{"id":35494,"date":"2024-12-26T01:00:01","date_gmt":"2024-12-25T17:00:01","guid":{"rendered":"https:\/\/17aitech.com\/?p=35494"},"modified":"2024-12-26T01:00:01","modified_gmt":"2024-12-25T17:00:01","slug":"%e5%a4%a7%e6%a8%a1%e5%9e%8b%e8%be%b9%e6%8e%a8%e7%90%86%e8%be%b9%e7%ba%a0%e9%94%99%ef%bc%8c%e6%9c%89%e5%8f%af%e8%83%bd%e5%81%9a%e5%88%b0%e5%90%97%ef%bc%9f%e8%bf%99%e6%98%aficml%e7%88%86%e7%81%ab","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=35494","title":{"rendered":"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2"},"content":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2<\/a><\/p>\n

<\/a><\/section>\n
\n
\n

AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/p>\n<\/section>\n<\/blockquote>\n

\u5373\u4fbf\u662f\u6700\u5f3a\u5927\u7684\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\uff0c\u4ecd\u4f1a\u5076\u5c14\u51fa\u73b0\u63a8\u7406\u9519\u8bef\u3002\u9664\u4e86\u901a\u8fc7\u63d0\u793a\u8bcd\u8ba9\u6a21\u578b\u8fdb\u884c\u4e0d\u592a\u53ef\u9760\u7684\u591a\u8f6e\u81ea\u6211\u7ea0\u9519\u5916\uff0c\u6709\u6ca1\u6709\u66f4\u7cfb\u7edf\u7684\u65b9\u6cd5\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\u5462\uff1f<\/p>\n

\u6765\u81ea Meta FAIR\u3001CMU \u548c MBZUAI \u7684\u53f6\u6dfb\u3001\u5f90\u5b50\u8bda\u3001\u674e\u8fdc\u5fd7\u3001\u6731\u6cfd\u56ed\u56e2\u961f\u5728\u6700\u65b0\u7684 arXiv \u8bba\u6587\u300a\u8bed\u8a00\u6a21\u578b\u7269\u7406\u5b66 Part 2.2\uff1a\u5982\u4f55\u4ece\u9519\u8bef\u4e2d\u5b66\u4e60\u300b\u4e2d\uff0c\u901a\u8fc7\u53ef\u63a7\u5b9e\u9a8c\uff0c\u63a2\u7d22\u4e86\u8ba9\u6a21\u578b\u300c\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\u300d<\/strong>\u7684\u53ef\u80fd\u6027\u3002<\/p>\n

\u4ed6\u4eec\u5728\u9884\u8bad\u7ec3\u4e2d\u52a0\u5165\u5927\u91cf\u300c\u9519\u8bef\u7684\u63a8\u7406\u300d\u548c\u300c\u9519\u8bef\u7684\u7ea0\u6b63\u300d\uff0c\u5c55\u793a\u4e86\u8fd9\u7c7b\u6570\u636e\u53ef\u4ee5\u63d0\u9ad8\u8bed\u8a00\u6a21\u578b\u7684\u63a8\u7406\u51c6\u786e\u6027\uff08\u65e0\u9700\u63d0\u793a\u8bcd\uff0c\u65e0\u9700\u591a\u8f6e\u5bf9\u8bdd\uff09\u3002\u6587\u7ae0\u8fd8\u6df1\u5165\u63a2\u8ba8\u4e86\u8bb8\u591a\u7ec6\u8282\uff0c\u4f8b\u5982\uff081\uff09\u8fd9\u79cd\u65b9\u6cd5\u4e0e beam search \u7684\u533a\u522b\uff0c\uff082\uff09\u5982\u4f55\u51c6\u5907\u6b64\u7c7b\u6570\u636e\uff0c\uff083\uff09\u662f\u5426\u9700\u8981\u5bf9\u9519\u8bef\u8fdb\u884c\u63a9\u7801\uff0c\uff084\uff09\u6240\u9700\u7684\u9519\u8bef\u6570\u91cf\uff0c\uff085\uff09\u6b64\u7c7b\u6570\u636e\u662f\u5426\u53ef\u7528\u4e8e\u5fae\u8c03\u7b49\u3002<\/p>\n

<\/a><\/section>\n

\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u56fe 1<\/sup><\/em><\/p>\n

\u4f5c\u8005\u9996\u5148\u5c55\u793a\u4e86\u4e00\u4e2a GPT-4o \u901a\u8fc7\u63d0\u793a\u8bcd\u548c\u591a\u8f6e\u5bf9\u8bdd\u8fdb\u884c\u7ea0\u9519\u7684\u793a\u4f8b\uff08\u56fe 2\uff09\uff0c\u53ef\u4ee5\u770b\u5230\u6210\u529f\u7387\u4e0d\u9ad8\uff0c\u800c\u4e14\u9700\u8981\u5f88\u957f\u7684\u5bf9\u8bdd\u624d\u80fd\u5b8c\u6210\u7ea0\u9519\u3002\u90a3\u4e48\uff0c\u5982\u679c\u6a21\u578b\u6700\u7ec8\u80fd\u7ea0\u9519\uff0c\u4e3a\u4ec0\u4e48\u4e0d\u5728\u7b2c\u4e00\u6b21\u72af\u9519\u65f6\u300c\u7acb\u5373\u6536\u56de\u5e76\u6539\u6b63\u300d\u5462\uff1f<\/p>\n

<\/a><\/section>\n

\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u56fe 2\uff1aGPT-4o \u901a\u8fc7\u63d0\u793a\u8bcd\u548c\u591a\u8f6e\u5bf9\u8bdd\u8fdb\u884c\u7ea0\u9519\u7684\u5b9e\u4f8b<\/sup><\/em><\/p>\n

\u4e3a\u6b64\uff0c\u4f5c\u8005\u4f7f\u7528\u63a2\u9488\uff08probing\uff09\u65b9\u6cd5\u7814\u7a76\u6a21\u578b\u7684\u5185\u90e8\u5de5\u4f5c\u673a\u5236\u3002\u901a\u8fc7 Part 2.1 \u5efa\u7acb\u7684 iGSM \u6570\u636e\u96c6\uff0c\u4f5c\u8005\u53d1\u73b0\u5f53\u6a21\u578b\u72af\u9519\u540e\uff0c\u5185\u90e8\u53c2\u6570\u5e38\u5e38\u8868\u73b0\u51fa\u300c\u5f88\u540e\u6094\u300d\u7684\u72b6\u6001\uff0c\u4e5f\u5c31\u662f\u8bf4\uff0c\u6a21\u578b\u53ef\u80fd\u5df2\u7ecf\u77e5\u9053\u81ea\u5df1\u72af\u4e86\u9519\uff0c\u4f46\u300c\u8986\u6c34\u96be\u6536\u300d\u3002<\/p>\n

\u90a3\u4e48\uff0c\u80fd\u5426\u7b80\u5355\u5730\u8ba9\u6a21\u578b\u300c\u540e\u6094\u5373\u91cd\u8bd5\uff08retry upon regret\uff09\u300d<\/strong>\uff1f\u5373\uff0c\u901a\u8fc7\u989d\u5916\u8bad\u7ec3\uff08\u5982\u5fae\u8c03\uff09\u5f97\u5230\u4e00\u4e2a\u68c0\u6d4b\u9519\u8bef\u7684\u6a21\u578b\uff0c\u53ea\u8981\u8be5\u6a21\u578b\u5224\u5b9a\u5f53\u524d\u6b65\u9aa4\u6709\u9519\uff0c\u5c31\u7acb\u5373\u9000\u683c\u56de\u5230\u4e0a\u4e00\u6b65\u9aa4\u7684\u672b\u5c3e\uff0c\u518d\u91cd\u65b0\u751f\u6210\u5462\uff1f<\/p>\n

\u5982\u56fe 3 \u6240\u793a\uff0c\u4f5c\u8005\u8fdb\u884c\u4e86\u6a2a\u5411\u5bf9\u6bd4\u3002\u5373\u4fbf\u9519\u8bef\u8bc6\u522b\u7387\u8d85\u8fc7 99%\uff0c\u8fd9\u79cd\u91cd\u8bd5\u65b9\u6cd5\u5728 iGSM \u6570\u636e\u96c6\u4e0a\u4e5f\u53ea\u80fd\u5c06\u63a8\u7406\u6b63\u786e\u7387\u63d0\u9ad8 2%\uff08\u867d\u7136\u6bd4 beam search \u597d\uff09\u3002\u4f5c\u8005\u603b\u7ed3\u4e86\u6b64\u65b9\u6cd5\u7684\u4e09\u4e2a\u4e0d\u8db3\u3002<\/p>\n

\u9996\u5148\uff0c\u5bf9\u6b63\u786e\u7387\u63d0\u9ad8\u6709\u9650\uff0c\u6bd5\u7adf\u9000\u683c\u540e\uff0c\u6a21\u578b\u4f9d\u7136\u662f\u968f\u673a\u751f\u6210\uff0c\u5e76\u6ca1\u6709\u7528\u9ad8\u7ea7\u7684\u65b9\u6cd5\u6539\u9519\u3002\u5176\u6b21\uff0c\u5bf9\u9519\u8bef\u8bc6\u522b\u7387\u7684\u8981\u6c42\u5f88\u9ad8\uff08\u540c\u7b49\u6761\u4ef6\u4e0b\uff0c\u9700\u8981 100% \u9519\u8bef\u8bc6\u522b\u7387\u624d\u80fd\u5c06\u63a8\u7406\u6b63\u786e\u7387\u63d0\u9ad8 8%\uff0c\u4f46\u8fd9\u592a\u4e0d\u73b0\u5b9e\uff09\u3002\u6700\u91cd\u8981\u7684\u662f\uff0c\u8fd9\u5e76\u4e0d\u80fd\u964d\u4f4e\u6a21\u578b\u751f\u6210\u6587\u672c\u7684\u65f6\u95f4\u590d\u6742\u5ea6\uff0c\u56e0\u4e3a\u4f9d\u7136\u9700\u8981\u4e00\u6b21\u6b21\u5730\u91cd\u65b0\u751f\u6210\u3002<\/p>\n

<\/a><\/section>\n

\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u56fe 3<\/sup><\/em><\/p>\n

\u63a5\u4e0b\u6765\uff0c\u4f5c\u8005\u66f4\u6362\u65b9\u6cd5\uff0c\u5728\u9884\u8bad\u7ec3\u6570\u636e\u4e2d\u52a0\u5165\u5927\u91cf\u7684\u9519\u8bef\u548c\u7ea0\u6b63\uff0c\u4f8b\u5982\u300cA=>B\uff0c\u54e6\u6211\u8bf4\u9519\u4e86\uff0c\u5e94\u8be5\u662f A=>C\u300d<\/strong>\u3002\u90a3\u4e48\uff0c\u8fd9\u80fd\u5426\u63d0\u5347\u6a21\u578b\u7684\u63a8\u7406\u6b63\u786e\u7387\u5462\uff1f\u4e4d\u4e00\u770b\uff0c\u8fd9\u4f3c\u4e4e\u4e0d\u5408\u7406\uff0c\u56e0\u4e3a\u589e\u52a0\u9519\u8bef\u7684\u540c\u65f6\uff0c\u6a21\u578b\u5c82\u4e0d\u662f\u88ab\u8feb\u5b66\u4e60\u8bf4\u9519\u8bef\u7684\u8bdd\uff08\u5373 A=>B\uff09\uff1f\u662f\u5426\u9700\u8981\u5c06\u9519\u8bef\u90e8\u5206\uff08\u8b6c\u5982\u300cA=>B\uff0c\u54e6\u6211\u8bf4\u9519\u4e86\uff0c\u5e94\u8be5\u662f\u300d\u8fd9\u51e0\u4e2a\u5b57\uff09\u901a\u8fc7\u63a9\u7801\uff08label masking\uff09\u4ece\u8bad\u7ec3\u6807\u7b7e\u4e2d\u5220\u9664\uff1f<\/p>\n

\u7b54\u6848\u662f\u4e0d\u9700\u8981\u3002\u4f9d\u7136\u901a\u8fc7 iGSM \u6570\u636e\u96c6\uff0c\u4f5c\u8005\u7528\u63a7\u5236\u53d8\u91cf\u6cd5\uff0c\u6a2a\u5411\u5bf9\u6bd4\u4e86\u8bf8\u591a\u53c2\u6570\u540e\u5f97\u51fa\u82e5\u5e72\u7ed3\u8bba\uff08\u56fe 4\uff09\u3002<\/p>\n

\u4f8b\u5982\uff0c\u5373\u4fbf\u9884\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u6bcf\u9053\u9898\u76ee\u6709 50% \u7684\u6b65\u9aa4\u5305\u542b\u9519\u8bef\uff0c\u6a21\u578b\u5728\u6d4b\u8bd5\u9636\u6bb5\u5e76\u4e0d\u4f1a\u523b\u610f\u72af\u9519\uff08\u5982\u4f7f\u7528 temp=0 \u751f\u6210\u65f6\uff09\u3002\u80cc\u540e\u7684\u539f\u56e0\u4e0e\u8bed\u8a00\u6a21\u578b\u5bf9\u8bed\u6cd5\u7684\u7ea0\u9519\u80fd\u529b\u6709\u5173\uff0c\u5177\u4f53\u53ef\u53c2\u89c1\u4f5c\u8005\u7684\u53e6\u4e00\u7bc7 Part 1 \u8bba\u6587\uff0c\u56e0\u6b64\u4e0d\u9700\u8981\u5bf9\u9519\u8bef\u8fdb\u884c\u63a9\u7801\u3002\u66f4\u795e\u5947\u7684\u662f\uff0c\u5728\u5408\u7406\u8303\u56f4\u5185\uff0c\u8bad\u7ec3\u96c6\u91cc\u7684\u9519\u8bef\u5176\u5b9e\u8d8a\u591a\u8d8a\u597d\uff0c\u4f8b\u5982\u5305\u542b 50% \u9519\u8bef\u7684\u6570\u636e\uff0c\u6bd4 10% \u9519\u8bef\u7684\u6570\u636e\u5728 iGSM \u6570\u636e\u96c6\u4e0a\u8fd8\u80fd\u518d\u63d0\u5347\u63a8\u7406\u6b63\u786e\u7387 4 \u4e2a\u767e\u5206\u70b9\u3002<\/p>\n

<\/a><\/section>\n

\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u56fe 4<\/sup><\/em><\/p>\n

\u63a5\u4e0b\u6765\uff0c\u4f5c\u8005\u7814\u7a76\u4e86\u5305\u542b\u300c\u9519\u8bef\u548c\u7ea0\u6b63\u300d\u7684\u6570\u636e\u80fd\u5426\u4f5c\u4e3a\u5fae\u8c03\u6570\u636e\u4f7f\u7528<\/strong>\u3002\u8fd9\u662f\u4e2a\u91cd\u8981\u95ee\u9898\uff0c\u56e0\u4e3a\u73b0\u6709\u7684\u5f00\u6e90\u5927\u6a21\u578b\u53ef\u80fd\u5e76\u4e0d\u5177\u5907\u5f88\u597d\u7684\u7ea0\u9519\u80fd\u529b\u3002\u5982\u679c\u6211\u4eec\u5236\u5907\u4e86\u5b8c\u7f8e\u7684\u9519\u8bef\u7ea0\u6b63\u6570\u636e\u96c6\uff0c\u80fd\u5426\u901a\u8fc7\u5c11\u91cf\u53c2\u6570\u5fae\u8c03\uff08\u5982\u4f7f\u7528 LoRA \u65b9\u6cd5\uff09\u8ba9\u73b0\u6709\u6a21\u578b\u5b66\u4f1a\u7ea0\u9519\uff1f<\/p>\n

\u7b54\u6848\u662f\u5426\u5b9a\u7684\u3002\u5982\u56fe 5 \u6240\u793a\uff0c\u4f5c\u8005\u5c1d\u8bd5\u4e86\u591a\u79cd LoRA \u53c2\u6570\uff0c\u53d1\u73b0\u6700\u591a\u53ea\u80fd\u5c06\u63a8\u7406\u6b63\u786e\u7387\u4ece 78% \u63d0\u9ad8\u5230 83%\u2014\u2014 \u751a\u81f3\u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\uff0c\u5982 LoRA \u7684 rank \u8f83\u5c0f\u65f6\uff0c\u6a21\u578b\u7684\u6b63\u786e\u7387\u8fdc\u4f4e\u4e8e 78%\u3002\u8fd9\u8bf4\u660e\u300c\u7ea0\u6b63\u9519\u8bef\u300d\u662f\u4e00\u4e2a\u9ad8\u7ea7\u80fd\u529b<\/strong>\uff0c\u4e0e\u6a21\u578b\u7684\u6b63\u5e38\u63a8\u7406\u4e0d\u540c\uff0c\u9700\u8981\u5927\u91cf\u53c2\u6570\u53d8\u5316\u624d\u80fd\u5b9e\u73b0\u3002\uff08\u8fd9\u4e5f\u5408\u7406\uff0c\u6bd5\u7adf\u5982\u679c\u4fee\u6539\u5c11\u91cf\u53c2\u6570\u5c31\u80fd\u5b8c\u6210\u7ea0\u9519\uff0c\u90a3\u4e48\u8ba9\u6a21\u578b\u300c\u540e\u6094\u5373\u91cd\u8bd5\uff08\u56fe 3\uff09\u300d\u6050\u6015\u65e9\u5c31\u80fd\u63d0\u9ad8\u63a8\u7406\u6b63\u786e\u7387\u4e86\u3002\uff09<\/p>\n

\u76f8\u5bf9\u800c\u8a00\uff0c\u300c\u9519\u8bef\u8bc6\u522b\u300d\u5e76\u4e0d\u662f\u9ad8\u7ea7\u80fd\u529b<\/strong>\uff0c\u53ef\u4ee5\u901a\u8fc7\u5fae\u91cf\u7684 LoRA \u5fae\u8c03\u5b66\u4f1a\u3002\u6b64\u5916\uff0c\u901a\u8fc7 beam search \u6a21\u578b\u4e5f\u80fd\u8fdb\u884c\u4e00\u5b9a\u7a0b\u5ea6\u7684\u91cd\u8bd5\uff0c\u4f46\u5bf9\u6b63\u786e\u7387\u7684\u63d0\u5347\u51e0\u4e4e\u4e3a\u96f6\u3002\u7efc\u5408\u4ee5\u4e0a\uff0c\u4f5c\u8005\u8ba4\u4e3a\uff0c\u5982\u679c\u80fd\u5236\u5907\u4f18\u8d28\u7684\u300c\u9519\u8bef\u548c\u7ea0\u6b63\u300d\u6570\u636e\uff0c\u5e94\u5c06\u6b64\u7c7b\u6570\u636e\u653e\u5165\u9884\u8bad\u7ec3\u6570\u636e\u96c6\u4e2d\uff0c\u800c\u4e0d\u662f\u7b49\u5230\u5fae\u8c03\u65f6\u518d\u4f7f\u7528\u3002<\/p>\n

<\/a><\/section>\n

\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u56fe 5<\/sup><\/em><\/p>\n

\u6700\u540e\uff0c\u4f5c\u8005\u7814\u7a76\u4e86\u5728\u5b9e\u9645\u751f\u6d3b\u4e2d\u5982\u4f55\u5236\u5907\u300c\u9519\u8bef\u548c\u7ea0\u6b63\u300d\u6570\u636e\u3002\u76ee\u524d\u4e3a\u6b62\uff0c\u6587\u7ae0\u90fd\u5728 iGSM \u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u53ef\u63a7\u5b9e\u9a8c\uff0c\u7531\u4e8e\u6b64\u6570\u636e\u96c6\u4e2d\u7684\u6570\u5b66\u9898\u6ee1\u8db3\u7edf\u4e00\u683c\u5f0f\uff0c\u53ef\u4ee5\u968f\u610f\u5220\u51cf\u62fc\u63a5\uff0c\u5236\u4f5c\u65e0\u9650\u91cf\u7684\u9519\u8bef\u548c\u7ea0\u6b63\u6570\u636e\u3002\u8fd9\u592a\u7406\u60f3\u5316\u4e86\u3002\u73b0\u5b9e\u751f\u6d3b\u4e2d\uff0c\u6709\u6ca1\u6709\u529e\u6cd5\u5728\u4e0d\u8981\u6c42\u7406\u89e3\u9898\u76ee\u7684\u57fa\u7840\u4e0a\u751f\u6210\u4e00\u4e9b\u300c\u5047\u9519\u8bef\u300d<\/strong>\uff1f<\/p>\n

\u4f5c\u8005\u5bf9\u6b64\u505a\u4e86\u4e00\u4e9b\u521d\u6b65\u5c1d\u8bd5\u3002\u4f8b\u5982\uff0c\u901a\u8fc7\u5c06\u89e3\u9898\u6b65\u9aa4\u4e2d\u9760\u540e\u7684\u7b2c Y \u6b65\u9aa4\u632a\u5230\u524d\u9762\u4f5c\u4e3a\u7b2c X \u6b65\u7684\u5047\u9519\u8bef\uff0c\u7136\u540e\u7528\u539f\u672c\u7684\u7b2c X \u6b65\u4f5c\u4e3a\u7ea0\u6b63\u3002\u8fd9\u4e00\u65b9\u6cd5\u5728 iGSM \u6570\u636e\u96c6\u4e0a\u4e5f\u80fd\u663e\u8457\u63d0\u5347\u6b63\u786e\u7387\uff08\u4ece 78% \u5230 91%\uff09\uff0c\u5982\u56fe 6 \u6240\u793a\u3002<\/p>\n

<\/a><\/section>\n

\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u56fe 6<\/sup><\/em><\/p>\n

\u636e\u6b64\uff0c\u4f5c\u8005\u5927\u80c6\u9884\u6d4b\uff0c\u5c3d\u7ba1\u672a\u6765\u7684 LLM \u53ef\u80fd\u4e0d\u4f1a\u76f4\u63a5\u5728 iGSM \u6570\u636e\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u4f46\u672c\u6587\u901a\u8fc7\u53ef\u63a7\u7684\u5bf9\u6bd4\u8bd5\u9a8c\uff0c\u7814\u7a76\u4e86\u5728\u901a\u5411 AGI \u7684\u9053\u8def\u4e0a\uff0c\u6211\u4eec\u9700\u8981\u5bf9\u6570\u636e\u8fdb\u884c\u54ea\u4e9b\u4fee\u6539\u548c\u5236\u5907\u3002<\/p>\n

\u4f8b\u5982\uff0c\u5229\u7528\u50cf Llama3-405B \u8fd9\u6837\u7684\u6a21\u578b\u6765\u6539\u5199\u6570\u5b66\u9898\uff0c\u5728\u6b63\u786e\u7684\u89e3\u9898\u6b65\u9aa4\u4e2d\u63d2\u5165\u8bb8\u591a\u9519\u8bef \u2014\u2014 \u751a\u81f3\u662f\u7b80\u5355\u7684\u5047\u9519\u8bef\uff0c\u4e5f\u6709\u671b\u6539\u53d8\u6a21\u578b\u7684\u7b54\u9898\u65b9\u5f0f\u3002\u8ba9\u6a21\u578b\u300c\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\u300d<\/strong>\uff0c\u800c\u4e0d\u662f\u901a\u8fc7\u989d\u5916\u7684\u63d0\u793a\u8bcd\u88ab\u52a8\u7ea0\u9519\uff0c\u6216\u8bb8\u662f\u4e00\u4e2a\u65b0\u7684\u601d\u8def\u3002\u4f5c\u8005\u9650\u4e8e GPU \u9650\u5236\uff0c\u65e0\u6cd5\u5bf9\u5982\u6b64\u65b9\u5411\u505a\u771f\u5b9e\u6570\u636e\u7684\u5927\u89c4\u6a21\u7814\u7a76\uff0c\u4f46\u6b22\u8fce\u8bfb\u8005\u6cbf\u7740\u8fd9\u4e00\u601d\u8def\u8bd5\u8bd5\u770b\u3002<\/p>\n

\u6700\u540e\uff0c\u8fd9\u7bc7 arXiv \u8bba\u6587\u662f\u300a\u8bed\u8a00\u6a21\u578b\u7269\u7406\u5b66\u300b\u7cfb\u5217\u4f5c\u54c1\u4e2d\u7684 Part 2.2\u3002\u6b64\u7cfb\u5217\u76ee\u524d\u5171 6 \u7bc7\u8bba\u6587\uff0c\u5728 ICML 2024 \u5927\u4f1a\u4e0a\u505a\u4e86 2 \u5c0f\u65f6\u7684\u6f14\u8bb2\uff0c\u6536\u83b7\u8bf8\u591a\u597d\u8bc4\uff08\u56fe 7\uff09\u3002<\/strong>\u6709\u5174\u8da3\u4e86\u89e3\u6574\u4e2a\u7cfb\u5217\u4f5c\u54c1\u7684\u5c0f\u4f19\u4f34\uff0c\u53ef\u4ee5\u79fb\u6b65 https:\/\/www.bilibili.com\/video\/BV1Yw4m1k7nH<\/p>\n

<\/a><\/section>\n

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709 […]<\/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":[68],"class_list":["post-35494","post","type-post","status-publish","format-standard","hentry","category-news","tag-68"],"yoast_head":"\n\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2 - \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=35494\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2\",\"datePublished\":\"2024-12-25T17:00:01+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494\"},\"wordCount\":83,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-e9822041001fcda34dc56c70577aafe6.png\",\"keywords\":[\"\u5927\u6a21\u578b\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=35494#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=35494\",\"name\":\"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-e9822041001fcda34dc56c70577aafe6.png\",\"datePublished\":\"2024-12-25T17:00:01+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=35494\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-e9822041001fcda34dc56c70577aafe6.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-e9822041001fcda34dc56c70577aafe6.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=35494#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2\"}]},{\"@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":"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2 - \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=35494","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=35494#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=35494"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2","datePublished":"2024-12-25T17:00:01+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=35494"},"wordCount":83,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=35494#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-e9822041001fcda34dc56c70577aafe6.png","keywords":["\u5927\u6a21\u578b"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=35494#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=35494","url":"https:\/\/17aitech.com\/?p=35494","name":"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=35494#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=35494#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-e9822041001fcda34dc56c70577aafe6.png","datePublished":"2024-12-25T17:00:01+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=35494#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=35494"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=35494#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-e9822041001fcda34dc56c70577aafe6.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-e9822041001fcda34dc56c70577aafe6.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=35494#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u5927\u6a21\u578b\u8fb9\u63a8\u7406\u8fb9\u7ea0\u9519\uff0c\u6709\u53ef\u80fd\u505a\u5230\u5417\uff1f\u8fd9\u662fICML\u7206\u706b\u7684\u6f14\u8bb2"}]},{"@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\/35494","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=35494"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/35494\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=35494"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=35494"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=35494"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}