{"id":5245,"date":"2024-12-04T10:31:55","date_gmt":"2024-12-04T02:31:55","guid":{"rendered":"https:\/\/nullthought.net\/?p=5245"},"modified":"2025-02-23T11:08:36","modified_gmt":"2025-02-23T03:08:36","slug":"star%ef%bc%88synthesis-of-tailored-architectures%ef%bc%89%ef%bc%9a%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e7%9a%84%e5%90%88%e6%88%90%e5%ae%9a%e5%88%b6%e5%8c%96%e6%9e%b6%e6%9e%84","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=5245","title":{"rendered":"STAR\uff08Synthesis of Tailored Architectures\uff09\uff1a\u6df1\u5ea6\u5b66\u4e60\u7684\u5408\u6210\u5b9a\u5236\u5316\u67b6\u6784"},"content":{"rendered":"\n<p>\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/2411.17800\" target=\"_blank\" rel=\"noreferrer noopener\">STAR: Synthesis of Tailored Architectures<\/a><\/strong>\uff08\u300aSTAR: \u5408\u6210\u5b9a\u5236\u5316\u67b6\u6784\u300b\uff09\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3aSTAR\u7684\u521b\u65b0\u65b9\u6cd5\uff0c\u7528\u4e8e\u4f18\u5316\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u67b6\u6784\u3002STAR\u91c7\u7528\u4e86\u4e00\u79cd\u7ed3\u5408\u65b0\u7684\u641c\u7d22\u7a7a\u95f4\u548c\u8fdb\u5316\u7b97\u6cd5\uff08evolutionary algorithm\uff09\u7684\u65b9\u6cd5\uff0c\u4f7f\u5f97\u5728\u591a\u4e2a\u76ee\u6807\uff08\u4f8b\u5982\u6a21\u578b\u8d28\u91cf\u3001\u53c2\u6570\u91cf\u3001\u63a8\u7406\u7f13\u5b58\u5927\u5c0f\u7b49\uff09\u4e0a\u4f18\u5316\u6a21\u578b\u6210\u4e3a\u53ef\u80fd\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aArmin W. Thomas, Rom Parnichkun, Alexander Amini, Stefano Massaroli, Michael Poli\uff0c\u5747\u6765\u81ea<strong><a href=\"https:\/\/www.liquid.ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">Liquid AI<\/a><\/strong>\u516c\u53f8\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e00\u3001\u5f15\u8a00<\/h5>\n\n\n\n<p>\u5728\u5f15\u8a00\u4e2d\uff0c\u4f5c\u8005\u6307\u51fa\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u4e2d\u7684\u6a21\u578b\u67b6\u6784\u8bbe\u8ba1\u5b58\u5728\u4e24\u4e2a\u4e3b\u8981\u7684\u4f18\u5316\u9014\u5f84\uff1a\u81ea\u52a8\u5316\u548c\u624b\u52a8\u8bbe\u8ba1\u3002\u867d\u7136\u624b\u52a8\u8bbe\u8ba1\u53ef\u4ee5\u5229\u7528\u4eba\u7c7b\u7684\u76f4\u89c9\u548c\u7ecf\u9a8c\uff0c\u4ea7\u751f\u6709\u6548\u7684\u6a21\u578b\u67b6\u6784\uff0c\u4f46\u8fd9\u79cd\u65b9\u6cd5\u901a\u5e38\u9700\u8981\u5927\u91cf\u7684\u8d44\u6e90\u548c\u65f6\u95f4\u3002\u800c\u81ea\u52a8\u5316\u67b6\u6784\u641c\u7d22\u65b9\u6cd5\uff0c\u5c3d\u7ba1\u5728\u67d0\u4e9b\u7279\u5b9a\u4efb\u52a1\u4e2d\u8868\u73b0\u826f\u597d\uff08\u4f8b\u5982\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u7684\u4f18\u5316\uff09\uff0c\u4f46\u7531\u4e8e\u641c\u7d22\u7a7a\u95f4\u7684\u5c40\u9650\u6027\u548c\u8fc7\u4e8e\u7b80\u5355\u7684\u6a21\u5f0f\uff0c\u5e38\u5e38\u65e0\u6cd5\u663e\u8457\u8d85\u8d8a\u73b0\u6709\u7684\u624b\u5de5\u8bbe\u8ba1\u7684\u6807\u51c6\u6a21\u578b\u3002\u8fd9\u4e5f\u4f53\u73b0\u5728Transformer\u65f6\u4ee3\uff0c\u6a21\u578b\u67b6\u6784\u7684\u540c\u8d28\u5316\u660e\u663e\uff0c\u51e0\u4e4e\u6240\u6709\u5927\u578b\u8bed\u8a00\u6a21\u578b\u90fd\u91c7\u7528\u4e86\u7c7b\u4f3c\u7684\u67b6\u6784\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3\u73b0\u6709\u65b9\u6cd5\u7684\u5c40\u9650\u6027\uff0c\u4f5c\u8005\u63d0\u51fa\u4e86STAR\uff08Synthesis of Tailored Architectures\uff09\u65b9\u6cd5\uff0c\u8fd9\u79cd\u65b9\u6cd5\u901a\u8fc7\u5f15\u5165\u65b0\u7684\u641c\u7d22\u7a7a\u95f4\u4ee5\u53ca\u4e0e\u8fdb\u5316\u7b97\u6cd5\u7684\u7ed3\u5408\uff0c\u81f4\u529b\u4e8e\u521b\u5efa\u9002\u5e94\u591a\u6837\u5316\u9700\u6c42\u7684\u6a21\u578b\u67b6\u6784\uff0c\u4ee5\u6ee1\u8db3\u4e0d\u540c\u4efb\u52a1\u5728\u6548\u7387\u548c\u8d28\u91cf\u65b9\u9762\u7684\u8981\u6c42\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e8c\u3001\u641c\u7d22\u7a7a\u95f4\u7684\u57fa\u7840<\/h5>\n\n\n\n<p>STAR\u65b9\u6cd5\u7684\u6838\u5fc3\u5728\u4e8e\u5b83\u7684\u641c\u7d22\u7a7a\u95f4\u8bbe\u8ba1\u3002\u4f20\u7edf\u81ea\u52a8\u5316\u67b6\u6784\u8bbe\u8ba1\u9762\u4e34\u7684\u4e00\u4e2a\u4e3b\u8981\u6311\u6218\u662f\u5982\u4f55\u5b9a\u4e49\u4e00\u4e2a\u65e2\u7a33\u5b9a\u53c8\u5168\u9762\u7684\u641c\u7d22\u7a7a\u95f4\u3002\u5728\u8fd9\u79cd\u80cc\u666f\u4e0b\uff0c\u4f5c\u8005\u63d0\u51fa\u4e86\u57fa\u4e8e\u7ebf\u6027\u8f93\u5165\u53d8\u52a8\u7cfb\u7edf\uff08LIV\uff0cLinear Input-Varying Systems\uff09\u7684\u641c\u7d22\u7a7a\u95f4\u3002LIV\u4e3a\u6df1\u5ea6\u5b66\u4e60\u4e2d\u7684\u5404\u7c7b\u8ba1\u7b97\u5355\u5143\uff08\u4f8b\u5982\u6ce8\u610f\u529b\u53d8\u4f53\u3001\u5377\u79ef\u3001\u7ebf\u6027\u56de\u5f52\u7b49\uff09\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5e7f\u4e49\u5316\u7684\u6846\u67b6\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">2.1 \u7ebf\u6027\u8f93\u5165\u53d8\u52a8\u7cfb\u7edf\uff08LIV\uff09\u7684\u5b9a\u4e49<\/h6>\n\n\n\n<p>LIV\u662f\u4e00\u7c7b\u53ef\u7528\u6765\u63cf\u8ff0\u8f93\u5165\u5e8f\u5217\u7684\u7cfb\u7edf\uff0c\u5176\u4e2d\u8f93\u5165\u662f\u4e00\u4e2a\u5411\u91cf\u5e8f\u5217\uff0c\u4f8b\u5982 {x<sub>0<\/sub>,x<sub>1<\/sub>,\u2026,x<sub>\u2113<\/sub>}\uff0c\u5176\u4e2d\u6bcf\u4e2a\u5411\u91cf\u90fd\u53ef\u4ee5\u770b\u4f5c\u662f\u4e00\u4e2atoken\uff0c\u6bcf\u4e2atoken\u6709\u591a\u4e2achannel\u7ec4\u6210\uff08\u7c7b\u4f3c\u4e8e\u4e00\u4e2a\u591a\u7ef4\u7279\u5f81\u5411\u91cf\uff09\u3002\u5728LIV\u6846\u67b6\u4e0b\uff0c\u8ba1\u7b97\u7531\u4e00\u4e2a\u7ebf\u6027\u64cd\u4f5c\u7b26\u63a7\u5236\uff0c\u4e14\u8be5\u64cd\u4f5c\u7b26\u7684\u884c\u4e3a\u7531\u8f93\u5165\u81ea\u8eab\u51b3\u5b9a\u3002\u4f8b\u5982\uff0c\u5728\u7ecf\u5178\u7684\u6ce8\u610f\u529b\u673a\u5236\u4e2d\uff0c\u64cd\u4f5c\u7b26\u7684\u6743\u91cd\u7531\u8f93\u5165\u901a\u8fc7\u67d0\u4e9b\u7ebf\u6027\u6216\u975e\u7ebf\u6027\u53d8\u6362\u8ba1\u7b97\u5f97\u5230\uff0c\u4ece\u800c\u5f62\u6210\u4e86LIV\u7684\u4e00\u4e2a\u5178\u578b\u5b9e\u4f8b\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">2.2 LIV\u7684\u5c42\u6b21\u5316\u63cf\u8ff0<\/h6>\n\n\n\n<p>LIV\u7cfb\u7edf\u53ef\u4ee5\u5728\u4e09\u4e2a\u5c42\u6b21\u4e0a\u63cf\u8ff0\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u7279\u5f81\u5316\uff08Featurization\uff09<\/strong>\uff1a\u5b9a\u4e49\u8f93\u5165\u4e0a\u4e0b\u6587\u5982\u4f55\u8c03\u5236\u7ebf\u6027\u8ba1\u7b97\u3002\u8fd9\u5305\u62ec\u4f7f\u7528\u8f93\u5165\u8ba1\u7b97\u6ce8\u610f\u529b\u6743\u91cd\u7684\u8fc7\u7a0b\uff0c\u4f8b\u5982\u901a\u8fc7\u7ebf\u6027\u53d8\u6362\u5f97\u5230query\u548ckey\u7684\u7279\u5f81\u3002<\/li>\n\n\n\n<li><strong>\u64cd\u4f5c\u7b26\u7ed3\u6784\uff08Operator Structure\uff09<\/strong>\uff1a\u5b9a\u4e49token\u548cchannel\u7684\u6df7\u5408\u7ed3\u6784\uff0c\u5373\u5982\u4f55\u5728\u8f93\u5165\u7684\u4e0d\u540ctoken\u548c\u4e0d\u540cchannel\u4e4b\u95f4\u8fdb\u884c\u7ebf\u6027\u7ec4\u5408\u3002<\/li>\n\n\n\n<li><strong>\u9aa8\u67b6\uff08Backbone\uff09<\/strong>\uff1a\u5b9a\u4e49\u591a\u4e2aLIV\u5982\u4f55\u7ec4\u6210\u4e00\u4e2a\u5b8c\u6574\u7684\u67b6\u6784\u3002\u5728STAR\u4e2d\uff0c\u8fd9\u79cd\u9aa8\u67b6\u7ed3\u6784\u53ef\u4ee5\u901a\u8fc7\u8fdb\u5316\u7b97\u6cd5\u8fdb\u884c\u4f18\u5316\uff0c\u4ece\u800c\u627e\u5230\u6700\u4f18\u7684LIV\u7ec4\u5408\u5f62\u5f0f\u3002<\/li>\n<\/ol>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e09\u3001STAR\u57fa\u56e0\u7ec4<\/h5>\n\n\n\n<p>\u5728STAR\u65b9\u6cd5\u4e2d\uff0c\u67b6\u6784\u901a\u8fc7\u57fa\u56e0\u7ec4\uff08Genome\uff09\u8fdb\u884c\u7f16\u7801\uff0c\u79f0\u4e3a\u201cSTAR\u57fa\u56e0\u7ec4\u201d\u3002\u57fa\u56e0\u7ec4\u7684\u7ed3\u6784\u662f\u5c42\u6b21\u5316\u7684\uff0c\u5206\u4e3a\u4e09\u4e2a\u4e3b\u8981\u90e8\u5206\uff1a\u7279\u5f81\u5316\u3001\u64cd\u4f5c\u7b26\u7ed3\u6784\u548c\u9aa8\u67b6\u3002\u6bcf\u4e00\u4e2a\u90e8\u5206\u90fd\u7528\u6574\u6570\u7f16\u7801\uff0c\u4f7f\u5f97\u5b83\u4eec\u53ef\u4ee5\u901a\u8fc7\u8fdb\u5316\u7b97\u6cd5\u8fdb\u884c\u4f18\u5316\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">3.1 \u9aa8\u67b6\u57fa\u56e0\u7ec4<\/h6>\n\n\n\n<p>\u9aa8\u67b6\u57fa\u56e0\u7ec4\u63cf\u8ff0\u4e86LIV\u4e4b\u95f4\u7684\u8fde\u63a5\u65b9\u5f0f\u3002\u6bcf\u4e2aLIV\u90fd\u88ab\u8868\u793a\u4e3a\u4e00\u4e2a5\u4f4d\u6574\u6570\u5e8f\u5217\uff0c\u5176\u4e2d\u5305\u62ec\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>LIV\u7c7b\u522b<\/strong>\uff1a\u8868\u793a\u5f53\u524dLIV\u7684\u7c7b\u578b\uff0c\u4f8b\u5982\u662f\u5377\u79ef\u3001\u6ce8\u610f\u529b\u8fd8\u662f\u67d0\u79cd\u7ebf\u6027\u53d8\u6362\u3002<\/li>\n\n\n\n<li><strong>\u7279\u5f81\u5316\u6743\u91cd\u5171\u4eab<\/strong>\uff1a\u786e\u5b9a\u5728\u9aa8\u67b6\u4e2d\u4e0d\u540c\u6df1\u5ea6\u7684LIV\u4e4b\u95f4\u662f\u5426\u5171\u4eab\u7279\u5f81\u5316\u6743\u91cd\u3002<\/li>\n\n\n\n<li><strong>\u7279\u5f81\u5316\u5171\u4eab\u7b56\u7565<\/strong>\uff1a\u5b9a\u4e49\u5982\u4f55\u5b9e\u73b0\u7279\u5f81\u5316\u7684\u5171\u4eab\uff0c\u4f8b\u5982\u4ec5\u5171\u4eab\u8ba1\u7b97\u67d0\u4e9b\u7279\u5f81\u7ec4\u7684\u6743\u91cd\u3002<\/li>\n\n\n\n<li><strong>\u7279\u5f81\u7ec4\u5171\u4eab<\/strong>\uff1aLIV\u4e4b\u95f4\u53ef\u4ee5\u76f4\u63a5\u5171\u4eab\u7279\u5f81\u7ec4\uff0c\u8fd9\u610f\u5473\u7740\u67d0\u4e9b\u8ba1\u7b97\u53ef\u4ee5\u91cd\u590d\u4f7f\u7528\uff0c\u4ece\u800c\u63d0\u9ad8\u6548\u7387\u3002<\/li>\n\n\n\n<li><strong>\u7279\u5f81\u7ec4\u5171\u4eab\u7b56\u7565<\/strong>\uff1a\u63cf\u8ff0\u5728\u6240\u6709\u7279\u5f81\u7ec4\u4e2d\uff0c\u54ea\u4e9b\u7279\u5f81\u7ec4\u53ef\u4ee5\u5171\u4eab\u3002<\/li>\n<\/ol>\n\n\n\n<p>\u8fd9\u79cd\u7f16\u7801\u65b9\u5f0f\u53ef\u4ee5\u4f7f\u5f97\u8fdb\u5316\u7b97\u6cd5\u5728\u4e0d\u540c\u7684\u5c42\u6b21\u4e0a\u5bf9\u67b6\u6784\u8fdb\u884c\u4f18\u5316\u3002\u4f8b\u5982\uff0c\u7279\u5f81\u7ec4\u5171\u4eab\u7b56\u7565\u7684\u53d8\u5316\u53ef\u4ee5\u4ea7\u751f\u65b0\u7684\u6a21\u578b\u67b6\u6784\uff0c\u800c\u4e0d\u9700\u8981\u6539\u53d8\u6574\u4f53\u7684\u67b6\u6784\u9aa8\u67b6\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">3.2 \u64cd\u4f5c\u7b26\u548c\u7279\u5f81\u5316\u57fa\u56e0\u7ec4<\/h6>\n\n\n\n<p>\u64cd\u4f5c\u7b26\u57fa\u56e0\u7ec4\u5b9a\u4e49\u4e86LIV\u7684\u5177\u4f53\u5f62\u5f0f\uff0c\u5305\u62ec\u7279\u5f81\u5316\u65b9\u6cd5\u3001token\u6df7\u5408\u7ed3\u6784\u7b49\u3002\u7279\u5f81\u5316\u57fa\u56e0\u7ec4\u5219\u8fdb\u4e00\u6b65\u63cf\u8ff0\u4e86\u6bcf\u4e2a\u7279\u5f81\u7ec4\u7684\u5177\u4f53\u7ed3\u6784\uff0c\u4f8b\u5982channel\u7684\u6269\u5c55\u56e0\u5b50\u3001\u91cd\u590d\u56e0\u5b50\u7b49\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u56db\u3001\u901a\u8fc7\u8fdb\u5316\u7b97\u6cd5\u5408\u6210\u67b6\u6784<\/h5>\n\n\n\n<p>STAR\u901a\u8fc7\u8fdb\u5316\u7b97\u6cd5\u5bf9\u57fa\u56e0\u7ec4\u8fdb\u884c\u4f18\u5316\uff0c\u4ee5\u627e\u5230\u6700\u4f18\u7684\u6a21\u578b\u67b6\u6784\u3002\u6574\u4e2a\u8fdb\u5316\u8fc7\u7a0b\u5305\u62ec\u4ee5\u4e0b\u6b65\u9aa4\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u8bc4\u4f30\uff08Assessment\uff09<\/strong>\uff1a\u9996\u5148\uff0c\u5bf9\u521d\u59cb\u79cd\u7fa4\u4e2d\u7684\u6bcf\u4e2a\u57fa\u56e0\u7ec4\u8fdb\u884c\u8bc4\u4f30\u3002\u8fd9\u5305\u62ec\u901a\u8fc7\u8bad\u7ec3\u6a21\u578b\u5e76\u6839\u636e\u6027\u80fd\u6307\u6807\uff08\u4f8b\u5982\u56f0\u60d1\u5ea6PPL\uff09\u5bf9\u5176\u8fdb\u884c\u6253\u5206\u3002\u9664\u4e86\u8d28\u91cf\u76ee\u6807\u5916\uff0c\u8fd8\u53ef\u4ee5\u8fdb\u884c\u9759\u6001\u5206\u6790\uff0c\u5982\u53c2\u6570\u6570\u91cf\u548c\u63a8\u7406\u7f13\u5b58\u5927\u5c0f\uff0c\u4ee5\u8bc4\u4f30\u6548\u7387\u3002<\/li>\n\n\n\n<li><strong>\u914d\u5bf9\uff08Pairing\uff09<\/strong>\uff1a\u7136\u540e\uff0c\u9009\u62e9\u8868\u73b0\u6700\u597d\u7684\u4e2a\u4f53\u8fdb\u884c\u914d\u5bf9\uff0c\u91c7\u7528\u9526\u6807\u8d5b\u9009\u62e9\uff08Tournament Selection\uff09\uff0c\u5373\u968f\u673a\u9009\u62e9\u4e00\u4e9b\u57fa\u56e0\u7ec4\u5e76\u6311\u9009\u8868\u73b0\u6700\u597d\u7684\u4e2a\u4f53\u3002<\/li>\n\n\n\n<li><strong>\u91cd\u7ec4\uff08Recombination\uff09\u548c\u53d8\u5f02\uff08Mutation\uff09<\/strong>\uff1a\u63a5\u7740\uff0c\u901a\u8fc7\u4ea4\u6362\u9057\u4f20\u7269\u8d28\u751f\u6210\u65b0\u7684\u5019\u9009\u57fa\u56e0\u7ec4\uff0c\u5e76\u5f15\u5165\u968f\u673a\u53d8\u5f02\u4ee5\u4fdd\u6301\u79cd\u7fa4\u7684\u591a\u6837\u6027\u3002\u91cd\u7ec4\u901a\u8fc7k\u70b9\u4ea4\u53c9\uff08k-point crossover\uff09\u5b8c\u6210\uff0c\u5373\u5728\u57fa\u56e0\u7ec4\u4e2d\u968f\u673a\u9009\u62e9k\u4e2a\u70b9\u8fdb\u884c\u4ea4\u6362\uff0c\u4ece\u800c\u751f\u6210\u65b0\u7684\u5019\u9009\u57fa\u56e0\u7ec4\u3002<\/li>\n\n\n\n<li><strong>\u4fee\u590d\uff08Repair\uff09<\/strong>\uff1a\u5982\u679c\u91cd\u7ec4\u6216\u53d8\u5f02\u5bfc\u81f4\u57fa\u56e0\u7ec4\u914d\u7f6e\u65e0\u6548\uff08\u4f8b\u5982\u5171\u4eab\u7b56\u7565\u4e0d\u517c\u5bb9\uff09\uff0c\u5219\u901a\u8fc7\u91cd\u65b0\u91c7\u6837\u6709\u6548\u7684\u57fa\u56e0\u7ec4\u503c\u6765\u4fee\u590d\u8fd9\u4e9b\u65e0\u6548\u8fde\u63a5\u3002<\/li>\n<\/ol>\n\n\n\n<p>\u901a\u8fc7\u8fd9\u4e9b\u6b65\u9aa4\uff0cSTAR\u9010\u6e10\u6f14\u5316\u51fa\u4f18\u5316\u7684\u67b6\u6784\uff0c\u6700\u7ec8\u4f7f\u5f97\u5408\u6210\u7684\u6a21\u578b\u5728\u591a\u4e2a\u8d28\u91cf\u548c\u6548\u7387\u76ee\u6807\u4e0a\u8fbe\u5230\u6700\u4f73\u3002\u4f5c\u8005\u6307\u51fa\uff0c\u901a\u8fc7\u7ed3\u5408LIV\u7684\u641c\u7d22\u7a7a\u95f4\u3001\u57fa\u56e0\u7ec4\u7f16\u7801\u4ee5\u53ca\u9488\u5bf9\u53d8\u5f02\u548c\u91cd\u7ec4\u7684\u6307\u5bfc\u539f\u5219\uff0cSTAR\u53ef\u4ee5\u6709\u6548\u5730\u83b7\u5f97\u7a33\u5b9a\u4e14\u6027\u80fd\u4f18\u826f\u7684\u5019\u9009\u67b6\u6784\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e94\u3001\u5b9e\u9a8c\u7ed3\u679c\u4e0e\u5206\u6790<\/h5>\n\n\n\n<p>\u5728\u5b9e\u9a8c\u4e2d\uff0c\u4f5c\u8005\u5c06STAR\u5e94\u7528\u4e8e\u81ea\u56de\u5f52\u8bed\u8a00\u5efa\u6a21\uff0c\u91cd\u70b9\u8bc4\u4f30\u4e86STAR\u5728\u591a\u4e2a\u8d28\u91cf\u548c\u6548\u7387\u76ee\u6807\u4e0a\u7684\u4f18\u5316\u80fd\u529b\uff0c\u5305\u62ec\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8d28\u91cf\uff08\u56f0\u60d1\u5ea6PPL\uff09<\/strong>\uff1a\u8fd9\u662f\u8861\u91cf\u8bed\u8a00\u6a21\u578b\u9884\u6d4b\u8d28\u91cf\u7684\u4e3b\u8981\u6307\u6807\u3002<\/li>\n\n\n\n<li><strong>\u6a21\u578b\u5927\u5c0f\uff08\u53c2\u6570\u6570\u91cf\uff09<\/strong>\uff1a\u4f18\u5316\u53c2\u6570\u6570\u91cf\u4ee5\u51cf\u5c11\u6a21\u578b\u7684\u5185\u5b58\u5360\u7528\u3002<\/li>\n\n\n\n<li><strong>\u63a8\u7406\u7f13\u5b58\u5927\u5c0f<\/strong>\uff1a\u8fd9\u5bf9\u4e8e\u63a8\u7406\u6548\u7387\u81f3\u5173\u91cd\u8981\uff0c\u5c24\u5176\u662f\u5728\u5904\u7406\u957f\u5e8f\u5217\u8f93\u5165\u65f6\u3002<\/li>\n<\/ul>\n\n\n\n<h6 class=\"wp-block-heading\">5.1 \u8d28\u91cf\u4f18\u5316\u5b9e\u9a8c<\/h6>\n\n\n\n<p>\u5b9e\u9a8c\u663e\u793a\uff0c\u7ecf\u8fc7STAR\u6f14\u5316\u7684\u67b6\u6784\u5728RedPajama\u6570\u636e\u96c6\u4e0a\u7684\u8bc4\u4f30\u56f0\u60d1\u5ea6\u663e\u8457\u4f4e\u4e8e\u624b\u52a8\u8bbe\u8ba1\u7684Transformer++\u548cStripe\u6df7\u5408\u6a21\u578b\u3002\u5728\u591a\u4e2a\u4e0b\u6e38\u4efb\u52a1\u4e0a\uff0c\u5982HellaSwag\u3001ARC-Easy\u3001Winogrande\u7b49\uff0cSTAR\u7684\u8868\u73b0\u8d85\u8d8a\u4e86\u6807\u51c6\u7684\u6df7\u5408\u67b6\u6784\uff0c\u663e\u793a\u4e86\u66f4\u9ad8\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">5.2 \u53c2\u6570\u9ad8\u6548\u6a21\u578b\u7684\u5408\u6210<\/h6>\n\n\n\n<p>\u5728\u540c\u65f6\u4f18\u5316\u8d28\u91cf\u548c\u6a21\u578b\u5927\u5c0f\u7684\u5b9e\u9a8c\u4e2d\uff0cSTAR\u6210\u529f\u51cf\u5c11\u4e86\u53c2\u6570\u6570\u91cf\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u4f18\u5f02\u7684\u8d28\u91cf\u8868\u73b0\u3002\u4e0e\u624b\u52a8\u8bbe\u8ba1\u7684\u57fa\u51c6\u6a21\u578b\u76f8\u6bd4\uff0cSTAR\u5408\u6210\u7684\u6a21\u578b\u53c2\u6570\u6570\u91cf\u51cf\u5c11\u4e86\u7ea613%\uff0c\u800c\u6027\u80fd\u4ecd\u7136\u8fbe\u5230\u6216\u8d85\u8fc7\u57fa\u51c6\u6a21\u578b\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">5.3 \u9ad8\u6548\u63a8\u7406\u7f13\u5b58\u6a21\u578b\u7684\u5408\u6210<\/h6>\n\n\n\n<p>\u4e3a\u4e86\u51cf\u5c11\u63a8\u7406\u7f13\u5b58\u7684\u5927\u5c0f\uff0cSTAR\u8fd8\u4f18\u5316\u4e86\u56f0\u60d1\u5ea6\u548c\u7f13\u5b58\u5927\u5c0f\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5728\u76f8\u540c\u7684\u5e8f\u5217\u957f\u5ea6\u4e0b\uff0cSTAR\u5408\u6210\u7684\u6a21\u578b\u7f13\u5b58\u5927\u5c0f\u6bd4\u6807\u51c6Transformer++\u51cf\u5c11\u4e8690%\uff0c\u5e76\u4e14\u5728\u591a\u4e2a\u4e0b\u6e38\u4efb\u52a1\u4e0a\u4fdd\u6301\u4e86\u9ad8\u8d28\u91cf\u3002<\/p>\n\n\n\n<h6 class=\"wp-block-heading\">5.4 \u53d1\u73b0\u67b6\u6784\u7684\u5408\u6210\u6a21\u5f0f<\/h6>\n\n\n\n<p>\u901a\u8fc7\u8fdb\u5316\u7b97\u6cd5\uff0cSTAR\u8fd8\u80fd\u591f\u81ea\u52a8\u53d1\u73b0\u67b6\u6784\u4e2d\u6709\u52a9\u4e8e\u6027\u80fd\u63d0\u5347\u7684\u6a21\u5f0f\uff08\u79f0\u4e3a\u201c\u67b6\u6784\u56fe\u6848\u201d\uff09\u3002\u8fd9\u4e9b\u56fe\u6848\u5305\u62ec\u4e86\u5bf9\u7279\u5b9aLIV\u7684\u504f\u597d\uff0c\u4f8b\u5982\u77ed\u5377\u79ef\u3001\u5206\u7ec4\u67e5\u8be2\u6ce8\u610f\u529b\u7b49\u3002\u8fd9\u4e9b\u67b6\u6784\u56fe\u6848\u5728\u591a\u4e2a\u4efb\u52a1\u4e0a\u90fd\u8868\u73b0\u51fa\u4e86\u8f83\u597d\u7684\u6027\u80fd\uff0c\u8fdb\u4e00\u6b65\u8bc1\u660e\u4e86STAR\u65b9\u6cd5\u5728\u67b6\u6784\u5408\u6210\u4e2d\u7684\u7075\u6d3b\u6027\u548c\u6709\u6548\u6027\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u516d\u3001\u603b\u7ed3\u4e0e\u672a\u6765\u5c55\u671b<\/h5>\n\n\n\n<p>\u8bba\u6587\u603b\u7ed3\u4e86STAR\u5728\u67b6\u6784\u5408\u6210\u4e2d\u7684\u4f18\u52bf\uff0c\u5e76\u5c55\u793a\u4e86\u5b83\u5728\u4f18\u5316\u6a21\u578b\u8d28\u91cf\u548c\u6548\u7387\u4e0a\u7684\u663e\u8457\u63d0\u5347\u3002\u76f8\u6bd4\u4e8e\u73b0\u6709\u7684\u624b\u52a8\u6216\u81ea\u52a8\u5316\u67b6\u6784\u8bbe\u8ba1\u65b9\u6cd5\uff0cSTAR\u7ed3\u5408\u4e86\u5c42\u6b21\u5316\u7684\u8bbe\u8ba1\u7a7a\u95f4\u548c\u8fdb\u5316\u7b97\u6cd5\uff0c\u4f7f\u5f97\u5728\u590d\u6742\u4efb\u52a1\u4e0a\u7684\u6a21\u578b\u67b6\u6784\u4f18\u5316\u6210\u4e3a\u53ef\u80fd\u3002\u6b64\u5916\uff0c\u4f5c\u8005\u6307\u51faSTAR\u5177\u5907\u826f\u597d\u7684\u53ef\u6269\u5c55\u6027\uff0c\u53ef\u7528\u4e8e\u66f4\u5927\u89c4\u6a21\u7684\u6a21\u578b\u6216\u66f4\u590d\u6742\u7684\u4efb\u52a1\uff0c\u4f8b\u5982\u5c06LIV\u67b6\u6784\u4e0e\u4f20\u7edf\u7684\u7f29\u653e\u6cd5\u5219\u7ed3\u5408\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u6a21\u578b\u5728\u5927\u89c4\u6a21\u8bad\u7ec3\u4e2d\u7684\u8868\u73b0\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">\u5173\u4e8e<strong><a href=\"https:\/\/www.liquid.ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">Liquid AI<\/a><\/strong>\u516c\u53f8<\/h5>\n\n\n\n<p>Liquid AI\u662f\u4e00\u5bb6\u7531\u9ebb\u7701\u7406\u5de5\u5b66\u9662\uff08MIT\uff09\u8ba1\u7b97\u673a\u79d1\u5b66\u4e0e\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\uff08CSAIL\uff09\u524d\u7814\u7a76\u4eba\u5458\u5171\u540c\u521b\u7acb\u7684\u521d\u521b\u516c\u53f8\uff0c\u4e13\u6ce8\u4e8e\u5f00\u53d1\u65b0\u578b\u4eba\u5de5\u667a\u80fd\u6a21\u578b\uff0c\u7279\u522b\u662f\u6db2\u6001\u795e\u7ecf\u7f51\u7edc\uff08Liquid Neural Networks\uff0cLNN\uff09\u3002Liquid AI\u7684\u521b\u59cb\u56e2\u961f\u5305\u62ecMIT CSAIL\u4e3b\u4efbDaniela Rus\u6559\u6388\uff0c\u4ee5\u53ca\u535a\u58eb\u540e\u7814\u7a76\u5458Ramin Hasani\u3001Mathias Lechner\u548c\u535a\u58eb\u751fAlexander Amini\u3002<\/p>\n\n\n\n<p>\u6db2\u6001\u795e\u7ecf\u7f51\u7edc\uff08LNN\uff09\u662f\u4e00\u79cd\u65b0\u578b\u7684\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\uff0c\u5176\u7279\u70b9\u662f\u7ed3\u6784\u52a8\u6001\u53ef\u53d8\uff0c\u80fd\u591f\u6839\u636e\u8f93\u5165\u6570\u636e\u5b9e\u65f6\u8c03\u6574\u81ea\u8eab\u53c2\u6570\uff0c\u4ece\u800c\u63d0\u9ad8\u5bf9\u65b0\u73af\u5883\u7684\u9002\u5e94\u6027\u3002 \u4e0e\u4f20\u7edf\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u76f8\u6bd4\uff0cLNN\u5728\u5904\u7406\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u548c\u9700\u8981\u5b9e\u65f6\u54cd\u5e94\u7684\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\u3002<\/p>\n\n\n\n<p>2024\u5e749\u6708\uff0cLiquid AI\u63a8\u51fa\u4e86\u9996\u6279\u591a\u6a21\u6001AI\u6a21\u578b\uff0c\u79f0\u4e3a\u6db2\u6001\u57fa\u7840\u6a21\u578b\uff08Liquid Foundation Models\uff0cLFM\uff09\u3002 \u8fd9\u4e9b\u6a21\u578b\u5e76\u975e\u57fa\u4e8e\u4f20\u7edf\u7684Transformer\u67b6\u6784\uff0c\u800c\u662f\u91c7\u7528\u4e86\u57fa\u4e8e\u52a8\u6001\u7cfb\u7edf\u7406\u8bba\u7684\u65b0\u578b\u8ba1\u7b97\u5355\u5143\u3002 LFM\u5728\u591a\u4e2a\u89c4\u6a21\u4e0a\u5b9e\u73b0\u4e86\u5148\u8fdb\u7684\u6027\u80fd\uff0c\u540c\u65f6\u4fdd\u6301\u8f83\u5c0f\u7684\u5185\u5b58\u5360\u7528\u548c\u66f4\u9ad8\u7684\u63a8\u7406\u6548\u7387\u3002<\/p>\n\n\n\n<p>Liquid AI\u7684\u6a21\u578b\u5728\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u4f18\u5f02\uff0c\u7279\u522b\u662f\u5728\u5904\u7406\u957f\u5e8f\u5217\u8f93\u5165\u65f6\uff0c\u80fd\u591f\u9ad8\u6548\u5730\u5229\u7528\u4e0a\u4e0b\u6587\u957f\u5ea6\uff0c\u663e\u8457\u51cf\u5c11\u5185\u5b58\u5360\u7528\u3002 \u4f8b\u5982\uff0cLFM-3B\u6a21\u578b\u4ec5\u970016GB\u5185\u5b58\uff0c\u800c\u540c\u7b49\u89c4\u6a21\u7684\u5176\u4ed6\u6a21\u578b\u53ef\u80fd\u9700\u8981\u8d85\u8fc748GB\u5185\u5b58\u3002<\/p>\n\n\n\n<p>Liquid AI\u7684\u6280\u672f\u9002\u7528\u4e8e\u591a\u79cd\u6570\u636e\u6a21\u6001\uff0c\u5305\u62ec\u6587\u672c\u3001\u97f3\u9891\u3001\u56fe\u50cf\u548c\u89c6\u9891\uff0c\u5177\u6709\u5e7f\u6cdb\u7684\u5e94\u7528\u6f5c\u529b\u3002 \u5176\u9ad8\u6548\u7684\u5185\u5b58\u4f7f\u7528\u548c\u63a8\u7406\u80fd\u529b\uff0c\u4f7f\u5176\u5728\u8d44\u6e90\u53d7\u9650\u7684\u73af\u5883\u4e2d\u4e5f\u80fd\u53d1\u6325\u4f5c\u7528\uff0c\u5982\u79fb\u52a8\u8bbe\u5907\u548c\u8fb9\u7f18\u8ba1\u7b97\u573a\u666f\u3002<\/p>\n\n\n\n<p>2023\u5e7412\u6708\uff0cLiquid AI\u5ba3\u5e03\u5b8c\u62103750\u4e07\u7f8e\u5143\u7684\u79cd\u5b50\u8f6e\u878d\u8d44\uff0c\u4f30\u503c\u8fbe\u52303.03\u4ebf\u7f8e\u5143\u3002 \u6295\u8d44\u8005\u5305\u62ecGitHub\u8054\u5408\u521b\u59cb\u4ebaTom Preston-Werner\u3001Shopify\u8054\u5408\u521b\u59cb\u4ebaTobias L\u00fctke\u548cRed Hat\u8054\u5408\u521b\u59cb\u4ebaBob Young\u7b49\u77e5\u540d\u4eba\u58eb\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587STAR: Synthesis of Tailored Architectures\uff08\u300aSTAR: \u5408\u6210\u5b9a\u5236 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","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":"","ast-disable-related-posts":"","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":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"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":""},"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":[8],"tags":[39,63,95,90,86],"class_list":{"0":"post-5245","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"hentry","6":"category-tech","7":"tag-ai","9":"tag-transformer","10":"tag-multimodal","11":"tag-llm"},"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u8bba\u6587STAR: Synthesis of Tailored Architectures\uff08\u300aSTAR: \u5408\u6210\u5b9a\u5236&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5245","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5245"}],"version-history":[{"count":1,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5245\/revisions"}],"predecessor-version":[{"id":5246,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5245\/revisions\/5246"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5245"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5245"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5245"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}