7801\u3001\u7f29\u653e\u5f0f\u6ce8\u610f\u529b\u3001\u591a\u5934\u6ce8\u610f\u529b\u3001\u5404\u5411\u540c\u6027\u7684\u7b80\u5355\u8bbe\u8ba1<\/strong>\u7b49\u3002\u5728\u6211\u770b\u6765\uff0c\u76f4\u5230\u4eca\u5929\uff08\u5927\u7ea6 7 \u5e74\u8fc7\u53bb\u4e86\uff09\uff0cTransformer \u57fa\u672c\u4e0a\u8fd8\u4fdd\u6301\u7740 2017 \u5e74\u7684\u5f62\u5f0f\uff0c\u53ea\u6709\u76f8\u5bf9\u8f83\u5c11\u7684\u5fae\u5c0f\u4fee\u6539\uff0c\u4e5f\u8bb8\u9664\u4e86\u4f7f\u7528\u66f4\u597d\u7684\u4f4d\u7f6e\u7f16\u7801\u65b9\u6848\uff08RoPE \u7b49\uff09\u3002<\/section>\n
<\/section>\n
\u603b\u4e4b\uff0c\u6211\u5148\u628a\u5b8c\u6574\u90ae\u4ef6\u8d34\u5728\u4e0b\u9762\uff0c\u5176\u4e2d\u4e5f\u6697\u793a\u4e86\u8fd9\u4e2a\u8fd0\u7b97\u4e00\u5f00\u59cb\u88ab\u79f0\u4e3a Attention \u7684\u539f\u56e0 \u2014\u2014 \u5b83\u6e90\u81ea\u5bf9\u6e90\u53e5\u5b50\u4e2d\u8bcd\u7684\u5173\u6ce8\uff08attending\uff09\u5e76\u540c\u65f6\u4ee5\u987a\u5e8f\u65b9\u5f0f\u8f93\u51fa\u7ffb\u8bd1\u7ed3\u679c\u7684\u8bcd\uff0c\u5e76\u4e14\u4e4b\u540e Yoshua Bengio \u5728 RNNSearch \u4e2d\u5c06\u5176\u5f15\u5165\u6210\u4e86\u4e00\u4e2a\u672f\u8bed\uff08\u611f\u8c22\u4e0a\u5e1d\uff1f:D\uff09\u3002\u540c\u6837\u6709\u8da3\u7684\u662f\uff0c\u8be5\u8bbe\u8ba1\u7684\u7075\u611f\u6765\u81ea\u4eba\u7c7b\u7684\u8ba4\u77e5\u8fc7\u7a0b\/\u7b56\u7565\uff0c\u5373\u6309\u987a\u5e8f\u6765\u56de\u5173\u6ce8\u4e00\u4e9b\u6570\u636e\u3002\u6700\u540e\uff0c\u4ece\u53d1\u5c55\u8fdb\u6b65\u7684\u672c\u8d28\u6765\u770b\uff0c\u8fd9\u4e2a\u6545\u4e8b\u76f8\u5f53\u6709\u8da3\u2014\u2014\u7c7b\u4f3c\u7684\u60f3\u6cd5\u548c\u8868\u8ff0\u300c\u65e9\u5c31\u5df2\u7ecf\u5728\u7a7a\u6c14\u4e2d\u56de\u8361\u300d\uff0c\u7279\u522b\u8981\u63d0\u5230\u5f53\u65f6 Alex Graves\uff08NMT\uff09\u548c Jason Weston\uff08\u8bb0\u5fc6\u7f51\u7edc\uff09\u7684\u5de5\u4f5c\u3002<\/section>\n
<\/section>\n
\u8c22\u8c22\u4f60\u7684\u6545\u4e8b @DBahdanau \uff01<\/section>\n
<\/section>\n
\u4e4b\u540e\uff0cKarpathy \u8fd8\u505a\u4e86\u4e00\u4e9b\u8865\u5145\uff1aChatGPT \u4ee5\u53ca\u7edd\u5927\u591a\u6570\u73b0\u4ee3 AI \u6a21\u578b\u90fd\u662f\u5de8\u578b Transformer\u3002\u300c\u6240\u4ee5 LLM \u7684\u6838\u5fc3\u795e\u5947\u4e4b\u5904\u6765\u81ea\u4e8e\u53cd\u590d\u5e94\u7528\u6ce8\u610f\u529b<\/strong>\uff0c\u4e00\u904d\u53c8\u4e00\u904d\u5730\u5173\u6ce8\u8f93\u5165\u7684 token\uff0c\u4ee5\u9884\u6d4b\u4e0b\u4e00\u4e2a token \u662f\u4ec0\u4e48\u3002\u300d<\/section>\n
<\/a><\/section>\n
Dzmitry Bahdanau \u7684\u539f\u59cb\u90ae\u4ef6\u5185\u5bb9<\/strong><\/section>\n
<\/section>\n
Karpathy \u4e5f\u4e00\u5e76\u5206\u4eab\u4e86 Dzmitry Bahdanau \u7684\u539f\u59cb\u90ae\u4ef6\u5185\u5bb9\uff1a<\/section>\n
<\/a><\/section>\n
\u55e8\uff0cAndrej\uff0c<\/section>\n
<\/section>\n
\u5f88\u9ad8\u5174\u544a\u8bc9\u4f60 8 \u5e74\u524d\u53d1\u751f\u7684\u6545\u4e8b\uff01<\/section>\n
<\/section>\n
\u6211\u5728\u96c5\u5404\u5e03\u5927\u5b66 Herbert Jaeger \u7684\u6307\u5bfc\u4e0b\u5b8c\u6210\u7855\u58eb\u8bfe\u7a0b\u7684\u7b2c\u4e00\u5e74\u540e\uff0c\u4f5c\u4e3a\u5b9e\u4e60\u751f\u6765\u5230\u4e86 Yoshua \u7684\u5b9e\u9a8c\u5ba4\u3002<\/section>\n
<\/section>\n
\u6211\u544a\u8bc9 Yoshua \u6211\u5f88\u4e50\u610f\u505a\u4efb\u4f55\u4e8b\u60c5\u3002Yoshua \u8ba9\u6211\u53c2\u4e0e\u673a\u5668\u7ffb\u8bd1<\/mark>\u9879\u76ee\uff0c\u4e0e Kyunghyun Cho \u548c\u56e2\u961f\u4e00\u8d77\u5de5\u4f5c\u3002\u6211\u5f53\u65f6\u975e\u5e38\u6000\u7591\u5c06\u8bcd\u5e8f\u5217\u585e\u5165\u5411\u91cf\u7684\u60f3\u6cd5\u3002\u4f46\u6211\u4e5f\u975e\u5e38\u60f3\u83b7\u5f97\u535a\u58eb\u5b66\u4f4d\u3002\u6240\u4ee5\u6211\u64b8\u8d77\u8896\u5b50\uff0c\u5f00\u59cb\u505a\u6211\u64c5\u957f\u7684\u4e8b\u60c5\u2014\u2014\u7f16\u5199\u4ee3\u7801\u3001\u4fee\u590d\u9519\u8bef\u7b49\u7b49\u3002\u5728\u67d0\u4e2a\u65f6\u5019\uff0c\u6211\u8868\u73b0\u5f97\u5f88\u4e86\u89e3\u6211\u505a\u7684\u4e1c\u897f\u4e86\uff0cYoshua \u9080\u8bf7\u6211\u653b\u8bfb\u535a\u58eb\u5b66\u4f4d\uff082014 \u5e74\u662f\u4e00\u4e2a\u5f88\u597d\u7684\u65f6\u673a\uff0c\u8868\u73b0\u5f97\u5f88\u4e86\u89e3\u5c31\u5df2\u7ecf\u8db3\u591f\u4e86\u2014\u2014\u7f8e\u597d\u7684\u65e7\u65f6\u5149\uff01\uff09\u3002\u6211\u975e\u5e38\u9ad8\u5174\uff0c\u6211\u8ba4\u4e3a\u53ef\u4ee5\u5f00\u59cb\u4eab\u53d7\u4e50\u8da3\u5e76\u53d1\u6325\u521b\u9020\u529b\u4e86\u3002<\/section>\n
<\/section>\n
\u6240\u4ee5\u6211\u5f00\u59cb\u601d\u8003\u5982\u4f55\u907f\u514d\u7f16\u7801\u5668\u548c\u89e3\u7801\u5668 RNN \u4e4b\u95f4\u7684\u74f6\u9888\u3002\u6211\u7684\u7b2c\u4e00\u4e2a\u60f3\u6cd5\u662f\u6784\u5efa\u4e00\u4e2a\u5e26\u6709\u4e24\u4e2a\u300c\u5149\u6807\u300d\u7684\u6a21\u578b\uff0c\u4e00\u4e2a\u5728\u6e90\u5e8f\u5217\u4e2d\u79fb\u52a8\uff08\u7531\u4e00\u4e2a BiRNN \u7f16\u7801\uff09\uff0c\u53e6\u4e00\u4e2a\u5728\u76ee\u6807\u5e8f\u5217\u4e2d\u79fb\u52a8\u3002\u4f7f\u7528\u52a8\u6001\u89c4\u5212<\/mark>\uff08dynamic programming\uff09\u53ef\u4ee5\u5c06\u5149\u6807\u8f68\u8ff9\u8fb9\u7f18\u5316\u3002KyungHyun Cho \u8ba4\u4e3a\u8fd9\u76f8\u5f53\u4e8e Alex Graves \u7684 RNN Transducer \u6a21\u578b\u3002\u4e4b\u540e\uff0c\u6211\u53ef\u80fd\u8fd8\u8bfb\u4e86 Graves \u7684\u624b\u5199\u8bc6\u522b\u8bba\u6587\u3002\u4e0d\u8fc7\uff0c\u8fd9\u79cd\u65b9\u6cd5\u770b\u8d77\u6765\u4e0d\u9002\u5408\u673a\u5668\u7ffb\u8bd1<\/mark>\u3002<\/section>\n
<\/section>\n
\u5728\u6211\u5b9e\u4e60\u7684\u5269\u4f59 5 \u5468\u5185\uff0c\u4e0a\u8ff0\u4f7f\u7528\u5149\u6807\u7684\u65b9\u6cd5\u5f88\u96be\u5b9e\u73b0\u3002\u6240\u4ee5\u6211\u5c1d\u8bd5\u4e86\u4e00\u79cd\u66f4\u7b80\u5355\u7684\u65b9\u6cd5\u2014\u2014\u4e24\u4e2a\u5149\u6807\u540c\u65f6\u540c\u6b65\u79fb\u52a8\uff08\u5b9e\u9645\u4e0a\u662f\u786c\u7f16\u7801\u7684\u5bf9\u89d2\u6ce8\u610f\u529b\uff09\u3002\u8fd9\u79cd\u65b9\u6cd5\u6709\u70b9\u6548\u679c\uff0c\u4f46\u65b9\u6cd5\u4e0d\u591f\u4f18\u96c5\u3002<\/section>\n
<\/section>\n
\u6240\u4ee5\u6709\u4e00\u5929\uff0c\u6211\u60f3\u5230\u5982\u679c\u80fd\u8ba9\u89e3\u7801\u5668 RNN \u5b66\u4f1a\u5728\u6e90\u5e8f\u5217\u4e2d\u641c\u7d22\u653e\u7f6e\u5149\u6807\u7684\u4f4d\u7f6e<\/strong>\u5c31\u597d\u4e86\u3002\u8fd9\u591a\u5c11\u53d7\u5230\u6211\u4e2d\u5b66\u65f6\u5b66\u4e60\u82f1\u8bed\u65f6\u7684\u7ffb\u8bd1\u7ec3\u4e60\u7684\u542f\u53d1\u3002\u7ffb\u8bd1\u65f6\uff0c\u4f60\u7684\u76ee\u5149\u4f1a\u5728\u6e90\u5e8f\u5217\u548c\u76ee\u6807\u5e8f\u5217\u4e4b\u95f4\u6765\u56de\u79fb\u52a8\u3002\u6211\u5c06\u8fd9\u79cd\u8f6f\u6027\u641c\u7d22\u8868\u793a\u4e3a softmax\uff0c\u7136\u540e\u5bf9 BiRNN \u72b6\u6001\u8fdb\u884c\u52a0\u6743\u5e73\u5747\u3002\u5b83\u7684\u6548\u679c\u5f88\u597d\uff0c\u4ece\u7b2c\u4e00\u6b21\u5c1d\u8bd5\uff0c\u5230\u540e\u6765\u632f\u594b\u4eba\u5fc3\u3002\u6211\u5c06\u8fd9\u4e2a\u67b6\u6784\u79f0\u4e3a RNNSearch\uff0c\u6211\u4eec\u6025\u4e8e\u53d1\u8868\u4e00\u7bc7 arXiv \u8bba\u6587\uff0c\u56e0\u4e3a\u6211\u4eec\u77e5\u9053\u8c37\u6b4c\u7684 Ilya \u548c\u540c\u4e8b\u9886\u5148\u4e8e\u6211\u4eec\uff0c\u4ed6\u4eec\u6709\u5de8\u5927\u7684 8 GPU LSTM \u6a21\u578b\uff08\u800c RNN Search \u4ecd\u5728 1 GPU \u4e0a\u8fd0\u884c\uff09\u3002<\/section>\n
<\/section>\n
\u540e\u6765\u53d1\u73b0\uff0c\u8fd9\u4e2a\u540d\u5b57\u5e76\u4e0d\u597d\u3002\u76f4\u5230\u6700\u540e\u51e0\u6b21\u8fc7\u8bba\u6587\u65f6\uff0cYoshua \u624d\u5c06\u66f4\u597d\u7684\u540d\u5b57\uff08attention\uff09\u6dfb\u52a0\u5230\u8bba\u6587\u7ed3\u8bba\u4e2d<\/strong>\u3002\u00a0<\/section>\n
<\/section>\n
\u4e00\u4e2a\u534a\u6708\u540e\uff0c\u6211\u4eec\u770b\u5230\u4e86 Alex Graves \u7684\u00a0NMT<\/strong> \u8bba\u6587\u3002\u8fd9\u786e\u5b9e\u662f\u5b8c\u5168\u76f8\u540c\u7684\u60f3\u6cd5\uff0c\u5c3d\u7ba1\u4ed6\u63d0\u51fa\u5b83\u7684\u52a8\u673a\u5b8c\u5168\u4e0d\u540c\u3002\u5728\u6211\u4eec\u7684\u60c5\u51b5\u4e0b\uff0c\u662f\u56e0\u4e3a\u9700\u8981\u800c\u4ea7\u751f\u4e86\u8fd9\u4e2a\u53d1\u660e\u3002\u5728\u4ed6\u7684\u60c5\u51b5\u4e0b\uff0c\u6211\u60f3\u5e94\u8be5\u662f\u5c06\u795e\u7ecf\u548c\u7b26\u53f7 AI \u8fde\u63a5\u8d77\u6765\u7684\u96c4\u5fc3\u5427\uff1fJason Weston \u53ca\u5176\u540c\u4e8b\u7684\u8bb0\u5fc6\u7f51\u7edc<\/strong>\u8bba\u6587\u4e5f\u91c7\u7528\u4e86\u7c7b\u4f3c\u7684\u673a\u5236\u3002<\/section>\n
<\/section>\n
\u6211\u6ca1\u6709\u8fdc\u89c1\u5730\u60f3\u5230\u6ce8\u610f\u529b\u53ef\u4ee5\u5728\u8f83\u4f4e\u7684\u5c42\u7ea7\u4f7f\u7528\uff0c\u4ee5\u4f5c\u4e3a\u8868\u5f81\u5b66\u4e60<\/mark>\u7684\u6838\u5fc3\u8fd0\u7b97\u3002\u4f46\u662f\u5f53\u6211\u770b\u5230 Transformer \u8bba\u6587\u65f6\uff0c\u6211\u7acb\u5373\u5411\u5b9e\u9a8c\u5ba4\u540c\u4e8b\u65ad\u8a00 RNN \u5df2\u6b7b\u3002<\/section>\n
<\/section>\n
\u56de\u5230\u4f60\u6700\u521d\u7684\u95ee\u9898\uff1a\u5728\u8499\u7279\u5229\u5c14 Yoshua \u7684\u5b9e\u9a8c\u5ba4\u53d1\u660e\u7684\u300c\u53ef\u5fae\u5206\u548c\u6570\u636e\u4f9d\u8d56\u52a0\u6743\u5e73\u5747\u300d\u72ec\u7acb\u4e8e\u795e\u7ecf\u56fe\u7075\u673a<\/mark>\u3001\u8bb0\u5fc6\u7f51\u7edc\u4ee5\u53ca 90 \u5e74\u4ee3\uff08\u751a\u81f3 70 \u5e74\u4ee3\uff09\u7684\u4e00\u4e9b\u76f8\u5173\u8ba4\u77e5\u79d1\u5b66\u8bba\u6587\u3002\u8fd9\u662f Yoshua \u9886\u5bfc\u63a8\u52a8\u5b9e\u9a8c\u5ba4\u8fdb\u884c\u96c4\u5fc3\u52c3\u52c3\u7684\u7814\u7a76\u7684\u7ed3\u679c\uff0cKyungHyun Cho \u5728\u8fd0\u884c\u4e00\u4e2a\u5927\u578b\u673a\u5668\u7ffb\u8bd1<\/mark>\u9879\u76ee\u65b9\u9762\u62e5\u6709\u9ad8\u8d85\u7684\u6280\u80fd\uff0c\u8be5\u9879\u76ee\u7531\u521d\u7ea7\u535a\u58eb\u751f\u548c\u5b9e\u4e60\u751f\u7ec4\u6210\uff1b\u6700\u540e\uff0c\u6211\u81ea\u5df1\u7684\u521b\u9020\u529b\u548c\u7f16\u7801\u6280\u80fd\u5728\u591a\u5e74\u7684\u7ade\u4e89\u6027\u7f16\u7a0b\u4e2d\u5f97\u5230\u4e86\u78e8\u7ec3\u3002\u4f46\u6211\u8ba4\u4e3a\u8fd9\u4e2a\u60f3\u6cd5\u5f88\u5feb\u5c31\u4f1a\u88ab\u53d1\u73b0\u3002\u5c31\u7b97\u6211\u3001Alex Graves \u548c\u8fd9\u4e2a\u6545\u4e8b\u4e2d\u7684\u5176\u4ed6\u89d2\u8272\u5f53\u65f6\u6ca1\u6709\u7814\u7a76\u6df1\u5ea6\u5b66\u4e60<\/mark>\u4e5f\u662f\u5982\u6b64\uff0c\u6ce8\u610f\u529b\u5c31\u662f\u6df1\u5ea6\u5b66\u4e60<\/mark>\u4e2d\u5b9e\u73b0\u7075\u6d3b\u7a7a\u95f4\u8fde\u63a5\u7684\u81ea\u7136\u65b9\u5f0f\u3002\u7b49\u5f85 GPU \u8db3\u591f\u5feb\uff0c\u8ba9\u4eba\u4eec\u6709\u52a8\u529b\u8ba4\u771f\u5bf9\u5f85\u6df1\u5ea6\u5b66\u4e60<\/mark>\u7814\u7a76\uff0c\u8fd9\u662f\u4e00\u4e2a\u663e\u800c\u6613\u89c1\u7684\u60f3\u6cd5\u3002\u81ea\u4ece\u6211\u610f\u8bc6\u5230\u8fd9\u4e00\u70b9\u4ee5\u6765\uff0c\u6211\u5728 AI \u9886\u57df\u7684\u62b1\u8d1f\u5c31\u662f\u542f\u52a8\u50cf\u673a\u5668\u7ffb\u8bd1<\/mark>\u9879\u76ee\u8fd9\u6837\u7684\u51fa\u8272\u7684\u5e94\u7528\u9879\u76ee\u3002\u76f8\u6bd4\u4e8e\u90a3\u4e9b\u7814\u7a76\u6240\u8c13\u7684\u300c\u771f\u6b63\u300dAI \u7684\u82b1\u54e8\u7406\u8bba\uff0c\u826f\u597d\u7684\u7814\u53d1\u5de5\u4f5c\u53ef\u4ee5\u4e3a\u57fa\u7840\u6280\u672f\u7684\u8fdb\u6b65\u505a\u51fa\u66f4\u5927\u8d21\u732e\u3002<\/section>\n
<\/section>\n
\u5c31\u8fd9\u4e9b\uff01\u6211\u975e\u5e38\u60f3\u66f4\u591a\u4e86\u89e3\u5173\u4e8e\u60a8\u7684\u6559\u80b2 AI \u9879\u76ee\u7684\u4fe1\u606f\uff08\u6211\u4ece Harm de Vries \u90a3\u91cc\u542c\u5230\u4e86\u4e00\u4e9b\u4f20\u8a00\uff1b)\uff09\u3002<\/section>\n
<\/section>\n
\u795d\u597d\uff0c<\/section>\n
Dima<\/section>\n
<\/section>\n
\u76f8\u5173\u8bba\u6587<\/strong><\/section>\n
<\/section>\n
\u6700\u540e\uff0c\u4e0b\u9762\u68b3\u7406\u4e86\u524d\u9762\u63d0\u53ca\u7684\u76f8\u5173\u8bba\u6587\uff0c\u4ee5\u5e2e\u52a9\u611f\u5174\u8da3\u7684\u8bfb\u8005\u505a\u8fdb\u4e00\u6b65\u7684\u6280\u672f\u63a2\u7d22\u3002<\/section>\n
<\/section>\n
Neural Machine Translation by Jointly Learning to Align and Translate<\/strong>\uff0c\u672a\u5f97\u5230\u5e94\u6709\u5173\u6ce8\u7684\u6ce8\u610f\u529b\u8bba\u6587<\/section>\n
\u4f5c\u8005\uff1aDzmitry Bahdanau, Kyunghyun Cho, Yoshua Bengio<\/section>\n
\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/1409.0473<\/section>\n
<\/section>\n
Attention is All You Need<\/strong>\uff0c\u8457\u540d\u7684 Transformer \u8bba\u6587<\/section>\n
\u4f5c\u8005\uff1aAshish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin<\/section>\n
\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/1706.03762<\/section>\n
<\/section>\n
Neural Turing Machines<\/strong>\uff0c\u5176\u4e2d\u51e0\u4e4e\u540c\u65f6\u63d0\u51fa\u4e86\u7c7b\u4f3c\u7684 soft pooling \u8fd0\u7b97<\/section>\n
\u4f5c\u8005\uff1aAlex Graves, Greg Wayne, Ivo Danihelka<\/section>\n
\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/1410.5401<\/section>\n
<\/section>\n
Generating Sequences With Recurrent Neural Networks<\/strong>\uff0c\u7814\u7a76\u4e86\u4f7f\u7528 LTSM \u5faa\u73af\u795e\u7ecf\u7f51\u7edc<\/mark>\u751f\u6210\u5177\u6709\u957f\u7a0b\u7ed3\u6784\u7684\u590d\u6742\u5e8f\u5217\uff0c\u5f88\u6709\u5f00\u521b\u6027<\/section>\n
\u4f5c\u8005\uff1aAlex Graves<\/section>\n
\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/1308.0850<\/section>\n
<\/section>\n
Memory Networks<\/strong>\uff0c\u8bb0\u5fc6\u7f51\u7edc<\/section>\n
\u4f5c\u8005\uff1aJason Weston, Sumit Chopra, Antoine Bordes<\/section>\n
\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/1410.3916<\/section>\n
<\/section>\n
Sequence to Sequence Learning with Neural Networks<\/strong>\uff0c\u6587\u4e2d\u63d0\u5230\u7684\u5f53\u65f6\u5728\u8c37\u6b4c\u7684 Ilya \u7b49\u4eba\u7684\u8bba\u6587<\/section>\n
\u4f5c\u8005\uff1aIlya Sutskever, Oriol Vinyals, Quoc V. Le<\/section>\n
\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/1409.3215<\/section>\n
<\/section>\n
\u53c2\u8003\u94fe\u63a5<\/sup><\/em><\/section>\n
https:\/\/x.com\/karpathy\/status\/1864023344435380613<\/em><\/sup><\/section>\n
https:\/\/x.com\/karpathy\/status\/1864030016457375916<\/sup><\/em><\/section>\n

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarp […]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[27],"tags":[65],"class_list":["post-36989","post","type-post","status-publish","format-standard","hentry","category-news","tag-65"],"yoast_head":"\n\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b - \u4e00\u8d77AI\u6280\u672f<\/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:\/\/17aitech.com\/?p=36989\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b\",\"datePublished\":\"2025-01-21T17:01:00+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989\"},\"wordCount\":307,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/frc-396a21b1d60ad06252f9daccc9ea7edd.png\",\"keywords\":[\"\u6df1\u5ea6\u5b66\u4e60\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=36989#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=36989\",\"name\":\"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/frc-396a21b1d60ad06252f9daccc9ea7edd.png\",\"datePublished\":\"2025-01-21T17:01:00+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=36989\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/frc-396a21b1d60ad06252f9daccc9ea7edd.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/frc-396a21b1d60ad06252f9daccc9ea7edd.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36989#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\",\"url\":\"https:\\\/\\\/17aitech.com\\\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/17aitech.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"url\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"contentUrl\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b - \u4e00\u8d77AI\u6280\u672f","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:\/\/17aitech.com\/?p=36989","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=36989#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=36989"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b","datePublished":"2025-01-21T17:01:00+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=36989"},"wordCount":307,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=36989#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/12\/frc-396a21b1d60ad06252f9daccc9ea7edd.png","keywords":["\u6df1\u5ea6\u5b66\u4e60"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=36989#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=36989","url":"https:\/\/17aitech.com\/?p=36989","name":"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=36989#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=36989#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/12\/frc-396a21b1d60ad06252f9daccc9ea7edd.png","datePublished":"2025-01-21T17:01:00+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=36989#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=36989"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=36989#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/12\/frc-396a21b1d60ad06252f9daccc9ea7edd.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/12\/frc-396a21b1d60ad06252f9daccc9ea7edd.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=36989#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u88ab\u5ffd\u7565\u7684\u8d77\u70b9\uff1fKarpathy\u63ed\u79d8\u6700\u521d\u7684\u6ce8\u610f\u529b\u8bba\u6587\u88abTransformer\u5149\u8292\u63a9\u76d6\u7684\u6545\u4e8b"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/36989","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=36989"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/36989\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=36989"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=36989"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=36989"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}