{"id":37755,"date":"2025-02-03T23:00:54","date_gmt":"2025-02-03T15:00:54","guid":{"rendered":"https:\/\/17aitech.com\/?p=37755"},"modified":"2025-02-03T23:00:54","modified_gmt":"2025-02-03T15:00:54","slug":"transformer%c2%b2%e8%a6%81%e5%81%9a%e3%80%8c%e6%b4%bb%e3%80%8d%e7%9a%84ai%e6%a8%a1%e5%9e%8b%ef%bc%8c%e5%8a%a8%e6%80%81%e8%b0%83%e6%95%b4%e6%9d%83%e9%87%8d%ef%bc%8c%e5%83%8f%e7%ab%a0%e9%b1%bc%e4%b8%80","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=37755","title":{"rendered":"Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-01-15-11\" target=\"_blank\">Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883<\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"67\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<section>\u81ea\u9002\u5e94 LLM \u53cd\u6620\u4e86\u795e\u7ecf\u79d1\u5b66\u548c\u8ba1\u7b97\u751f\u7269\u5b66\u4e2d\u4e00\u4e2a\u516c\u8ba4\u7684\u539f\u7406\uff0c\u5373\u5927\u8111\u6839\u636e\u5f53\u524d\u4efb\u52a1\u6fc0\u6d3b\u7279\u5b9a\u533a\u57df\uff0c\u5e76\u52a8\u6001\u91cd\u7ec4\u5176\u529f\u80fd\u7f51\u7edc\u4ee5\u54cd\u5e94\u4e0d\u65ad\u53d8\u5316\u7684\u4efb\u52a1\u9700\u6c42\u3002<\/section>\n<\/section>\n<\/blockquote>\n<section><\/section>\n<section>\u5728\u81ea\u7136\u754c\uff0c\u300c\u9002\u5e94\u300d\u662f\u4e00\u79cd\u975e\u5e38\u666e\u904d\u7684\u73b0\u8c61\u3002\u4f8b\u5982\uff0c\u7ae0\u9c7c\u80fd\u591f\u8fc5\u901f\u6539\u53d8\u81ea\u8eab\u7684\u80a4\u8272\u548c\u7eb9\u7406\uff0c\u4ee5\u878d\u5165\u5468\u56f4\u73af\u5883\uff0c\u4ece\u800c\u8eb2\u907f\u5929\u654c\u548c\u6355\u6349\u730e\u7269\uff1b\u4eba\u8111\u5728\u53d7\u4f24\u540e\u80fd\u591f\u91cd\u65b0\u8fde\u63a5\u81ea\u8eab\u795e\u7ecf\u56de\u8def\uff0c\u4f7f\u4e2a\u4f53\u80fd\u591f\u6062\u590d\u5931\u53bb\u7684\u529f\u80fd\u5e76\u9002\u5e94\u65b0\u7684\u601d\u7ef4\u65b9\u5f0f\u6216\u884c\u52a8\u65b9\u5f0f\u3002\u751f\u7269\u4f53\u5c55\u73b0\u51fa\u7684\u9002\u5e94\u80fd\u529b\u4f7f\u5f97\u751f\u547d\u80fd\u591f\u5728\u4e0d\u65ad\u53d8\u5316\u7684\u73af\u5883\u4e2d\u84ec\u52c3\u53d1\u5c55\u3002<\/section>\n<section><\/section>\n<section>\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57df\uff0c\u9002\u5e94\u7684\u6982\u5ff5\u540c\u6837\u5177\u6709\u5de8\u5927\u7684\u5438\u5f15\u529b\u3002\u60f3\u8c61\u4e00\u4e2a\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\uff0c\u5b83\u80fd\u591f\u52a8\u6001\u5730\u8c03\u6574\u81ea\u8eab\u7684\u6743\u91cd\u4ee5\u5728\u964c\u751f\u7684\u73af\u5883\u4e2d\u4e0d\u65ad\u5b66\u4e60\u3001\u8fdb\u5316\u3002\u4e0e\u90e8\u7f72\u5728\u73af\u5883\u4e2d\u7684\u9759\u6001 AI \u6a21\u578b\u76f8\u6bd4\uff0c\u8fd9\u79cd\u6709\u81ea\u9002\u5e94\u80fd\u529b\u7684\u6a21\u578b\u660e\u663e\u5b66\u4e60\u6548\u7387\u66f4\u9ad8\uff0c\u800c\u4e14\u6709\u671b\u6210\u4e3a\u4e0e\u73b0\u5b9e\u4e16\u754c\u52a8\u6001\u672c\u8d28\u59cb\u7ec8\u4fdd\u6301\u4e00\u81f4\u7684\u7ec8\u751f\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section>\u65e5\u672c AI \u521d\u521b\u516c\u53f8 Sakana AI \u7684\u4e00\u9879\u6210\u679c\u5c31\u662f\u5bf9\u8fd9\u4e00\u65b9\u5411\u7684\u63a2\u7d22\u3002\u5728\u8bba\u6587\u4e2d\uff0c\u4ed6\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u53ef\u4ee5\u6839\u636e\u4e0d\u540c\u4efb\u52a1\u52a8\u6001\u8c03\u6574\u6a21\u578b\u6743\u91cd\u7684\u673a\u5668\u5b66\u4e60\u7cfb\u7edf \u2014\u2014Transformer^2\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png\"><\/a><\/section>\n<section>Transformer^2 \u8fd9\u4e2a\u540d\u79f0\u53cd\u6620\u4e86\u5b83\u7684\u4e24\u6b65\u8fc7\u7a0b\uff1a\u9996\u5148\uff0c\u6a21\u578b\u5206\u6790\u4f20\u5165\u7684\u4efb\u52a1\u4ee5\u4e86\u89e3\u5176\u8981\u6c42\uff0c\u7136\u540e\u5e94\u7528\u7279\u5b9a\u4e8e\u4efb\u52a1\u7684\u8c03\u6574\u6765\u751f\u6210\u6700\u4f73\u7ed3\u679c\u3002\u901a\u8fc7\u6709\u9009\u62e9\u5730\u8c03\u6574\u6a21\u578b\u6743\u91cd\u7684\u5173\u952e\u7ec4\u6210\u90e8\u5206\uff0c\u8be5\u6846\u67b6\u5141\u8bb8 LLM \u5b9e\u65f6\u52a8\u6001\u5730\u9002\u5e94\u65b0\u4efb\u52a1\u3002<\/section>\n<section><\/section>\n<section>Transformer^2 \u5728\u5404\u79cd\u4efb\u52a1\uff08\u4f8b\u5982\u6570\u5b66\u3001\u7f16\u7a0b\u3001\u63a8\u7406\u548c\u89c6\u89c9\u7406\u89e3\uff09\u4e0a\u5c55\u793a\u4e86\u663e\u8457\u8fdb\u6b65\uff0c\u5728\u6548\u7387\u548c\u7279\u5b9a\u4e8e\u4efb\u52a1\u7684\u6027\u80fd\u65b9\u9762\u4f18\u4e8e LoRA \u7b49\u4f20\u7edf\u9759\u6001\u65b9\u6cd5\uff0c\u540c\u65f6\u9700\u8981\u7684\u53c2\u6570\u5c11\u5f97\u591a\u3002<\/section>\n<section><\/section>\n<section>\u4f5c\u8005\u8868\u793a\uff0c\u8fd9\u9879\u7814\u7a76\u4e3a\u4eba\u4eec\u63d0\u4f9b\u4e86\u4e00\u4e2a\u672a\u6765 AI \u6a21\u578b\u4e0d\u518d\u9759\u6001\u7684\u521d\u6b65\u5c55\u671b\u3002\u8fd9\u4e9b\u7cfb\u7edf\u5c06\u5728\u6d4b\u8bd5\u65f6\u52a8\u6001\u5730\u8c03\u6574\u5176\u8ba1\u7b97\u80fd\u529b\uff0c\u4ee5\u9002\u5e94\u5b83\u4eec\u6240\u9047\u5230\u7684\u4efb\u52a1\u7684\u590d\u6742\u6027\uff0c\u4f53\u73b0\u51fa\u80fd\u591f\u6301\u7eed\u53d8\u5316\u548c\u7ec8\u751f\u5b66\u4e60\u7684\u300c\u6d3b\u300d\u7684\u667a\u80fd\u3002<\/section>\n<section><\/section>\n<section>\u6709\u4eba\u5c31\u6b64\u5c55\u671b\u8bf4\uff0c\u300c\u672a\u6765\uff0c\u300e\u9884\u8bad\u7ec3\u300f\u548c\u300e\u540e\u8bad\u7ec3\u300f\u4e4b\u95f4\u7684\u754c\u9650\u5c06\u4f1a\u6d88\u5931\uff0c\u6211\u4eec\u7684\u6a21\u578b\u548c\u667a\u80fd\u4f53\u5c06\u4e0d\u65ad\u9002\u5e94\u548c\u81ea\u6211\u6539\u8fdb\u3002\u50cf\u8fd9\u6837\u7684\u7cfb\u7edf\u5c06\u4e3a\u65b0\u4e00\u4ee3\u81ea\u9002\u5e94\u4eba\u5de5\u667a\u80fd\u94fa\u5e73\u9053\u8def\uff0c\u8fd9\u79cd\u4eba\u5de5\u667a\u80fd\u80fd\u591f\u4fee\u6539\u81ea\u8eab\u7684\u6743\u91cd\u548c\u67b6\u6784\uff0c\u4ee5\u9002\u5e94\u5b83\u4eec\u5728\u73af\u5883\u4e2d\u9047\u5230\u7684\u4efb\u52a1\u4e0d\u65ad\u53d8\u5316\u7684\u672c\u8d28\u3002\u300d<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-3b099dd8b8bb08219f9fc0d6307352bc.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-3b099dd8b8bb08219f9fc0d6307352bc.png\"><\/a><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aTRANSFORMER2 : SELF-ADAPTIVE LLMS<\/section>\n<\/li>\n<li>\n<section>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2501.06252<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u8fd9\u7bc7\u8bba\u6587\u5171\u6709\u4e09\u4f4d\u5171\u540c\u4e00\u4f5c\uff0c\u5176\u4e2d\u4e24\u4f4d\u662f\u534e\u4eba\u3002Qi Sun \u5728\u4e1c\u4eac\u5de5\u4e1a\u5927\u5b66\u62c5\u4efb\u7814\u7a76\u52a9\u7406\uff0c\u540c\u65f6\u5728 Sakana AI \u517c\u804c\uff0c\u7814\u7a76\u65b9\u5411\u662f\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u8bad\u7ec3\u4e0e\u8bc4\u4f30\u3001\u5927\u89c4\u6a21\u5408\u6210\u6570\u636e\u751f\u6210\u7b49\u3002Yujin Tang \u66fe\u5728\u8c37\u6b4c\u5de5\u4f5c\u591a\u5e74\uff0c\u73b0\u5728\u662f Sakana AI \u7684\u7814\u7a76\u79d1\u5b66\u5bb6\uff0c\u7814\u7a76\u65b9\u5411\u662f\u5f3a\u5316\u5b66\u4e60\u548c\u673a\u5668\u4eba\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-110a0a1a29bae8986a91aafff3695417.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-110a0a1a29bae8986a91aafff3695417.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-3e807742960aaef8b122fc368d864cb4.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-3e807742960aaef8b122fc368d864cb4.png\"><\/a><\/section>\n<section><strong>\u8bba\u6587\u6982\u89c8<\/strong><\/section>\n<section><\/section>\n<section>\u81ea\u9002\u5e94\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u5c06\u4ee3\u8868 AI \u9886\u57df\u7684\u4e00\u4e2a\u91cd\u8981\u8fdb\u5c55\uff0c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u4f7f\u6a21\u578b\u80fd\u591f\u5b9e\u65f6\u9002\u5e94\u4e0d\u540c\u4efb\u52a1\u548c\u52a8\u6001\u73af\u5883\u7684\u6846\u67b6\u3002\u867d\u7136\u7ec4\u5408\u6027\u548c\u53ef\u6269\u5c55\u6027\u5bf9\u4e8e\u6709\u6548\u9002\u5e94\u81f3\u5173\u91cd\u8981\uff0c\u4f46\u5f53\u524d\u7684 LLM \u8bad\u7ec3\u65b9\u6cd5\u96be\u4ee5\u540c\u65f6\u5b9e\u73b0\u8fd9\u4e24\u4e2a\u7279\u6027\u3002Sakana AI \u7684\u7814\u7a76\u65e8\u5728\u63d0\u51fa\u4e00\u4e2a\u5f00\u521b\u6027\u7684\u89e3\u51b3\u65b9\u6848\u6765\u5b9e\u73b0\u8fd9\u4e00\u613f\u666f\u5e76\u89e3\u51b3\u8fd9\u4e9b gap\u3002<\/section>\n<section><\/section>\n<section>\u4f20\u7edf\u4e0a\uff0cLLM \u540e\u8bad\u7ec3\u8bd5\u56fe\u5728\u5355\u6b21\u5927\u89c4\u6a21\u8bad\u7ec3\u4e2d\u4f18\u5316\u6a21\u578b\u7684\u5e7f\u6cdb\u80fd\u529b\u3002\u867d\u7136\u8fd9\u79cd\u300c\u4e00\u6b21\u6027\u300d\u5fae\u8c03\u6846\u67b6\u4ece\u7b80\u5355\u6027\u7684\u89d2\u5ea6\u6765\u770b\u662f\u7406\u60f3\u7684\uff0c\u4f46\u5728\u5b9e\u8df5\u4e2d\u5f88\u96be\u5b9e\u73b0\u3002\u4f8b\u5982\uff0c\u540e\u8bad\u7ec3\u4ecd\u7136\u975e\u5e38\u6d88\u8017\u8d44\u6e90\uff0c\u5bfc\u81f4\u5de8\u5927\u7684\u8ba1\u7b97\u6210\u672c\u548c\u8d85\u957f\u7684\u8bad\u7ec3\u65f6\u95f4\u3002\u6b64\u5916\uff0c\u5728\u5f15\u5165\u989d\u5916\u6570\u636e\u5e7f\u5ea6\u65f6\u5f80\u5f80\u5b58\u5728\u660e\u663e\u7684\u6027\u80fd\u6743\u8861\uff0c\u8fd9\u4f7f\u5f97\u540c\u65f6\u514b\u670d\u8fc7\u62df\u5408\u548c\u4efb\u52a1\u5e72\u6270\u53d8\u5f97\u5177\u6709\u6311\u6218\u6027\u3002<\/section>\n<section><\/section>\n<section>\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u81ea\u9002\u5e94\u6a21\u578b\u63d0\u4f9b\u4e86\u66f4\u7075\u6d3b\u548c\u9ad8\u6548\u7684\u65b9\u6cd5\u3002\u4e0e\u5176\u8bd5\u56fe\u4e00\u6b21\u6027\u8bad\u7ec3 LLM \u5b8c\u6210\u6240\u6709\u4efb\u52a1\uff0c\u4e13\u5bb6\u6a21\u5757\u53ef\u4ee5\u79bb\u7ebf\u5f00\u53d1\u5e76\u6309\u9700\u589e\u5f3a\u5230\u57fa\u7840 LLM \u4e2d\u3002\u8fd9\u4f7f\u6a21\u578b\u80fd\u591f\u6839\u636e\u5f53\u524d\u4efb\u52a1\u52a8\u6001\u4fee\u6539\u5176\u884c\u4e3a\uff0c\u800c\u65e0\u9700\u4e0d\u65ad\u91cd\u65b0\u8c03\u6574\u3002\u9664\u4e86\u5177\u6709\u72ec\u7acb\u7ec4\u4ef6\u7684\u597d\u5904\u5916\uff0c\u8fd9\u79cd\u6a21\u5757\u5316\u8fd8\u652f\u6301\u6301\u7eed\u5b66\u4e60\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u968f\u65f6\u95f4\u589e\u52a0\u65b0\u6280\u80fd\u800c\u4e0d\u4f1a\u51fa\u73b0\u707e\u96be\u6027\u9057\u5fd8\u3002\u6b64\u5916\uff0c\u81ea\u9002\u5e94 LLM \u53cd\u6620\u4e86\u795e\u7ecf\u79d1\u5b66\u548c\u8ba1\u7b97\u751f\u7269\u5b66\u4e2d\u4e00\u4e2a\u516c\u8ba4\u7684\u539f\u7406\uff0c\u5373\u5927\u8111\u6839\u636e\u5f53\u524d\u4efb\u52a1\u6fc0\u6d3b\u7279\u5b9a\u533a\u57df\uff0c\u5e76\u52a8\u6001\u91cd\u7ec4\u5176\u529f\u80fd\u7f51\u7edc\u4ee5\u54cd\u5e94\u4e0d\u65ad\u53d8\u5316\u7684\u4efb\u52a1\u9700\u6c42\u3002<\/section>\n<section><\/section>\n<section>\u539f\u5219\u4e0a\uff0c\u5b9e\u73b0\u81ea\u9002\u5e94 LLM \u7684\u7b2c\u4e00\u6b65\u53ef\u4ee5\u901a\u8fc7\u5f00\u53d1\u4e13\u95e8\u7684\u4e13\u5bb6\u6a21\u5757\u6765\u5b9e\u73b0\uff0c\u6bcf\u4e2a\u6a21\u5757\u90fd\u901a\u8fc7 LoRA \u7b49\u6280\u672f\u8fdb\u884c\u5fae\u8c03\u3002\u7136\u540e\u8fd9\u4e9b\u4e13\u5bb6\u6a21\u5757\u53ef\u4ee5\u6839\u636e\u4efb\u52a1\u9700\u6c42\u5728\u8fd0\u884c\u65f6\u52a8\u6001\u7ec4\u5408\uff0c\u8fd9\u4e2a\u8fc7\u7a0b\u53ef\u4ee5\u901a\u8fc7 MoE \u7c7b\u7cfb\u7edf\u9ad8\u6548\u7ba1\u7406\u3002\u7136\u800c\uff0c\u8981\u4f7f\u8fd9\u79cd\u65b9\u6cd5\u65e2\u53ef\u6269\u5c55\u53c8\u5177\u6709\u7ec4\u5408\u6027\uff0c\u9700\u8981\u89e3\u51b3\u51e0\u4e2a\u6311\u6218\u3002\u9996\u5148\uff0c\u5fae\u8c03 LLM \u4ee5\u521b\u5efa\u591a\u4e2a\u4e13\u5bb6\u6a21\u5757\u663e\u8457\u589e\u52a0\u4e86\u9700\u8981\u8bad\u7ec3\u7684\u53c2\u6570\u6570\u91cf\u3002\u5b9e\u9645\u4e0a\uff0c\u5373\u4f7f\u4f7f\u7528 LoRA \u7b49\u53c2\u6570\u9ad8\u6548\u7684\u65b9\u6cd5\uff0c\u8fd9\u4e9b\u6a21\u5757\u7684\u7d2f\u79ef\u5927\u5c0f\u4e5f\u4f1a\u5feb\u901f\u589e\u52a0\uff0c\u5bfc\u81f4\u5b58\u50a8\u548c\u8ba1\u7b97\u9700\u6c42\u589e\u52a0\u3002\u5176\u6b21\uff0c\u8fd9\u4e9b\u4e13\u5bb6\u6a21\u5757\u5f80\u5f80\u5bb9\u6613\u8fc7\u62df\u5408\uff0c\u8fd9\u79cd\u73b0\u8c61\u5728\u8f83\u5c0f\u6570\u636e\u96c6\u6216\u7a84\u4efb\u52a1\u9886\u57df\u8bad\u7ec3\u65f6\u5c24\u4e3a\u666e\u904d\u3002\u7b2c\u4e09\uff0c\u8fd9\u4e9b\u4e13\u5bb6\u6a21\u5757\u7684\u7075\u6d3b\u7ec4\u5408\u4e5f\u5e26\u6765\u4e86\u76ee\u524d\u5c1a\u672a\u89e3\u51b3\u7684\u6311\u6218\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u514b\u670d\u8fd9\u4e9b\u9650\u5236\uff0c\u4f5c\u8005\u9996\u5148\u63d0\u51fa\u4e86\u5947\u5f02\u503c\u5fae\u8c03\uff08SVF\uff09\uff0c\u8fd9\u662f\u4e00\u79cd\u65b0\u7684\u53c2\u6570\u9ad8\u6548\u5fae\u8c03\uff08PEFT\uff09\u65b9\u6cd5\uff0c\u7528\u4e8e\u83b7\u5f97\u81ea\u9002\u5e94\u7684\u6709\u6548\u6784\u5efa\u5757\u3002SVF \u901a\u8fc7\u4ec5\u63d0\u53d6\u548c\u8c03\u6574\u6a21\u578b\u6743\u91cd\u77e9\u9635\u4e2d\u7684\u5947\u5f02\u503c\u6765\u5de5\u4f5c\u3002\u901a\u8fc7\u4e13\u6ce8\u4e8e\u8fd9\u79cd\u539f\u5219\u6027\u7684\u53c2\u6570\u5316\uff0c\u4ed6\u4eec\u63d0\u51fa\u7684\u65b9\u6cd5\u964d\u4f4e\u4e86\u8fc7\u62df\u5408\u98ce\u9669\uff0c\u5927\u5e45\u51cf\u5c11\u4e86\u8ba1\u7b97\u9700\u6c42\uff0c\u5e76\u5141\u8bb8\u56fa\u6709\u7684\u7ec4\u5408\u6027\u3002\u4ed6\u4eec\u8bc1\u660e\u8fd9\u4e9b\u7279\u6027\u4f7f\u4ed6\u4eec\u80fd\u591f\u901a\u8fc7\u5728\u7a84\u6570\u636e\u96c6\u4e0a\u4f7f\u7528\u5f3a\u5316\u5b66\u4e60\u8fdb\u884c\u8bad\u7ec3\u6765\u5ec9\u4ef7\u5730\u83b7\u5f97\u4e00\u7ec4\u6709\u6548\u7684\u9886\u57df\u7279\u5b9a\u300c\u4e13\u5bb6\u300d\u5411\u91cf\uff0c\u76f4\u63a5\u4f18\u5316\u5404\u4e2a\u4e3b\u9898\u7684\u4efb\u52a1\u6027\u80fd\u3002<\/section>\n<section><\/section>\n<section>\u7136\u540e\uff0c\u4f5c\u8005\u5f15\u5165\u4e86\u5b8c\u6574\u7684 Transformer^2 \u6846\u67b6\uff0c\u901a\u8fc7\u81ea\u9002\u5e94\u7684\u57fa\u672c\u539f\u5219\u6765\u589e\u5f3a LLM\u3002\u7ed9\u5b9a\u6765\u81ea\u672a\u77e5\u4efb\u52a1\u7684\u63d0\u793a\uff0cTransformer^2 \u91c7\u7528\u4e24\u9636\u6bb5\u63a8\u7406\u673a\u5236\uff0c\u5982\u56fe 1 \u6240\u793a\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-92482e740381b7d5f58a47aefadf2add.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-92482e740381b7d5f58a47aefadf2add.png\"><\/a><\/section>\n<section>\u5728\u7b2c\u4e00\u9636\u6bb5\uff0cTransformer^2 \u6267\u884c\u6a21\u578b\u5e76\u89c2\u5bdf\u5176\u6d4b\u8bd5\u65f6\u884c\u4e3a\uff0c\u6536\u96c6\u76f8\u5173\u4fe1\u606f\u4ee5\u7406\u89e3\u89e3\u51b3\u5f53\u524d\u95ee\u9898\u6240\u9700\u7684\u6280\u80fd\u3002\u5728\u7b2c\u4e8c\u9636\u6bb5\uff0cTransformer^2 \u6846\u67b6\u4f7f\u7528\u8fd9\u4e9b\u4fe1\u606f\u7ec4\u5408\u53ef\u7528\u7684\u4e13\u5bb6\u5411\u91cf\uff0c\u5e76\u5bf9 LLM \u7684\u57fa\u7840\u6743\u91cd\u63d0\u4f9b\u4e13\u95e8\u9488\u5bf9\u5176\u6d4b\u8bd5\u65f6\u6761\u4ef6\u7684\u65b0\u4fee\u6539\u3002\u4f5c\u8005\u5728 Transformer^2 \u4e2d\u8bbe\u8ba1\u4e86\u4e09\u79cd\u4e0d\u540c\u7684\u9002\u5e94\u7b56\u7565\uff0c\u5e76\u8bc1\u660e\u8fd9\u4e9b\u7b56\u7565\u968f\u7740\u5bf9\u6d4b\u8bd5\u65f6\u6761\u4ef6\u7684\u8bbf\u95ee\u589e\u52a0\u800c\u63d0\u4f9b\u5355\u8c03\u7684\u6027\u80fd\u63d0\u5347\u3002<\/section>\n<section><\/section>\n<section>\u4f5c\u8005\u901a\u8fc7\u5728\u5404\u79cd LLM \u548c\u4efb\u52a1\u4e0a\u7684\u5e7f\u6cdb\u5b9e\u9a8c\u8bc4\u4f30\u4e86 SVF \u548c\u5b8c\u6574\u7684 Transformer^2 \u6846\u67b6\u3002\u9996\u5148\uff0c\u5728\u9886\u57df\u7279\u5b9a\u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u65f6\uff0c\u4ed6\u4eec\u5c55\u793a\u4e86 SVF \u59cb\u7ec8\u4f18\u4e8e\u4f20\u7edf\u7684\u9ad8\u6548\u5fae\u8c03\u7b56\u7565\uff08\u5982 LoRA\uff09\uff0c\u540c\u65f6\u53c2\u6570\u91cf\u51cf\u5c11\u4e86\u6570\u4e2a\u6570\u91cf\u7ea7\u3002\u7136\u540e\uff0c\u4ed6\u4eec\u5c55\u793a\u4e86 Transformer^2 \u80fd\u591f\u8fdb\u4e00\u6b65\u63d0\u9ad8\u6027\u80fd\uff0c\u5373\u4f7f\u5728\u5b8c\u5168\u5206\u5e03\u5916\u7684\u5e94\u7528\uff08\u5982\u89c6\u89c9\u95ee\u7b54\uff09\u4e2d\u4e5f\u80fd\u6709\u6548\u8c03\u6574\u57fa\u7840\u6a21\u578b\u7684\u6743\u91cd\u3002\u6700\u540e\uff0c\u4ed6\u4eec\u5206\u6790\u4e86\u65b0\u6846\u67b6\u7684\u7279\u6027\uff0c\u9a8c\u8bc1\u4e86\u5b83\u5728\u83b7\u5f97\u66f4\u591a\u5f53\u524d\u6d4b\u8bd5\u65f6\u6761\u4ef6\u8bbf\u95ee\u6743\u9650\u65f6\u63d0\u4f9b\u589e\u91cf\u6536\u76ca\uff0c\u751a\u81f3\u5141\u8bb8\u8de8\u6a21\u578b\u67b6\u6784\u91cd\u7528\u9884\u8bad\u7ec3\u7684 SVF \u4e13\u5bb6\u3002<\/section>\n<section><\/section>\n<section><strong>\u65b9\u6cd5\u6982\u89c8<\/strong><\/section>\n<section><\/section>\n<section><strong>\u5947\u5f02\u503c\u5fae\u8c03\uff08SVF\uff09<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u5c31\u50cf\u4eba\u7c7b\u5927\u8111\u901a\u8fc7\u4e92\u8fde\u7684\u795e\u7ecf\u901a\u8def\u5b58\u50a8\u77e5\u8bc6\u548c\u5904\u7406\u4fe1\u606f\u4e00\u6837\uff0cLLM \u5728\u5176\u6743\u91cd\u77e9\u9635\u4e2d\u5b58\u50a8\u77e5\u8bc6\u3002\u8fd9\u4e9b\u77e9\u9635\u662f LLM \u7684\u300c\u5927\u8111\u300d\uff0c\u4fdd\u5b58\u7740\u5b83\u4ece\u8bad\u7ec3\u6570\u636e\u4e2d\u5b66\u5230\u7684\u7cbe\u9ad3\u3002<\/section>\n<section><\/section>\n<section>\u8981\u7406\u89e3\u8fd9\u4e2a\u300c\u5927\u8111\u300d\u5e76\u786e\u4fdd\u5b83\u80fd\u591f\u6709\u6548\u5730\u9002\u5e94\u65b0\u4efb\u52a1\uff0c\u9700\u8981\u4ed4\u7ec6\u7814\u7a76\u5176\u5185\u90e8\u7ed3\u6784\u3002\u8fd9\u5c31\u8981\u4f7f\u7528\u5947\u5f02\u503c\u5206\u89e3\uff08SVD\uff09\uff0cSVD \u5c06\u5b58\u50a8\u5728 LLM \u4e2d\u5e9e\u5927\u3001\u590d\u6742\u7684\u77e5\u8bc6\u5206\u89e3\u6210\u66f4\u5c0f\u7684\u3001\u6709\u610f\u4e49\u7684\u3001\u72ec\u7acb\u7684\u90e8\u5206\uff08\u4f8b\u5982\u6570\u5b66\u3001\u8bed\u8a00\u7406\u89e3\u7b49\u4e0d\u540c\u7684\u7ec4\u4ef6\uff09\u3002<\/section>\n<section><\/section>\n<section>Transformer^2 \u7684\u6838\u5fc3\u662f\u80fd\u591f\u52a8\u6001\u8c03\u6574\u5176\u6743\u91cd\u77e9\u9635\u7684\u5173\u952e\u7ec4\u4ef6\u3002\u5728\u8bad\u7ec3\u65f6\uff0c\u8be5\u7814\u7a76\u5f15\u5165\u5947\u5f02\u503c\u5fae\u8c03\uff08SVF\uff09\uff0c\u8fd9\u662f\u4e00\u79cd\u4f7f\u7528\u5f3a\u5316\u5b66\u4e60\u6765\u589e\u5f3a \/ \u6291\u5236\u6765\u81ea\u4e0d\u540c\u300c\u5927\u8111\u300d\u7ec4\u4ef6\u7684\u4fe1\u53f7\u4ee5\u7528\u4e8e\u5404\u79cd\u4e0b\u6e38\u4efb\u52a1\u7684\u65b9\u6cd5\u3002\u5728\u63a8\u7406\u65f6\uff0c\u8be5\u7814\u7a76\u91c7\u7528\u4e09\u79cd\u4e0d\u540c\u7684\u7b56\u7565\u6765\u68c0\u6d4b\u4efb\u52a1\u7684\u8eab\u4efd\u5e76\u76f8\u5e94\u5730\u8c03\u6574\u6a21\u578b\u7684\u6743\u91cd\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-b82bf83b918e5e431ec1464442767c5a.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-b82bf83b918e5e431ec1464442767c5a.png\"><\/a><\/section>\n<section><strong>\u4f7f\u7528 SVF \u548c RL \u8fdb\u884c\u8bad\u7ec3<\/strong><\/section>\n<section><\/section>\n<section>\u5728\u8bad\u7ec3\u65f6\uff0cSVF \u5b66\u4e60\u4e00\u7ec4 z \u5411\u91cf\uff0c\u6bcf\u4e2a\u4e0b\u6e38\u4efb\u52a1\u4e00\u4e2a\u3002\u6bcf\u4e2a z \u5411\u91cf\u53ef\u4ee5\u88ab\u89c6\u4e3a\u4efb\u52a1\u7684\u4e13\u5bb6\uff0c\u662f\u4e00\u4e2a\u7d27\u51d1\u7684\u8868\u5f81\uff0c\u8d1f\u8d23\u6307\u5b9a\u6743\u91cd\u77e9\u9635\u4e2d\u6bcf\u4e2a\u7ec4\u4ef6\u7684\u6240\u9700\u5f3a\u5ea6\uff0c\u5f62\u6210\u4e00\u7ec4\u300c\u653e\u5927\u5668\u300d\u6216\u300c\u8870\u51cf\u5668\u300d\u6765\u8c03\u8282\u4e0d\u540c\u7ec4\u4ef6\u5bf9\u6a21\u578b\u884c\u4e3a\u7684\u5f71\u54cd\u3002<\/section>\n<section><\/section>\n<section>SVF \u4f7f\u7528 RL \u5728\u9884\u5b9a\u4e49\u7684\u4e0b\u6e38\u4efb\u52a1\u96c6\u4e0a\u5b66\u4e60\u8fd9\u4e9b z \u5411\u91cf\u3002\u5b66\u5f97\u7684 z \u5411\u91cf\u4f7f Transformer^2 \u80fd\u591f\u9002\u5e94\u5404\u79cd\u65b0\u7684\u4e0b\u6e38\u4efb\u52a1\uff0c\u540c\u65f6\u53ea\u5f15\u5165\u6700\u5c11\u91cf\u7684\u989d\u5916\u53c2\u6570\uff08\u5373 z \u5411\u91cf\u3002<\/section>\n<section><\/section>\n<section><strong>\u81ea\u9002\u5e94<\/strong><\/section>\n<section><\/section>\n<section>\u5728\u63a8\u7406\u65f6\uff0c\u8be5\u7814\u7a76\u4e3a\u6846\u67b6\u8bbe\u8ba1\u4e86\u4e00\u4e2a\u4e24\u9636\u6bb5\u9002\u5e94\u7b56\u7565\uff0c\u4ee5\u6709\u6548\u5730\u7ec4\u5408\u4efb\u52a1\u7279\u5b9a\u7684 z \u5411\u91cf\u96c6\u3002\u5728\u7b2c\u4e00\u6b21\u63a8\u7406\u65f6\uff0c\u7ed9\u5b9a\u4efb\u52a1\u6216\u5355\u4e2a\u8f93\u5165\u63d0\u793a\uff0cTransformer^2 \u4f7f\u7528\u4ee5\u4e0b\u4e09\u79cd\u9002\u5e94\u65b9\u6cd5\u4e4b\u4e00\u5206\u6790\u5176\u6d4b\u8bd5\u65f6\u6761\u4ef6\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u57fa\u4e8e\u63d0\u793a\u7684\u9002\u5e94\uff1a\u4e13\u95e8\u8bbe\u8ba1\u7684\u9002\u5e94\u63d0\u793a\uff0c\u5bf9\u4efb\u52a1\u8fdb\u884c\u5206\u7c7b\uff08\u4f8b\u5982\u6570\u5b66\u3001\u7f16\u7a0b\uff09\u5e76\u9009\u62e9\u9884\u8bad\u7ec3\u7684 z \u5411\u91cf\u3002<\/section>\n<\/li>\n<li>\n<section>\u57fa\u4e8e\u5206\u7c7b\u5668\u7684\u9002\u5e94\uff1a\u4f7f\u7528 SVF \u8bad\u7ec3\u7684\u4efb\u52a1\u5206\u7c7b\u5668\uff0c\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u8bc6\u522b\u4efb\u52a1\u5e76\u9009\u62e9\u5408\u9002\u7684 z \u5411\u91cf\u3002<\/section>\n<\/li>\n<li>\n<section>\u5c11\u6837\u672c\u9002\u5e94\uff1a\u901a\u8fc7\u52a0\u6743\u63d2\u503c\u7ec4\u5408\u591a\u4e2a\u9884\u8bad\u7ec3\u7684 z \u5411\u91cf\u3002\u7b80\u5355\u7684\u4f18\u5316\u7b97\u6cd5\u6839\u636e\u5c11\u6837\u672c\u8bc4\u4f30\u96c6\u4e0a\u7684\u6027\u80fd\u8c03\u6574\u8fd9\u4e9b\u6743\u91cd\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u7b2c\u4e8c\u6b21\u63a8\u7406\u65f6\uff0cTransformer^2 \u901a\u8fc7\u7ec4\u5408 z \u5411\u91cf\u76f8\u5e94\u5730\u8c03\u5236\u6743\u91cd\uff0c\u4e3a\u5176\u65b0\u8bbe\u7f6e\u4ea7\u751f\u6700\u76f8\u5173\u7684\u6700\u7ec8\u54cd\u5e94\u3002<\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u7ed3\u679c<\/strong><\/section>\n<section><\/section>\n<section><strong>SVF \u6027\u80fd<\/strong><\/section>\n<section><\/section>\n<section>\u8868 1 \u63d0\u4f9b\u4e86\u5728 LLAMA3-8B-INSTRUCT\u3001MISTRAL-7B-INSTRUCT-V0.3 \u548c LLAMA3-70B-INSTRUCT \u57fa\u7840\u6a21\u578b\u4e0a\u5bf9\u6bcf\u4e2a\u4efb\u52a1\u8fdb\u884c\u8bad\u7ec3\u540e\u7684\u7ed3\u679c\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-e214746fda9b8c39079a1748f79973c7.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-e214746fda9b8c39079a1748f79973c7.png\"><\/a><\/section>\n<section>\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0cSVF \u5728\u51e0\u4e4e\u6240\u6709\u4efb\u52a1\u548c\u57fa\u7840\u6a21\u578b\u4e0a\u90fd\u63d0\u4f9b\u4e86\u663e\u8457\u4e14\u4e00\u81f4\u7684\u6027\u80fd\u63d0\u5347\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cLoRA \u4e13\u5bb6\u4ea7\u751f\u7684\u6536\u76ca\u8f83\u5c0f\uff0c\u751a\u81f3\u51fa\u73b0\u4e86\u96f6\u661f\u7684\u6027\u80fd\u4e0b\u964d\u3002<\/section>\n<section><\/section>\n<section>\u8fd9\u79cd\u8d8b\u52bf\u4e5f\u53ef\u4ee5\u6269\u5c55\u5230\u89c6\u89c9 &#8211; \u8bed\u8a00\u9886\u57df\uff0c\u56e0\u4e3a\u7528 SVF \u5fae\u8c03 LLAMA3-LLAVA-NEXT-8B \u5c06\u57fa\u7840\u6a21\u578b\u7684\u6027\u80fd\u63d0\u5347\u4e86\u8d85\u8fc7 39%\uff08\u89c1\u56fe 5\uff09\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-ec88cc71847fb59e3a053567d8e9310f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-ec88cc71847fb59e3a053567d8e9310f.png\"><\/a><\/section>\n<section><strong>\u9002\u5e94\u6027\u80fd<\/strong><\/section>\n<section><\/section>\n<section>\u8be5\u7814\u7a76\u4f7f\u7528 SVF \u8bad\u7ec3\u7684 z \u5411\u91cf\u8bc4\u4f30\u4e86 Transformer^2 \u5728\u672a\u89c1\u4efb\u52a1\u4e0a\u7684\u81ea\u9002\u5e94\u80fd\u529b\u3002<\/section>\n<section><\/section>\n<section>\u5982\u8868 2 \u6240\u793a\uff0c\u6240\u6709\u7684 Transformer^2 \u9002\u5e94\u7b56\u7565\u90fd\u5728 LLAMA3-8B-INSTRUCT \u57fa\u7840\u6a21\u578b\u7684\u6240\u6709\u4efb\u52a1\u4e0a\u8868\u73b0\u51fa\u6027\u80fd\u63d0\u5347\uff0c\u5728 MISTRAL-7B-INSTRUCT-V0.3 \u548c LLAMA3-70B-INSTRUCT \u7684\u4e09\u4e2a\u4efb\u52a1\u4e2d\u81f3\u5c11\u6709\u4e24\u4e2a\u4efb\u52a1\u6709\u6240\u6539\u8fdb\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u5373\u4f7f\u662f\u6700\u4f73\u8bad\u7ec3 LoRA \u4e5f\u53ea\u5728 ARC-Challenge \u4efb\u52a1\u4e0a\u63d0\u4f9b\u4e86\u6539\u8fdb\uff0c\u5728 MATH \u548c Humaneval \u4e0a\u663e\u8457\u964d\u4f4e\u4e86\u6027\u80fd\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-a2796b19c9fd4e26643e3dca21a40f83.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-a2796b19c9fd4e26643e3dca21a40f83.png\"><\/a><\/section>\n<section>\u8fd9\u79cd\u5dee\u5f02\u8868\u660e LoRA \u7684\u53c2\u6570\u5316\u548c\u4f18\u5316\u53ef\u80fd\u7279\u522b\u5bb9\u6613\u8fc7\u62df\u5408\uff0c\u7279\u522b\u662f\u5728\u4f7f\u7528\u8f83\u5c0f\u7684 GSM8K \u548c MBPP-Pro \u6570\u636e\u96c6\u8bad\u7ec3\u65f6\u3002<\/section>\n<section><\/section>\n<section>\u5728\u56fe 5 \u4e2d\uff0c\u57fa\u7840 LLAMA3-LLAVA-NEXT-8B VLM \u7684\u6027\u80fd\u4ec5\u5728\u5e94\u7528 Transformer^2 \u540e\u5f97\u5230\u6539\u5584\u3002\u7814\u7a76\u56e2\u961f\u6ce8\u610f\u5230\u5728\u8fd9\u79cd\u8bbe\u7f6e\u4e2d\uff0cTransformer^2 \u4ec5\u4ece GSM8K\u3001MBPP-Pro \u548c ARC-Easy \u7684\u4e13\u5bb6\u5411\u91cf\u4e2d\u8fdb\u884c\u81ea\u9002\u5e94\u3002\u56e0\u6b64\uff0c\u8fd9\u4e00\u7ed3\u679c\u8fdb\u4e00\u6b65\u5f3a\u8c03\u4e86\u81ea\u9002\u5e94\u7684\u9ad8\u5ea6\u7075\u6d3b\u6027\uff0c\u57fa\u4e8e\u8bed\u8a00\u4efb\u52a1\u7684\u77e5\u8bc6\u4e5f\u53ef\u4ee5\u8fc1\u79fb\u5230\u4e0d\u76f8\u5173\u7684\u57fa\u4e8e\u89c6\u89c9\u7684\u95ee\u9898\u4e0a\u3002<\/section>\n<section><\/section>\n<section>\u901a\u8fc7\u5bf9\u4e09\u79cd\u9002\u5e94\u7b56\u7565\u7684\u6bd4\u8f83\uff0c\u4f5c\u8005\u53d1\u73b0\u4e86\u4e00\u4e2a\u660e\u663e\u7684\u5355\u8c03\u8d8b\u52bf \u2014\u2014 \u5373\u968f\u7740\u7b56\u7565\u7684\u589e\u52a0\u548c\u6d4b\u8bd5\u65f6\u95f4\u6761\u4ef6\u7684\u589e\u52a0\uff0c\u81ea\u9002\u5e94\u7684\u6548\u679c\u8d8a\u6765\u8d8a\u660e\u663e\u3002\u7279\u522b\u662f\uff0c\u5177\u6709\u5c11\u6837\u672c\u81ea\u9002\u5e94\u7684 Transformer^2 \u51e0\u4e4e\u603b\u662f\u5f97\u5206\u6700\u9ad8\u7684\u65b9\u6cd5\uff0c\u5728\u6240\u6709\u6d4b\u8bd5\u8bbe\u7f6e\u4e2d\u90fd\u63d0\u4f9b\u4e86\u663e\u8457\u6539\u8fdb\uff0c\u9664\u4e86 LLAMA3-70B-INSTRUCT @MATH\u3002\u7531\u4e8e GPU \u8d44\u6e90\u6709\u9650\uff0c\u4f5c\u8005\u53ea\u5bf9\u4e00\u534a\u7684\u5c42\u8fdb\u884c\u4e86 SVF \u8c03\u4f18\u3002\u8fd9\u79cd\u8d8b\u52bf\u8868\u660e\uff0c\u63d0\u4f9b\u989d\u5916\u6216\u4e0d\u540c\u7c7b\u578b\u7684\u4fe1\u606f\u4f3c\u4e4e\u5bf9 Transformer^2 \u6846\u67b6\u975e\u5e38\u6709\u76ca\uff0c\u8868\u660e \u00a0Transformer^2 \u53ef\u4ee5\u4e3a\u57fa\u7840\u6a21\u578b\u63d0\u4f9b\u5728\u7ec8\u8eab\u8bbe\u7f6e\u4e2d\u90e8\u7f72\u65f6\u6301\u7eed\u6539\u8fdb\u6027\u80fd\u7684\u65b0\u65b9\u6cd5\u3002<\/section>\n<section><\/section>\n<section>\u8868 3 \u62a5\u544a\u4e86 Transformer^2 \u7684\u63d0\u793a\u9002\u5e94\u7b56\u7565\u6240\u9700\u7684\u63a8\u7406\u65f6\u95f4\uff0c\u5206\u522b\u5c55\u793a\u4e86\u7b2c\u4e00\u9636\u6bb5\u548c\u7b2c\u4e8c\u51e0\u6bb5\u89e3\u51b3\u6574\u4e2a\u95ee\u9898\u96c6\u6240\u82b1\u8d39\u7684\u65f6\u95f4\u3002\u6ce8\u610f\uff0c\u300c2nd pass\u300d\u63a8\u7406\u65f6\u95f4\u662f\u89e3\u51b3\u95ee\u9898\u6240\u82b1\u8d39\u7684\u65f6\u95f4\uff0c\u300c1st pass\u300d\u63a8\u7406\u65f6\u95f4\u662f\u81ea\u9002\u5e94\u7684\u65f6\u95f4\u3002\u62ec\u53f7\u4e2d\u662f\u300c1st pass\u300d\u5360\u300c2nd pass\u300d\u63a8\u7406\u65f6\u95f4\u7684\u6bd4\u7387\u3002\u867d\u7136\u989d\u5916\u7684\u63a8\u7406\u9636\u6bb5\u53ef\u80fd\u770b\u8d77\u6765\u4f1a\u4f7f\u6574\u4f53\u8fd0\u884c\u65f6\u95f4\u7ffb\u500d\uff0c\u4f46\u91cd\u8981\u7684\u662f\u8981\u6ce8\u610f\u63a8\u7406\u65f6\u95f4\u4e3b\u8981\u53d6\u51b3\u4e8e\u751f\u6210\u7684 token \u6570\u91cf\u3002\u5728\u8bba\u6587\u7684\u8bbe\u7f6e\u4e2d\uff0c\u5b83\u662f O (n)\uff0c\u5176\u4e2d n \u662f\u8f93\u5165\u7684\u957f\u5ea6\u3002ARC-challenge \u62ec\u53f7\u4e2d\u7684\u6570\u503c\u8f83\u5927\uff0c\u56e0\u4e3a\u5b83\u4eec\u662f\u5355\u9009\u9898\uff0c\u56e0\u6b64\u300c2nd pass\u300d\u7684\u6210\u672c\u4e5f\u662f O (n)\u3002\u5728\u4e00\u822c\u8bbe\u7f6e\u4e2d\uff0c\u4f5c\u8005\u8ba4\u4e3a\u8fd9\u4e2a\u6bd4\u7387\u66f4\u63a5\u8fd1 MATH \u548c Humaneval \u7684\u6bd4\u7387\u662f\u5408\u7406\u7684\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-b1b01d72b655107ab288181559613889.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-b1b01d72b655107ab288181559613889.png\"><\/a><\/section>\n<section>\u66f4\u591a\u5185\u5bb9\u8bf7\u53c2\u89c1\u539f\u8bba\u6587\u3002<\/section>\n<section><\/section>\n<section><em><sup>\u53c2\u8003\u94fe\u63a5\uff1ahttps:\/\/sakana.ai\/transformer-squared\/<\/sup><\/em><\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-01-15-11\" target=\"_blank\">Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:Transformer [&hellip;]<\/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":[55],"class_list":["post-37755","post","type-post","status-publish","format-standard","hentry","category-news","tag-55"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883 - \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=37755\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=37755\",\"url\":\"https:\/\/17aitech.com\/?p=37755\",\"name\":\"Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=37755#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=37755#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png\",\"datePublished\":\"2025-02-03T15:00:54+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=37755#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=37755\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=37755#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=37755#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883\"}]},{\"@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\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"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":"Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883 - \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=37755","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=37755","url":"https:\/\/17aitech.com\/?p=37755","name":"Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=37755#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=37755#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png","datePublished":"2025-02-03T15:00:54+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=37755#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=37755"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=37755#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-fc81bcfa8be876714c89dbb3f595764d.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=37755#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"Transformer\u00b2\u8981\u505a\u300c\u6d3b\u300d\u7684AI\u6a21\u578b\uff0c\u52a8\u6001\u8c03\u6574\u6743\u91cd\uff0c\u50cf\u7ae0\u9c7c\u4e00\u6837\u9002\u5e94\u73af\u5883"}]},{"@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":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","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\/37755","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=37755"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/37755\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=37755"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=37755"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=37755"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}