{"id":1158,"date":"2021-10-12T16:26:05","date_gmt":"2021-10-12T08:26:05","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=1158"},"modified":"2021-10-12T16:26:05","modified_gmt":"2021-10-12T08:26:05","slug":"%e5%9c%b0%e8%a1%a8%e6%9c%80%e5%bc%ba%ef%bc%81%e5%be%ae%e8%bd%af%e5%92%8c%e8%8b%b1%e4%bc%9f%e8%be%be%e6%8e%a8%e5%87%ba%e8%ae%ad%e7%bb%83%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8bmt-nlg%ef%bc%9a5300%e4%ba%bf","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2021\/10\/12\/1158\/","title":{"rendered":"\u5730\u8868\u6700\u5f3a\uff01\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u63a8\u51fa\u8bad\u7ec3\u8bed\u8a00\u6a21\u578bMT-NLG\uff1a5300\u4ebf\u53c2\u6570\u91cf\uff0c\u662fGPT-3\u76843\u500d"},"content":{"rendered":"<section><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-1159\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/uploads\/2021\/10\/4ffce04d92a4d6cb21c1494cdfcd6dc1-42.jpg\" width=\"1080\" height=\"720\" alt=\"\u56fe\u7247\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2021\/10\/4ffce04d92a4d6cb21c1494cdfcd6dc1-42.jpg 1080w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2021\/10\/4ffce04d92a4d6cb21c1494cdfcd6dc1-42-300x200.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2021\/10\/4ffce04d92a4d6cb21c1494cdfcd6dc1-42-1024x683.jpg 1024w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2021\/10\/4ffce04d92a4d6cb21c1494cdfcd6dc1-42-768x512.jpg 768w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/section>\n<section>\u6574\u7406 | \u6838\u5b50\u53ef\u4e50\u3001\u51cc\u654f<\/section>\n<section>OpenAI \u6700\u5f3a\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b GPT-3 \u5177\u6709 1750 \u4ebf\u4e2a\u53c2\u6570\uff0cMT-NLG \u53c2\u6570\u91cf\u662f\u524d\u8005\u7684 3 \u500d\u3002<\/section>\n<section>\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b MT-NLG \u95ee\u4e16<\/section>\n<p>10 \u6708 11 \u65e5\uff0c\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u6b63\u5f0f\u63a8\u51fa\u7531 DeepSpeed \u548c Megatron \u9a71\u52a8\u7684 Megatron-Turing \u81ea\u7136\u8bed\u8a00\u751f\u6210\u6a21\u578b\uff08MT-NLG\uff09\uff0c\u8fd9\u662f\u8fc4\u4eca\u4e3a\u6b62\u8bad\u7ec3\u7684\u6700\u5927\u6700\u5f3a\u7684\u89e3\u7801\u8bed\u8a00\u6a21\u578b\u3002\u636e\u4e86\u89e3\uff0c\u8be5\u6a21\u578b\u5177\u6709 5300 \u4ebf\u4e2a\u53c2\u6570\uff0c\u662f\u73b0\u6709\u6700\u5927\u7684\u6a21\u578b GPT-3 \u7684 3 \u500d\u3002<\/p>\n<p>\u4fe1\u606f\u663e\u793a\uff0c\u8fd1\u5e74\u6765\uff0c\u4f7f\u7528\u57fa\u4e8e Transformer \u8bed\u8a00\u6a21\u578b\u5b9e\u73b0\u7684\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u5df2\u7ecf\u5728\u5927\u89c4\u6a21\u8ba1\u7b97\u3001\u5927\u578b\u6570\u636e\u96c6\u4ee5\u53ca\u7528\u4e8e\u6a21\u578b\u8bad\u7ec3\u7684\u9ad8\u7ea7\u7b97\u6cd5\u4e0e\u8f6f\u4ef6\u7684\u5f3a\u52bf\u63a8\u52a8\u4e0b\uff0c\u8fce\u6765\u5feb\u901f\u53d1\u5c55\u3002\u6b64\u5916\uff0c\u6700\u5148\u8fdb\u7684 NLP \u6a21\u578b\u4e2d\u53c2\u6570\u7684\u6570\u91cf\u4e5f\u4fdd\u6301\u7740\u6307\u6570\u7ea7\u589e\u957f\u8d8b\u52bf\u3002<\/p>\n<section><img decoding=\"async\" class=\"img_loading\" src=\"data:image\/gif;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVQImWNgYGBgAAAABQABh6FO1AAAAABJRU5ErkJggg==\" alt=\"\u56fe\u7247\" crossorigin=\"anonymous\" data-ratio=\"0.6455078125\" data-src=\"https:\/\/mmbiz.qpic.cn\/mmbiz_jpg\/ZBjVrHIdkOmbz6bJicffQw8XI45UiblKMqicoNeSa96PrMibljnZNma8LrbXINovWHKsbfgQDnAfa24mSEoHtUS45A\/640?wx_fmt=jpeg\" data-type=\"jpeg\" data-w=\"1024\" \/>\u56fe\u4e00\uff1a\u6700\u5148\u8fdb NLP \u6a21\u578b\u968f\u65f6\u95f4\u63a8\u79fb\u7684\u89c4\u6a21\u53d8\u5316\u8d8b\u52bf<\/p>\n<\/section>\n<p>\u4e0d\u8fc7\uff0c\u8bad\u7ec3\u8fd9\u6837\u7684\u5927\u578b\u6a21\u578b\u5f80\u5f80\u6781\u5177\u6311\u6218\uff0c\u4e3b\u8981\u539f\u56e0\u5305\u62ec\u4ee5\u4e0b\u4e24\u70b9\uff1a<\/p>\n<p>\u5373\u4f7f\u662f\u6700\u5f3a\u5927\u7684 GPU \u4e5f\u65e0\u6cd5\u5728\u5185\u5b58\u4e2d\u76f4\u63a5\u62df\u5408\u8fd9\u4e9b\u6a21\u578b\u7684\u53c2\u6570\u3002\u5982\u679c\u4e0d\u9ad8\u5ea6\u5173\u6ce8\u7b97\u6cd5\u3001\u8f6f\u4ef6\u53ca\u786c\u4ef6\u5806\u6808\u7684\u4f18\u5316\uff0c\u90a3\u4e48\u4efb\u52a1\u4e2d\u6d89\u53ca\u7684\u5927\u91cf\u8ba1\u7b97\u64cd\u4f5c\u53ef\u80fd\u5c06\u8bad\u7ec3\u65f6\u957f\u63d0\u5347\u5230\u5931\u53bb\u5b9e\u9645\u610f\u4e49\u7684\u6c34\u5e73\u3002<\/p>\n<p>\u636e\u4ecb\u7ecd\uff0c\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u5c06\u6700\u5148\u8fdb\u7684 GPU \u52a0\u901f\u8bad\u7ec3\u57fa\u7840\u8bbe\u65bd\u4e0e\u5c16\u7aef\u5206\u5e03\u5f0f\u5b66\u4e60\u8f6f\u4ef6\u5806\u6808\u8fdb\u884c\u878d\u5408\uff0c\u5b9e\u73b0\u4e86\u6781\u9ad8\u7684\u8bad\u7ec3\u6548\u7387\u3002\u6b64\u5916\uff0c\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u8fd8\u4f7f\u5efa\u7acb\u4e86\u9ad8\u8d28\u91cf\u7684\u81ea\u7136\u8bed\u8a00\u8bad\u7ec3\u8bed\u6599\u5e93\uff0c\u5176\u4e2d\u5305\u542b\u6570\u5343\u4ebf\u4e2a\u6807\u8bb0\uff0c\u5e76\u5171\u540c\u5f00\u53d1\u4e86\u8bad\u7ec3\u914d\u65b9\uff0c\u4ee5\u63d0\u9ad8\u4f18\u5316\u6548\u7387\u548c\u7a33\u5b9a\u6027\u3002<\/p>\n<p>\u5728\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u8054\u5408\u53d1\u5e03\u7684\u4e00\u7bc7\u8bba\u6587\u4e2d\uff0c\u4e8c\u8005\u5bf9 MT-NLG \u8bad\u7ec3\u6d41\u7a0b\u4e2d\u7684\u65b9\u6cd5\u4e0e\u7ed3\u679c\u505a\u4e86\u8be6\u7ec6\u4ecb\u7ecd\u3002<\/p>\n<section>\u5927\u89c4\u6a21\u8bad\u7ec3\u57fa\u7840\u8bbe\u65bd<\/section>\n<p>\u5728\u82f1\u4f1f\u8fbe A100 Tensor Core GPU \u4e0e HDR InfiniBand \u7f51\u7edc\u7684\u652f\u6301\u4e0b\uff0c\u82f1\u4f1f\u8fbe Selene \u4e0e\u5fae\u8f6f Azure NDv4 \u7b49\u8d85\u7ea7\u8ba1\u7b97\u96c6\u7fa4\u5df2\u7ecf\u83b7\u5f97\u5145\u8db3\u7684\u7b97\u529b\uff0c\u80fd\u591f\u5728\u5408\u7406\u7684\u65f6\u95f4\u5468\u671f\u4e4b\u5185\u8bad\u7ec3\u51fa\u5305\u542b\u6570\u4e07\u4ebf\u4e2a\u53c2\u6570\u7684\u6a21\u578b\u3002\u7136\u800c\uff0c\u8981\u5145\u5206\u53d1\u6325\u8fd9\u4e9b\u8d85\u7ea7\u8ba1\u7b97\u673a\u7684\u6f5c\u529b\uff0c\u9700\u8981\u5728\u6570\u5343\u4e2a GPU \u4e4b\u95f4\u5b9e\u73b0\u826f\u597d\u7684\u5e76\u884c\u6027\uff0c\u540c\u65f6\u4fdd\u8bc1\u8d1f\u8f7d\u80fd\u591f\u5728\u5185\u5b58\u4e0e\u8ba1\u7b97\u5c42\u9762\u540c\u65f6\u83b7\u5f97\u9ad8\u6548\u6027\u4e0e\u53ef\u6269\u5c55\u6027\u3002\u9057\u61be\u7684\u662f\uff0c\u73b0\u6709\u5e76\u884c\u7b56\u7565\uff08\u4f8b\u5982\u6570\u636e\u3001\u7ba1\u9053\u6216\u5f20\u91cf\u5207\u7247\uff09\u5f80\u5f80\u5728\u5185\u5b58\u4e0e\u8ba1\u7b97\u6548\u7387\u4e4b\u95f4\u5b58\u5728\u51b2\u7a81\u4e0e\u53d6\u820d\uff0c\u65e0\u6cd5\u652f\u6491\u8d77\u5982\u6b64\u89c4\u6a21\u7684\u6a21\u578b\u8bad\u7ec3\u4efb\u52a1\u3002<\/p>\n<p>\u6570\u636e\u5e76\u884c\u80fd\u591f\u5e26\u6765\u826f\u597d\u7684\u8ba1\u7b97\u6548\u7387\uff0c\u4f46\u9700\u8981\u590d\u5236\u6a21\u578b\u72b6\u6001\u4e14\u65e0\u6cd5\u5229\u7528\u5206\u5e03\u5f0f\u5185\u5b58\u805a\u5408\u673a\u5236\u3002\u5f20\u91cf\u5207\u7247\u8981\u6c42\u5404 GPU \u4e4b\u95f4\u8fdb\u884c\u5927\u91cf\u901a\u4fe1\uff0c\u8fd9\u5c31\u9650\u5236\u4e86\u9ad8\u5e26\u5bbd NVLink \u652f\u6301\u4e0b\u7684\u5355\u8282\u70b9\u573a\u666f\u4e4b\u5916\u7684\u8ba1\u7b97\u6548\u7387\u3002\u7ba1\u9053\u5e76\u884c\u53ef\u4ee5\u5b9e\u73b0\u8de8\u8282\u70b9\u9ad8\u6548\u6269\u5c55\u3002\u4f46\u4e3a\u4e86\u63d0\u9ad8\u8ba1\u7b97\u6548\u7387\uff0c\u8fd9\u79cd\u65b9\u6cd5\u9700\u8981\u5927\u6279\u6b21\u3001\u7c97\u7c92\u5ea6\u7684\u5e76\u884c\u673a\u5236\u518d\u52a0\u4e0a\u5b8c\u7f8e\u7684\u8d1f\u8f7d\u5747\u8861\uff0c\u8fd9\u4e9b\u8981\u6c42\u5728 MT-NLG \u5e9e\u5927\u7684\u89c4\u6a21\u4e4b\u4e0b\u6839\u672c\u65e0\u6cd5\u5b9e\u73b0\u3002<\/p>\n<p>\u8f6f\u4ef6\u8bbe\u8ba1\u65b9\u9762\uff0c\u5728 DeepSpeed \u548c Megatron-LM \u7684\u9a71\u52a8\u4e0b\uff0c\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u521b\u5efa\u4e86\u4e00\u5957\u9ad8\u6548\u4e14\u53ef\u6269\u5c55\u7684 3D \u5e76\u884c\u7cfb\u7edf\uff0c\u80fd\u591f\u5c06\u6570\u636e\u3001\u7ba1\u9053\u4ee5\u53ca\u57fa\u4e8e\u5f20\u91cf\u5207\u7247\u7684\u5e76\u884c\u673a\u5236\u7ed3\u5408\u8d77\u6765\uff0c\u5171\u540c\u5e94\u5bf9\u6311\u6218\u3002<\/p>\n<p>\u5177\u4f53\u6765\u8bb2\uff0c\u8fd9\u5957\u7cfb\u7edf\u4f7f\u7528 Megatron-LM \u7684\u5f20\u91cf\u5207\u7247\u6269\u5c55\u8282\u70b9\u5185\u6a21\u578b\uff0c\u518d\u4f7f\u7528 DeepSpeed \u7684\u7ba1\u9053\u5e76\u884c\u5b9e\u73b0\u8de8\u8282\u70b9\u6a21\u578b\u6269\u5c55\u3002\u4f8b\u5982\uff0c\u5bf9\u4e8e 5300 \u4ebf\u53c2\u6570\u7684\u6a21\u578b\uff0c\u6bcf\u5957\u6a21\u578b\u526f\u672c\u90fd\u8de8\u8d8a 280 \u4e2a\u82f1\u4f1f\u8fbe A100 GPU\uff0c\u91c7\u7528\u7684\u5219\u662f\u8282\u70b9\u5185 8 \u8def\u5f20\u91cf\u5207\u7247\u4e0e\u8de8\u8282\u70b9 35 \u8def\u7ba1\u9053\u5e76\u884c\u3002\u63a5\u4e0b\u6765\uff0c\u4f7f\u7528 DeepSpeed \u7684\u6570\u636e\u5e76\u884c\u8fdb\u4e00\u6b65\u5c06\u8d1f\u8f7d\u6269\u5c55\u5230\u6570\u5343\u4e2a GPU\u3002<\/p>\n<p>\u786c\u4ef6\u7cfb\u7edf\u65b9\u9762\uff0c\u6a21\u578b\u8bad\u7ec3\u662f\u5728\u57fa\u4e8e\u82f1\u4f1f\u8fbe DGX SuperPOD \u7684 Selene \u8d85\u7ea7\u8ba1\u7b97\u673a\u4e0a\u4ee5\u6df7\u5408\u7cbe\u5ea6\u5b8c\u6210\u7684\u3002\u8fd9\u53f0\u8d85\u7ea7\u8ba1\u7b97\u673a\u7531 560 \u4e2a DGX A100 \u670d\u52a1\u5668\u6784\u5efa\u800c\u6210\uff0c\u91c7\u7528\u5b8c\u6574\u7684\u80d6\u6811\u62d3\u6251\uff08FatTree\uff09\u52a0 HDR InfiniBand \u8054\u7f51\u914d\u7f6e\u3002\u6bcf\u4e2a DGX A100 \u5305\u542b 8 \u4e2a\u82f1\u4f1f\u8fbe A100 80GB Tensor Core GPU\uff0c\u901a\u8fc7 NVLink \u4e0e NVSwitch \u5b9e\u73b0\u5b8c\u5168\u4e92\u8fde\u3002\u5fae\u8f6f\u4e5f\u5728 Azure NDv4 \u4e91\u8d85\u7ea7\u8ba1\u7b97\u673a\u4e2d\u91c7\u7528\u4e86\u7c7b\u4f3c\u7684\u53c2\u8003\u67b6\u6784\u3002<\/p>\n<p>\u7814\u7a76\u4eba\u5458\u5206\u522b\u5728 Selene \u4e0a\u4f7f\u7528 280\u3001350 \u4ee5\u53ca 420 \u53f0 DGX A100 \u670d\u52a1\u5668\u8bd5\u9a8c\u6279\u6b21\u5927\u5c0f\u4e3a 1920 \u7684 5300 \u4ebf\u53c2\u6570\u6a21\u578b\u7684\u7aef\u5230\u7aef\u541e\u5410\u91cf\uff0c\u5e76\u89c2\u5bdf\u5230\u5176\u8fed\u4ee3\u65f6\u95f4\u5206\u522b\u4e3a 60.1 \u79d2\u300150.2 \u79d2\u4ee5\u53ca 44.4 \u79d2\uff1b\u8fd9\u4e9b\u7ed3\u679c\u5404\u81ea\u5bf9\u5e94\u6bcf GPU \u6bcf\u79d2 126\u3001121 \u4ee5\u53ca 113 \u4e07\u4ebf\u6b21\u8ba1\u7b97\u3002<\/p>\n<section>\u8bad\u7ec3\u6570\u636e\u96c6\u4e0e\u6a21\u578b\u914d\u7f6e<\/section>\n<p>\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u7684\u7814\u7a76\u4eba\u5458\u4f7f\u7528 Transformer \u89e3\u7801\u5668\u67b6\u6784\uff0c\u8fd9\u662f\u4e00\u5957\u4ece\u5de6\u5230\u53f3\u751f\u6210\u7684\u57fa\u4e8e Transformer \u7684\u8bed\u8a00\u6a21\u578b\uff0c\u5171\u5305\u542b 5300 \u4ebf\u4e2a\u53c2\u6570\u3002\u5176\u5c42\u6570\u3001\u9690\u85cf\u7ef4\u5ea6\u4e0e\u6ce8\u610f\u529b\u5934\u6570\uff08attention heads\uff09\u5206\u522b\u4e3a 105\u300120480 \u4e0e 128\u3002<\/p>\n<p>\u7814\u7a76\u4eba\u5458\u4f7f\u7528 8 \u8def\u5f20\u91cf\u4e0e 35 \u8def\u7ba1\u9053\u5e76\u884c\uff0c\u5e8f\u5217\u957f\u5ea6\u4e3a 2048\uff0c\u5168\u5c40\u6279\u6b21\u5927\u5c0f\u4e3a 1920\u3002\u5728\u524d 120 \u4ebf\u4e2a\u8bad\u7ec3 token \u4e2d\uff0c\u4ee5 32 \u4e3a\u5355\u4f4d\u9010\u6b65\u589e\u52a0\u6279\u6b21\u5927\u5c0f\uff0c\u4ece 32 \u5f00\u59cb\u6162\u6162\u589e\u52a0\u81f3\u6700\u7ec8\u6279\u6b21\u5927\u5c0f 1920\u3002\u5171\u4f7f\u7528 10 \u4ebf\u4e2a token \u5bf9\u8bad\u7ec3\u5b66\u4e60\u7387\u8fdb\u884c\u9884\u70ed\u3002<\/p>\n<p>\u65b0\u7684\u8bad\u7ec3\u6570\u636e\u96c6\u4e3b\u8981\u57fa\u4e8e\u4e4b\u524d\u7684\u5de5\u4f5c\u6210\u679c The Pile \u6784\u5efa\u800c\u6210\u3002\u7814\u7a76\u4eba\u5458\u9996\u5148\u4ece The Pile \u5f53\u4e2d\u9009\u62e9\u8d28\u91cf\u76f8\u5bf9\u6700\u9ad8\u7684\u6570\u636e\u5b50\u96c6\uff08\u56fe\u4e8c\u4e2d\u7684\u524d 11 \u884c\uff09\uff0c\u4e4b\u540e\u91c7\u7528\u4e0e\u751f\u6210 Pile-CC \u7c7b\u4f3c\u7684\u65b9\u6cd5\u4e0b\u8f7d\u5e76\u8fc7\u6ee4\u4e24\u4efd\u6700\u8fd1\u7684 Common Crawl\uff08CC\uff09\u5feb\u7167\u3002\u5e76\u5bf9 CC \u6570\u636e\u6267\u884c\u4e86\u5206\u6b65\u64cd\u4f5c\uff0c\u5305\u62ec\u4ece\u539f\u59cb HTML \u6587\u4ef6\u4e2d\u63d0\u53d6\u6587\u672c\u3001\u5229\u7528\u5728\u9ad8\u8d28\u91cf\u6570\u636e\u4e0a\u8bad\u7ec3\u5b8c\u6210\u7684\u5206\u7c7b\u5668\u5bf9\u63d0\u53d6\u5230\u7684\u6587\u6863\u8fdb\u884c\u8bc4\u5206\uff0c\u518d\u6839\u636e\u8bc4\u5206\u8fc7\u6ee4\u6587\u6863\u7b49\u3002\u4e3a\u4e86\u4fdd\u8bc1\u8bad\u7ec3\u591a\u6837\u5316\uff0c\u7814\u7a76\u4eba\u5458\u8fd8\u4ece RealNews \u4e0e CC-Stories \u6570\u636e\u96c6\u4e2d\u8fdb\u884c\u4e86\u91c7\u6837\u3002<\/p>\n<p>\u5728\u6784\u5efa\u8bad\u7ec3\u6570\u636e\u96c6\u65f6\uff0c\u6587\u6863\u53bb\u91cd\u4e5f\u662f\u975e\u5e38\u91cd\u8981\u7684\u4e00\u73af\uff0c\u6bd5\u7adf\u4e0d\u540c\u6570\u636e\u96c6\u7684\u591a\u4e2a\u6587\u6863\u4e2d\u53ef\u80fd\u5305\u542b\u76f8\u540c\u7684\u5185\u5bb9\u3002\u7814\u7a76\u4eba\u5458\u4f7f\u7528 min-hash LSH \u5728\u6587\u6863\u5c42\u7ea7\u4f7f\u7528\u6a21\u7cca\u91cd\u590d\u6570\u636e\u5220\u9664\u6d41\u7a0b\uff0c\u7528\u4ee5\u8ba1\u7b97\u7a00\u758f\u6587\u6863\u56fe\u53ca\u5176\u4e2d\u7684\u8fde\u63a5\u90e8\u5206\uff0c\u7531\u6b64\u5feb\u901f\u8bc6\u522b\u51fa\u91cd\u590d\u6587\u6863\u3002\u53c8\u4ece\u5404\u8fde\u63a5\u90e8\u5206\u7684\u91cd\u590d\u6587\u6863\u4e2d\u9009\u5b9a\u4ee3\u8868\u6027\u6587\u6863\uff0c\u5e76\u6839\u636e\u6570\u636e\u96c6\u7684\u8d28\u91cf\u8fdb\u884c\u4f18\u5148\u7ea7\u6392\u5e8f\uff0c\u5e76\u4f7f\u7528\u57fa\u4e8e n-gram \u7684\u8fc7\u6ee4\u5668\u4ece\u8bad\u7ec3\u6570\u636e\u96c6\u4e2d\u5220\u9664\u4e0b\u6e38\u4efb\u52a1\u6570\u636e\u4ee5\u907f\u514d\u6c61\u67d3\u3002<\/p>\n<p>\u6700\u540e\uff0c\u7814\u7a76\u4eba\u5458\u7528\u4e00\u7ec4 15 \u5957\u6570\u636e\u96c6\u5b8c\u6210\u4e86\u8bad\u7ec3\uff0c\u5176\u4e2d\u5171\u5305\u542b 3390 \u4ebf\u4e2a token\u3002\u5728\u8bad\u7ec3\u671f\u95f4\uff0c\u6839\u636e\u56fe\u4e8c\u7ed9\u51fa\u7684\u53ef\u53d8\u91c7\u6837\u6743\u91cd\u5c06\u6570\u636e\u96c6\u6df7\u5408\u81f3\u5f02\u6784\u6279\u6b21\u5f53\u4e2d\uff0c\u91cd\u70b9\u653e\u5728\u8d28\u91cf\u66f4\u9ad8\u7684\u6570\u636e\u96c6\u8eab\u4e0a\u3002\u6700\u7ec8\uff0c\u6574\u4e2a\u6a21\u578b\u7684\u8bad\u7ec3\u4f7f\u7528\u5230 2700 \u4ebf\u4e2a token\u3002<\/p>\n<table cellspacing=\"0\" cellpadding=\"0\">\n<tbody>\n<tr>\n<td width=\"134\"><strong>\u6570\u636e\u96c6<\/strong><\/td>\n<td width=\"224\"><strong>\u6570\u636e\u96c6\u6765\u6e90<\/strong><\/td>\n<td width=\"170\"><strong>Token (<\/strong><strong>\u5355\u4f4d\u4e3a1<\/strong><strong>0<\/strong><strong>\u4ebf<\/strong><strong>)<\/strong><\/td>\n<td width=\"78\"><strong>\u6743\u91cd<\/strong><strong>\u00a0(%)<\/strong><\/td>\n<td width=\"44\"><strong>\u8f6e\u6b21<\/strong><\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>Books3<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">25.7<\/td>\n<td width=\"78\">14.3<\/td>\n<td width=\"44\">1.5<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>OpenWebText2<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">14.8<\/td>\n<td width=\"78\">19.3<\/td>\n<td width=\"44\">3.6<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>Stack Exchange<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">11.6<\/td>\n<td width=\"78\">5.7<\/td>\n<td width=\"44\">1.4<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>PubMed Abstracts<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">4.4<\/td>\n<td width=\"78\">2.9<\/td>\n<td width=\"44\">1.8<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>Wikipedia<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">4.2<\/td>\n<td width=\"78\">4.8<\/td>\n<td width=\"44\">3.2<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>Gutenberg (PG-19)<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">2.7<\/td>\n<td width=\"78\">0.9<\/td>\n<td width=\"44\">0.9<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>BookCorpus2<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">1.5<\/td>\n<td width=\"78\">1.0<\/td>\n<td width=\"44\">1.8<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>NIH\u00a0<\/strong><strong>ExPorter<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">0.3<\/td>\n<td width=\"78\">0.2<\/td>\n<td width=\"44\">1.8<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>Pile-CC<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">49.8<\/td>\n<td width=\"78\">9.4<\/td>\n<td width=\"44\">0.5<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>ArXiv<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">20.8<\/td>\n<td width=\"78\">1.4<\/td>\n<td width=\"44\">0.2<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>GitHub<\/strong><\/td>\n<td width=\"224\">Pile\u6570\u636e\u96c6<\/td>\n<td width=\"170\">24.3<\/td>\n<td width=\"78\">1.6<\/td>\n<td width=\"44\">0.2<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>CC-2020-50<\/strong><\/td>\n<td width=\"224\">Common Crawl (CC)\u5feb\u7167<\/td>\n<td width=\"170\">68.7<\/td>\n<td width=\"78\">13.0<\/td>\n<td width=\"44\">0.5<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>CC-2021-04<\/strong><\/td>\n<td width=\"224\">Common Crawl (CC)\u00a0\u5feb\u7167<\/td>\n<td width=\"170\">82.6<\/td>\n<td width=\"78\">15.7<\/td>\n<td width=\"44\">0.5<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>RealNews<\/strong><\/td>\n<td width=\"224\">RealNews<\/td>\n<td width=\"170\">21.9<\/td>\n<td width=\"78\">9.0<\/td>\n<td width=\"44\">1.1<\/td>\n<\/tr>\n<tr>\n<td width=\"134\"><strong>CC-Stories<\/strong><\/td>\n<td width=\"224\">Common Crawl (CC) stories<\/td>\n<td width=\"170\">5.3<\/td>\n<td width=\"78\">0.9<\/td>\n<td width=\"44\">0.5<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u56fe\u4e8c\uff1a\u7528\u4e8e\u8bad\u7ec3MT-NLG\u6a21\u578b\u7684\u6570\u636e\u96c6<\/p>\n<section>\u7ed3\u679c\u4e0e\u6210\u5c31<\/section>\n<p>\u8fd1\u671f\u5728\u8bed\u8a00\u6a21\u578b\uff08LM\uff09\u9886\u57df\u7684\u5de5\u4f5c\u8868\u660e\uff0c\u5f3a\u5927\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u5f80\u5f80\u80fd\u591f\u5728\u65e0\u9700\u5fae\u8c03\u7684\u60c5\u51b5\u4e0b\u5b9e\u73b0\u6781\u5177\u7ade\u4e89\u529b\u7684\u5e7f\u6cdb NLP \u4efb\u52a1\u5904\u7406\u80fd\u529b\u3002<\/p>\n<p>\u4e3a\u4e86\u7406\u89e3\u7eb5\u5411\u6269\u5c55\u8bed\u8a00\u6a21\u578b\u5728\u96f6\u6837\u672c\u6216\u5c11\u6837\u672c\u573a\u666f\u4e0b\u5b9e\u73b0\u7684\u5177\u4f53\u5b66\u4e60\u80fd\u529b\u589e\u5f3a\u6c34\u5e73\uff0c\u5fae\u8f6f\u548c\u82f1\u4f1f\u8fbe\u7684\u7814\u7a76\u4eba\u5458\u5bf9 MT-NLG \u5f00\u5c55\u8bc4\u4f30\uff0c\u5e76\u8bc1\u660e\u5b83\u5728\u591a\u79cd NLP \u4efb\u52a1\u4e2d\u5747\u83b7\u5f97\u4e86\u6700\u4f73\u7ed3\u679c\u3002\u4e3a\u4e86\u4fdd\u8bc1\u8bc4\u4f30\u7684\u5168\u9762\u6027\uff0c\u7814\u7a76\u4eba\u5458\u9009\u62e9\u4e86\u8de8\u4e94\u5927\u9886\u57df\u7684\u516b\u9879\u4e0d\u540c\u4efb\u52a1\uff1a<\/p>\n<p>\u5728\u6587\u672c\u9884\u6d4b\u4efb\u52a1 LAMBADA \u4e2d\uff0c\u6a21\u578b\u9884\u6d4b\u7279\u5b9a\u6bb5\u843d\u7684\u6700\u540e\u4e00\u4e2a\u5355\u8bcd\u3002\u5728\u9605\u8bfb\u7406\u89e3\u4efb\u52a1 RACE-h \u4e0e BooIQ \u4e2d\uff0c\u6a21\u578b\u6839\u636e\u7279\u5b9a\u6bb5\u843d\u751f\u6210\u95ee\u9898\u7684\u7b54\u6848\u3002\u5728\u5e38\u8bc6\u63a8\u7406\u4efb\u52a1 PiQA\u3001HellaSwag \u4e0e Winogrande \u4e2d\uff0c\u5404\u9879\u4efb\u52a1\u90fd\u9700\u8981\u4e00\u5b9a\u7a0b\u5ea6\u7684\u5e38\u8bc6\u6307\u5bfc\uff0c\u8d85\u51fa\u4e86\u4f20\u7edf\u8bed\u8a00\u7edf\u8ba1\u6a21\u5f0f\u7684\u89e3\u51b3\u8303\u7574\u3002\u5728\u81ea\u7136\u8bed\u8a00\u63a8\u7406\u65b9\u9762\uff0c\u91c7\u7528 ANLI-R2 \u4e0e HANS \u4e24\u9879\u8fc7\u5f80\u6a21\u578b\u666e\u904d\u65e0\u6cd5\u901a\u8fc7\u7684\u786c\u57fa\u51c6\u6d4b\u8bd5\u3002\u8bcd\u4e49\u6d88\u6b67\u4efb\u52a1 WiC\uff0c\u7ed3\u5408\u4e0a\u4e0b\u6587\u8bc4\u4f30\u591a\u4e49\u8bcd\u7406\u89e3\u3002<\/p>\n<p>\u4e3a\u4e86\u9a8c\u8bc1\u53ef\u91cd\u590d\u6027\uff0c\u7814\u7a76\u4eba\u5458\u8fd8\u4ee5\u5f00\u6e90\u9879\u76ee Im-evaluation-harness \u4e3a\u57fa\u7840\u5f00\u5c55\u8bc4\u4f30\u5e76\u505a\u51fa\u4e86\u9002\u5f53\u8c03\u6574\uff0c\u786e\u4fdd\u8bbe\u7f6e\u4e0e\u4e4b\u524d\u7684\u5de5\u4f5c\u8d1f\u8f7d\u4fdd\u6301\u9ad8\u5ea6\u4e00\u81f4\u3002\u5728\u96f6\u6837\u672c\u3001\u5355\u6837\u672c\u4e0e\u5c11\u6837\u672c\u7b49\u573a\u666f\u4e0b\u8bc4\u4f30\u4e86 MT-NLG\uff0c\u4e14\u6ca1\u6709\u6267\u884c\u6700\u4f73\u6837\u672c\u6570\u641c\u7d22\u3002<\/p>\n<p>\u56fe\u4e09\u4e3a\u51c6\u786e\u5ea6\u6307\u6807\u7ed3\u679c\u3002\u5982\u679c\u6d4b\u8bd5\u96c6\u4e3a\u516c\u5f00\u53ef\u7528\uff0c\u4f1a\u5728\u6d4b\u8bd5\u96c6\u4e0a\u76f4\u63a5\u8bc4\u4f30\uff1b\u5bf9\u4e8e\u975e\u516c\u5f00\u6d4b\u8bd5\u96c6\uff0c\u5219\u76f4\u63a5\u62a5\u544a\u4f7f\u7528\u5f00\u53d1\u6570\u636e\u96c6\u83b7\u5f97\u7684\u6570\u5b57\u3002\u8bc4\u4f30\u4e2d\u53ea\u6709 LAMBADA\u3001RACE-h \u4ee5\u53ca ANLI-R2 \u5c5e\u4e8e\u516c\u5f00\u6d4b\u8bd5\u96c6\uff0c\u5176\u4ed6\u5747\u5c5e\u4e8e\u5f00\u53d1\u6570\u636e\u96c6\u3002<\/p>\n<table cellspacing=\"0\" cellpadding=\"0\">\n<tbody>\n<tr>\n<td width=\"310\"><strong>\u7c7b\u522b<\/strong><\/td>\n<td width=\"90\"><strong>\u4efb\u52a1<\/strong><\/td>\n<td width=\"83\"><strong>\u96f6\u6837\u672c<\/strong><\/td>\n<td width=\"83\"><strong>\u5355\u6837\u672c<\/strong><\/td>\n<td width=\"83\"><strong>\u5c11\u6837\u672c<\/strong><\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u8865\u5168\u9884\u6d4b<\/td>\n<td width=\"90\"><strong>Lambada<\/strong><\/td>\n<td width=\"83\">0.766*<\/td>\n<td width=\"83\">0.731*<\/td>\n<td width=\"83\">0.872*<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u9605\u8bfb\u7406\u89e3<\/td>\n<td width=\"90\"><strong>BoolQ<\/strong><\/td>\n<td width=\"83\">0.782<\/td>\n<td width=\"83\">0.825<\/td>\n<td width=\"83\">0.848<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u9605\u8bfb\u7406\u89e3<\/td>\n<td width=\"90\"><strong>RACE-h<\/strong><\/td>\n<td width=\"83\">0.479<\/td>\n<td width=\"83\">0.484<\/td>\n<td width=\"83\">0.479<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u5e38\u8bc6\u63a8\u7406<\/td>\n<td width=\"90\"><strong>PiQA<\/strong><\/td>\n<td width=\"83\">0.820*<\/td>\n<td width=\"83\">0.810*<\/td>\n<td width=\"83\">0.832*<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u5e38\u8bc6\u63a8\u7406<\/td>\n<td width=\"90\"><strong>HellaSwag<\/strong><\/td>\n<td width=\"83\">0.802<\/td>\n<td width=\"83\">0.802<\/td>\n<td width=\"83\">0.824<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u5e38\u8bc6\u63a8\u7406<\/td>\n<td width=\"90\"><strong>WinoGrande<\/strong><\/td>\n<td width=\"83\">0.730<\/td>\n<td width=\"83\">0.737<\/td>\n<td width=\"83\">0.789<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u81ea\u7136\u8bed\u8a00\u63a8\u7406<\/td>\n<td width=\"90\"><strong>ANLI-R2<\/strong><\/td>\n<td width=\"83\">0.366<\/td>\n<td width=\"83\">0.397<\/td>\n<td width=\"83\">0.396<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u81ea\u7136\u8bed\u8a00\u63a8\u7406<\/td>\n<td width=\"90\"><strong>HANS<\/strong><\/td>\n<td width=\"83\">0.607<\/td>\n<td width=\"83\">0.649<\/td>\n<td width=\"83\">0.702<\/td>\n<\/tr>\n<tr>\n<td width=\"310\">\u8bcd\u4e49\u6d88\u6b67<\/td>\n<td width=\"90\"><strong>WiC<\/strong><\/td>\n<td width=\"83\">0.486<\/td>\n<td width=\"83\">0.513<\/td>\n<td width=\"83\">0.585<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u56fe\u4e09\uff1a\u4f7f\u7528NT-NLG\u8fdb\u884c\u96f6\u6837\u672c\u3001\u5355\u6837\u672c\u4e0e\u5c11\u6837\u672c\u8bc4\u4f30\u65f6\u83b7\u5f97\u7684\u51c6\u786e\u7387\u7ed3\u679c\u3002MT-NLG\u5728\u6240\u6709\u4e09\u79cd\u8bbe\u7f6e\uff08\u4ee5*\u8868\u793a\uff09\u4e2d\u5747\u5728PiQA\u5f00\u53d1\u96c6\u4e0eLAMBADA\u6d4b\u8bd5\u96c6\u4e0a\u83b7\u5f97\u4e86\u6700\u9ad8\u5f97\u5206\uff0c\u5e76\u5728\u5176\u4ed6\u7c7b\u522b\u7684\u540c\u7c7b\u5355\u4f53\u6a21\u578b\u4e2d\u540c\u6837\u62ff\u4e0b\u6700\u4f73\u7ed3\u679c\u3002<\/p>\n<p>\u4ee5\u5c11\u6837\u672c\u6027\u80fd\u4e3a\u4f8b\uff0c\u7814\u7a76\u4eba\u5458\u89c2\u5bdf\u5230 MT-NLG \u5177\u5907\u4e0d\u9519\u7684\u63d0\u5347\u6548\u679c\uff0c\u7279\u522b\u662f\u5728 WiC \u4e0e ANLI \u8fd9\u7c7b\u5bf9\u5148\u524d\u6a21\u578b\u6781\u5177\u6311\u6218\u7684\u53e5\u5b50\u95f4\u5173\u7cfb\u6bd4\u8f83\u6216\u8054\u7cfb\u7b49\u4efb\u52a1\u5f53\u4e2d\u3002\u6b64\u5916\uff0c\u5728\u96f6\u6837\u672c\u4e0e\u5355\u6837\u672c\u8bc4\u4f30\u4e2d\u540c\u6837\u770b\u5230\u4e86\u5e7f\u6cdb\u7684\u6027\u80fd\u63d0\u5347\u6548\u679c\u3002\u53e6\u5916\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u4e0e\u4ee5\u5f80\u6a21\u578b\u76f8\u6bd4\uff0cMT-NLG \u6a21\u578b\u5bf9\u8bad\u7ec3\u6570\u636e\u96c6\u4e2d\u7684\u6807\u8bb0\u4f9d\u8d56\u6027\u66f4\u4f4e\uff0c\u8fd9\u4ee3\u8868\u7740\u6a21\u578b\u89c4\u6a21\u8d8a\u5927\u3001\u5b66\u4e60\u901f\u5ea6\u53cd\u800c\u8d8a\u5feb\u3002<\/p>\n<p>\u5bf9\u4e8e HANS \u6570\u636e\u96c6\uff0c\u7814\u7a76\u4eba\u5458\u6ca1\u6709\u627e\u5230\u4efb\u4f55\u62a5\u544a\u6570\u636e\u96c6\u8303\u56f4\u6307\u6807\u7684\u57fa\u51c6\u3002\u4f46\u6839\u636e HANS \u8bba\u6587\u7684\u5206\u6790\uff0c\u5728 MNLI \u4e0a\u8bad\u7ec3\u7684 BERT \u57fa\u51c6\u80fd\u591f\u5728\u7ea6\u534a\u6570\u5b50\u7c7b\u4e0a\u83b7\u5f97\u8fd1\u4e4e\u5b8c\u7f8e\u7684\u7ed3\u679c\uff0c\u4f46\u5728\u53e6\u4e00\u534a\u4e0a\u5f97\u5206\u5374\u8d8b\u8fd1\u4e8e\u96f6\u3002\u8fd9\u8868\u660e\u539f\u6709\u6a21\u578b\u9ad8\u5ea6\u4f9d\u8d56\u4e8e\u8bba\u6587\u4e2d\u786e\u5b9a\u7684\u4f2a\u53e5\u6cd5\u542f\u53d1\u5f0f\u65b9\u6cd5\u3002<\/p>\n<p>\u5728\u96f6\u6837\u672c\u3001\u5355\u6837\u672c\u4e0e\u5c11\u6837\u672c\u573a\u666f\u4e2d\uff0cMT-NLG \u5728 PiQA \u5f00\u53d1\u96c6\u4e0e LAMBADA \u6d4b\u8bd5\u96c6\u4e0a\u90fd\u62ff\u4e0b\u4e86\u6700\u9ad8\u5206\u3002\u9664\u4e86\u62a5\u544a\u57fa\u51c6\u4efb\u52a1\u7684\u6c47\u603b\u6307\u6807\u4e4b\u5916\uff0c\u7814\u7a76\u4eba\u5458\u8fd8\u5bf9\u6a21\u578b\u8f93\u51fa\u8fdb\u884c\u4e86\u5b9a\u6027\u5206\u6790\uff0c\u5e76\u5f97\u51fa\u4e86\u66f4\u591a\u6709\u8da3\u53d1\u73b0\uff1a\u5373\u4f7f\u7b26\u53f7\u7ecf\u8fc7\u4e25\u91cd\u6df7\u6dc6\uff08\u793a\u4f8b 2\uff09\uff0c\u6a21\u578b\u4ecd\u7136\u53ef\u4ee5\u7ed3\u5408\u4e0a\u4e0b\u6587\u63a8\u7406\u51fa\u57fa\u7840\u6570\u5b66\u8fd0\u7b97\uff08\u793a\u4f8b 1\uff09\u3002\u867d\u7136\u8fd8\u4e0d\u80fd\u8bf4 MT-NLG \u771f\u6b63\u5177\u5907\u81ea\u8ff0\u80fd\u529b\uff0c\u4f46\u8be5\u6a21\u578b\u4f3c\u4e4e\u5df2\u7ecf\u8d85\u51fa\u4e86\u8bb0\u5fc6\u7b97\u672f\u7684\u8303\u7574\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u7814\u7a76\u4eba\u5458\u8fd8\u5c55\u793a\u4e86\u6765\u81ea HANS \u4efb\u52a1\u7684\u6837\u672c\u3002\u5728\u6b64\u4efb\u52a1\u4e2d\uff0c\u7814\u7a76\u4eba\u5458\u5c06\u5305\u542b\u7b80\u5355\u53e5\u6cd5\u7ed3\u6784\u7684\u4efb\u52a1\u4f5c\u4e3a\u95ee\u9898\uff0c\u63d0\u793a\u6a21\u578b\u7ed9\u51fa\u7b54\u6848\u3002\u5c3d\u7ba1\u7ed3\u6784\u975e\u5e38\u7b80\u5355\uff0c\u4f46\u73b0\u6709\u81ea\u7136\u8bed\u8a00\u63a8\u7406\uff08NLI\uff09\u6a21\u578b\u5f80\u5f80\u5f88\u96be\u5904\u7406\u6b64\u7c7b\u8f93\u5165\u3002\u5fae\u8c03\u6a21\u578b\u5219\u9700\u8981\u4ece NLI \u6570\u636e\u96c6\u4e2d\u7684\u7cfb\u7edf\u5ff5\u5934\u4e2d\u63d0\u53d6\u67d0\u4e9b\u53e5\u6cd5\u7ed3\u6784\u4e0e\u8574\u542b\u5173\u7cfb\u4e4b\u95f4\u7684\u4f2a\u5173\u8054\u3002\u4f46 MT-NLG \u65e0\u9700\u4efb\u4f55\u5fae\u8c03\uff0c\u5c31\u80fd\u5728\u8fd9\u7c7b\u573a\u666f\u4e0b\u5e26\u6765\u9887\u5177\u7ade\u4e89\u529b\u7684\u8868\u73b0\u3002<\/p>\n<table cellspacing=\"0\" cellpadding=\"0\">\n<tbody>\n<tr>\n<td width=\"144\"><strong>\u4efb\u52a1<\/strong><\/td>\n<td width=\"263\"><strong>\u793a\u4f8b1<\/strong><\/td>\n<td width=\"244\"><strong>\u793a\u4f8b2<\/strong><\/td>\n<\/tr>\n<tr>\n<td width=\"144\">\u63a8\u7406\u6570\u5b66\u8fd0\u7b97<\/td>\n<td width=\"263\"><strong>\u63d0\u793a<\/strong>:<\/p>\n<p>Q: 11 # 2 = ?<\/p>\n<p>A: 5.5<\/p>\n<p>Q: 18 # 9 = ?<\/p>\n<p>A: 2<\/p>\n<p>Q: 35 # 4 = ?<\/p>\n<p>A: 8.75<\/p>\n<p>Q: 27 # 9 = ?<\/p>\n<p>A: 3<\/p>\n<p>Q: 75 # 6 = ?<\/p>\n<p>A:<strong>MT-NLG<\/strong>:12.5<\/p>\n<p>Q: 45 # 5 = ?<\/p>\n<p>A: 9<\/p>\n<p>Q: 18 # 6 = ?<\/p>\n<p>A: 3<\/td>\n<td width=\"244\">\u63d0\u793a:<\/p>\n<p>Q: 11 $^% 2 = ?<\/p>\n<p>A: 22<\/p>\n<p>Q: 3 $^% 9\u00a0= ?<\/p>\n<p>A: 27<\/p>\n<p>Q: 12 $^% 4 = ?<\/p>\n<p>A: 48<\/p>\n<p>Q: 7 $^% 7 = ?<\/p>\n<p>A: 49<\/p>\n<p>Q: 8 $^% 5 = ?<\/p>\n<p>A:<strong>MT-NLG<\/strong>:40<\/p>\n<p>Q: 9 $^% 6 = ?<\/p>\n<p>A: 54<\/p>\n<p>Q: 10 $^% 5 = ?<\/p>\n<p>A: 50<\/td>\n<\/tr>\n<tr>\n<td width=\"144\">\u4e0d\u540c\u53e5\u6cd5\u7ed3\u6784\u7684\u81ea\u7136\u8bed\u8a00\u63a8\u7406<\/td>\n<td width=\"263\">\u63d0\u793a\uff1a<\/p>\n<p>\u94f6\u884c\u5bb6\u5fd8\u8bb0\u4e86\u6cd5\u5b98\u7ed9\u6559\u6388\u4eec\u63d0\u4f9b\u7684\u5efa\u8bae\u3002<\/p>\n<p>\u95ee\uff1a\u6cd5\u5b98\u7ed9\u6559\u6388\u4eec\u5efa\u8bae\u3002\u5bf9\u6216\u9519\uff1f\u7b54\u6848\uff1a<\/p>\n<p>MT-NLG\uff1a\u662f\u7684\u3002\u94f6\u884c\u5bb6\u5fd8\u8bb0\u4e86\u6cd5\u5b98\u7ed9\u6559\u6388\u4eec\u63d0\u4f9b\u4e86\u5efa\u8bae\u3002<\/td>\n<td width=\"244\">\u63d0\u793a\uff1a<\/p>\n<p>\u8457\u540d\u6559\u6388\u652f\u6301\u79d8\u4e66\u3002<\/p>\n<p>\u95ee\uff1a\u6559\u6388\u652f\u6301\u79d8\u4e66\u3002\u5bf9\u6216\u9519\uff1f<\/p>\n<p>\u7b54\u6848\uff1a<\/p>\n<p>MT-NLG\uff1a\u662f\u7684\u3002\u79d8\u4e66\u5f97\u5230\u4e86\u8457\u540d\u6559\u6388\u7684\u652f\u6301\u3002<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u56fe\u56db\uff1aNT-NLG\u5728\u4e0d\u540c\u53e5\u6cd5\u7ed3\u6784\u4e0b\u7684\u6570\u5b66\u8fd0\u7b97\u4e0e\u81ea\u7136\u8bed\u8a00\u63a8\u7406\u793a\u4f8b\u3002<\/p>\n<p>\u867d\u7136\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b\u786e\u5b9e\u63a8\u52a8\u7740\u8bed\u8a00\u751f\u6210\u6280\u672f\u7684\u4e0d\u65ad\u8fdb\u6b65\uff0c\u4f46\u5176\u4e2d\u4e5f\u5b58\u5728\u7740\u4e25\u91cd\u7684\u504f\u5dee \/ \u504f\u89c1\u4e0e\u6c61\u67d3\u95ee\u9898\u3002\u7814\u7a76\u4eba\u5458\u53d1\u73b0\uff0cMT-NLG \u6a21\u578b\u7ee7\u627f\u4e86\u5b83\u4ece\u8bad\u7ec3\u6570\u636e\u4e2d\u83b7\u5f97\u7684\u523b\u677f\u5370\u8c61\u4e0e\u504f\u89c1\uff0c\u5f53\u524d\uff0c\u5fae\u8f6f\u4e0e\u82f1\u4f1f\u8fbe\u6b63\u5728\u52aa\u529b\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u3002\u6b64\u5916\uff0c\u7814\u7a76\u4eba\u5458\u8868\u793a\uff0c\u5728\u751f\u4ea7\u573a\u666f\u4e2d\u4f7f\u7528 MT-NLG \u8fd8\u9700\u8981\u914d\u5408\u9002\u5f53\u63aa\u65bd\uff0c\u7528\u4ee5\u786e\u4fdd\u7f13\u89e3\u5e76\u6700\u5927\u7a0b\u5ea6\u964d\u4f4e\u5bf9\u7528\u6237\u9020\u6210\u7684\u6f5c\u5728\u4f24\u5bb3\u3002<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_1158\" class=\"pvc_stats total_only  \" data-element-id=\"1158\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u6574\u7406 | \u6838\u5b50\u53ef\u4e50\u3001\u51cc\u654f OpenAI \u6700\u5f3a\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b GPT-3 \u5177\u6709 1750 \u4ebf\u4e2a\u53c2\u6570\uff0cMT-NLG [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_1158\" class=\"pvc_stats total_only  \" data-element-id=\"1158\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":1,"featured_media":1159,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-1158","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/1158","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=1158"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/1158\/revisions"}],"predecessor-version":[{"id":1160,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/1158\/revisions\/1160"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/1159"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=1158"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=1158"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=1158"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}