{"id":39832,"date":"2025-04-02T01:02:42","date_gmt":"2025-04-01T17:02:42","guid":{"rendered":"https:\/\/17aitech.com\/?p=39832"},"modified":"2025-04-02T01:02:42","modified_gmt":"2025-04-01T17:02:42","slug":"deepseek-r1%e4%b9%8b%e5%90%8e%e6%8e%a8%e7%90%86%e6%a8%a1%e5%9e%8b%e5%8f%91%e5%b1%95%e5%a6%82%e4%bd%95%ef%bc%9fraschka%e9%95%bf%e6%96%87%e6%a2%b3%e7%90%86%e5%90%8er1%e6%97%b6%e4%bb%a314%e7%af%87","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=39832","title":{"rendered":"DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-04-01-9\" target=\"_blank\">DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587<\/a><\/p>\n<section>\u8fd1\u65e5\uff0cDeepseek R1 \u7b49\u4e00\u7cfb\u5217\u63a8\u7406\u5927\u8bed\u8a00\u6a21\u578b\u5df2\u6210\u4e3a 2025 \u5e74\u6700\u70ed\u95e8\u7684\u8bdd\u9898\u4e4b\u4e00\u3002\u5728\u8fc7\u53bb\u7684\u51e0\u5468\u91cc\uff0c\u7814\u7a76\u4eba\u5458\u63a8\u51fa\u4e86\u8bb8\u591a\u6539\u8fdb\u63a8\u7406\u7684\u65b0\u7b56\u7565\uff0c\u5305\u62ec\u6269\u5c55\u7b80\u5355\u6d4b\u8bd5 &#8211; \u65f6\u95f4\u89c4\u6a21\u5316\uff08S1\uff09\u3001\u5173\u8054\u601d\u7ef4\u94fe\uff08Chain of Associated thoughts\uff09\u3001Inner Transformer \u7b49\u65b9\u6cd5\u3002<\/section>\n<section><\/section>\n<section>\u4e0d\u4ec5\u5982\u6b64\uff0c\u8fd8\u6709\u6765\u81ea\u817e\u8baf\u5b9e\u9a8c\u5ba4\u7684\u7814\u7a76\u4eba\u5458\u63a2\u7d22\u4e86 Thoughts Are All Over the Place\uff0c\u901a\u8fc7\u8861\u91cf\u4e0d\u6b63\u786e\u7b54\u6848\u4e2d\u7684 token \u6548\u7387\u6765\u9f13\u52b1\u6a21\u578b\u5bf9\u6bcf\u6761\u63a8\u7406\u8def\u5f84\u8fdb\u884c\u66f4\u6df1\u5165\u7684\u63a2\u7d22\uff1b\u6765\u81ea\u7f8e\u56fd\u9a6c\u91cc\u5170\u5927\u5b66\u548c\u6a61\u6811\u5cad\u56fd\u5bb6\u5b9e\u9a8c\u5ba4\u7b49\u673a\u6784\u8054\u5408\u63d0\u51fa\u4e86 Recurrent Block\uff0c\u901a\u8fc7\u91cd\u590d\u8c03\u7528\u540c\u4e00\u4e2a\u5faa\u73af\u4f53\u7684\u65b9\u5f0f\uff0c\u5728\u63a8\u7406\u9636\u6bb5\u53ef\u4ee5\u8fed\u4ee3\u4efb\u610f\u591a\u6b21\uff1b\u4ee5\u53ca\u6765\u81ea\u7f8e\u56fd Zoom \u89c6\u9891\u901a\u8baf\u516c\u53f8\u7684\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u4e86 Chain of Draft\uff08CoD\uff09\uff0c\u57fa\u4e8e\u66f4\u63a5\u8fd1\u4eba\u7c7b\u63a8\u7406\u7684\u63d0\u793a\u7b56\u7565\u63d0\u51fa\u4e86\u8349\u7a3f\u56fe\uff0c\u8fd9\u662f\u4e00\u79cd\u4f18\u5148\u8003\u8651\u6548\u7387\u548c\u63a8\u7406\u7684\u63d0\u793a\u8bcd\u7b56\u7565\u3002\u00a0<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png\"><\/a><\/section>\n<section>\u76ee\u524d\u5927\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u6280\u672f\u7814\u7a76\u867d\u5728\u7ec6\u5206\u9886\u57df\u53d6\u5f97\u7a81\u7834\uff0c\u4f46\u7814\u7a76\u8def\u5f84\u5206\u6563\u3001\u6210\u679c\u788e\u7247\u5316\uff0c\u7f3a\u5c11\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u7814\u7a76\u73b0\u72b6\u7684\u603b\u7ed3\uff0c\u56e0\u6b64\u7cfb\u7edf\u6027\u603b\u7ed3\u6210\u7834\u5c40\u5173\u952e\u3002<\/section>\n<section><\/section>\n<section>\u524d\u7edf\u8ba1\u5b66\u6559\u6388\uff0c\u73b0 AI\/ML \u7814\u7a76\u5458 Sebastian Raschka \u5728\u7efc\u8ff0\u300aThe State of LLM Reasoning Models\u300b\u4e2d\u63a2\u8ba8\u5e76\u603b\u7ed3\u4e86\u63a8\u7406 LLM \u7684\u6700\u65b0\u7814\u7a76\u8fdb\u5c55\uff0c\u7279\u522b\u5173\u6ce8\u81ea DeepSeek R1 \u53d1\u5e03\u4ee5\u6765\u51fa\u73b0\u7684\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-880b2cd5eb9d23072c04ec6cfa54a98f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-880b2cd5eb9d23072c04ec6cfa54a98f.png\"><\/a><\/section>\n<section>\u539f\u6587\u5730\u5740\uff1ahttps:\/\/magazine.sebastianraschka.com\/p\/state-of-llm-reasoning-and-inference-scaling<\/section>\n<section><\/section>\n<section>\u987a\u5e26\u4e00\u63d0\uff0cSebastian Raschka \u524d\u6bb5\u65f6\u95f4\u8fd8\u66fe\u5199\u8fc7\u53e6\u4e00\u7bc7\u4e0e\u63a8\u7406\u6a21\u578b\u76f8\u5173\u7684\u957f\u7bc7\u535a\u5ba2\uff0c\u611f\u5174\u8da3\u7684\u8bfb\u8005\u53ef\u8bbf\u95ee\u300a<a data-itemshowtype=\"0\" data-linktype=\"2\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzA3MzI4MjgzMw==&amp;mid=2650954131&amp;idx=2&amp;sn=1d90cf8ef42ebe40d91b27f5ceca4b47&amp;scene=21#wechat_redirect\" target=\"_blank\">Sebastian Raschka\uff1a\u5173\u4e8e DeepSeek R1 \u548c\u63a8\u7406\u6a21\u578b\uff0c\u6211\u6709\u51e0\u70b9\u770b\u6cd5<\/a>\u300b\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-fdd6efbf15dfae04d563d447bb86bd35.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-fdd6efbf15dfae04d563d447bb86bd35.png\"><\/a><\/section>\n<section>\u9996\u5148\u8be5\u6587\u7ae0\u7b80\u8981\u4ecb\u7ecd\u4e86\u4ec0\u4e48\u662f LLM \u63a8\u7406\u6a21\u578b\uff1f<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-f1260cf5b6f479187b8ebf6d0988a9ac.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-f1260cf5b6f479187b8ebf6d0988a9ac.png\"><\/a><\/section>\n<section>\u76f8\u8f83\u4e8e\u4ec5\u5206\u4eab\u6700\u7ec8\u7b54\u6848\u7684\u7b80\u5355\u95ee\u7b54\u5f0f LLM \u4e0d\u540c\uff0c\u63a8\u7406\u6a21\u578b\u662f\u4e00\u79cd\u901a\u8fc7\u751f\u6210\u4e2d\u95f4\u6b65\u9aa4\u6216\u663e\u5f0f\u300c\u601d\u7ef4\u300d\u8fc7\u7a0b\u6765\u89e3\u51b3\u591a\u6b65\u9aa4\u95ee\u9898\u7684 LLM\u3002\u5176\u6838\u5fc3\u7a81\u7834\u5728\u4e8e\u6a21\u62df\u4eba\u7c7b\u601d\u7ef4\u8fc7\u7a0b\uff0c\u4e3b\u8981\u5448\u73b0\u4e09\u5927\u7279\u5f81\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8fc7\u7a0b\u900f\u660e\u5316\uff1a\u901a\u8fc7\u601d\u7ef4\u94fe\uff08CoT\uff09\u7b49\u6280\u672f\uff0c\u5c06\u95ee\u9898\u62c6\u89e3\u4e3a\u53ef\u89e3\u91ca\u7684\u63a8\u7406\u6b65\u9aa4\uff0c\u4f7f\u6a21\u578b\u51b3\u7b56\u8def\u5f84\u53ef\u89c6\u5316\u3002<\/section>\n<\/li>\n<li>\n<section>\u8ba1\u7b97\u52a8\u6001\u5316\uff1a\u91c7\u7528\u6d4b\u8bd5\u65f6\u95f4\u6269\u5c55\uff08Test-Time Scaling\uff09\u7b49\u7b56\u7565\u5728\u63a8\u7406\u9636\u6bb5\u52a8\u6001\u5206\u914d\u66f4\u591a\u8ba1\u7b97\u8d44\u6e90\u5904\u7406\u590d\u6742\u5b50\u95ee\u9898\u3002<\/section>\n<\/li>\n<li>\n<section>\u8bad\u7ec3\u5f3a\u5316\uff1a\u7ed3\u5408\u5f3a\u5316\u5b66\u4e60\uff08\u5982 RLHF\uff09\u3001\u5bf9\u6297\u8bad\u7ec3\u7b49\u65b9\u6cd5\uff0c\u5229\u7528\u9ad8\u96be\u5ea6\u63a8\u7406\u4efb\u52a1\u6570\u636e\u96c6\uff08\u5982 MATH\u3001CodeContests\uff09\u8fdb\u884c\u5fae\u8c03\uff0c\u63d0\u5347\u7b26\u53f7\u63a8\u7406\u4e0e\u903b\u8f91\u8fde\u8d2f\u6027\u3002<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-203b00eaa0dac03315250c063b02ffc7.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-203b00eaa0dac03315250c063b02ffc7.png\"><\/a><\/section>\n<section>\u901a\u8fc7\u6211\u4eec\u65e5\u5e38\u548c Deepseek \u7684\u4ea4\u6d41\u53ef\u4ee5\u770b\u5230\uff0c\u63a8\u7406\u6a21\u578b\u660e\u786e\u7684\u5c55\u793a\u4e86\u5176\u601d\u7ef4\u8fc7\u7a0b\uff0c\u7ed3\u5408\u5b9e\u9645\u5e94\u7528\u573a\u666f\u4e2d\u6765\u770b\uff0c\u8fd9\u6709\u52a9\u4e8e\u6211\u4eec\u7406\u89e3\u6a21\u578b\u7684\u51b3\u7b56\u8fc7\u7a0b\uff0c\u8fd9\u5728\u9700\u8981\u9ad8\u5ea6\u4fe1\u4efb\u7684\u5e94\u7528\u573a\u666f\u4e2d\u5c24\u4e3a\u91cd\u8981\uff0c\u6bd4\u5982\u533b\u7597\u8bca\u65ad\u6216\u91d1\u878d\u6295\u8d44\u3002<\/section>\n<section><\/section>\n<section>\u90a3\u4e48\u5982\u4f55\u63d0\u9ad8\u5927\u6a21\u578b\u63a8\u7406\u7684\u63a8\u7406\u80fd\u529b\u5462\uff1f<\/section>\n<section><\/section>\n<section>\u8be5\u6587\u7ae0\u8868\u793a\u4e00\u822c\u6765\u8bf4\u6709\u4e24\u79cd\u6838\u5fc3\u7b56\u7565\uff1a<\/section>\n<section><\/section>\n<ol>\n<li>\n<section>\u589e\u52a0\u8bad\u7ec3\u8ba1\u7b97\uff0c\u5373\u901a\u8fc7\u6269\u5c55\u8bad\u7ec3\u6570\u636e\u91cf\u3001\u5f3a\u5316\u5b66\u4e60\u6216\u9488\u5bf9\u7279\u5b9a\u4efb\u52a1\u7684\u5fae\u8c03\u6765\u589e\u5f3a\u6a21\u578b\u80fd\u529b\uff1b<\/section>\n<\/li>\n<li>\n<section>\u589e\u52a0\u63a8\u7406\u8ba1\u7b97\uff0c\u4e5f\u79f0\u4e3a\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u6216\u6d4b\u8bd5\u65f6\u95f4\u6269\u5c55\uff0c\u5373\u5728\u6a21\u578b\u751f\u6210\u8f93\u51fa\u65f6\u5206\u914d\u66f4\u591a\u8ba1\u7b97\u8d44\u6e90\uff0c\u5141\u8bb8\u6a21\u578b \u201c\u601d\u8003\u66f4\u957f\u65f6\u95f4\u201d \u6216\u6267\u884c\u66f4\u590d\u6742\u7684\u63a8\u7406\u6b65\u9aa4\u3002<\/section>\n<\/li>\n<\/ol>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-63e433de0d97dbec6a2246f9404794d2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-63e433de0d97dbec6a2246f9404794d2.png\"><\/a><\/section>\n<section>\u5982\u4e0a\u56fe\u6240\u793a\uff0c\u7814\u7a76\u4eba\u5458\u5206\u522b\u901a\u8fc7\u8bad\u7ec3\u65f6\u95f4\u8ba1\u7b97\u6216\u6d4b\u8bd5\u65f6\u95f4\u8ba1\u7b97\u6765\u6539\u8fdb\u63a8\u7406\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u867d\u7136\u8fd9\u4e24\u79cd\u7b56\u7565\u53ef\u4ee5\u72ec\u7acb\u4f7f\u7528\uff0c\u4f46\u5b9e\u9645\u5e94\u7528\u4e2d\uff0cLLM \u7684\u63a8\u7406\u80fd\u529b\u4f18\u5316\u901a\u5e38\u9700\u8981\u7ed3\u5408\u4e24\u8005\u3002<\/section>\n<section><\/section>\n<section>\u5177\u4f53\u6765\u8bf4\uff0c\u901a\u8fc7\u5927\u91cf\u8bad\u7ec3\u8ba1\u7b97\uff08\u5982\u4f7f\u7528\u5f3a\u5316\u5b66\u4e60\u6216\u4e13\u95e8\u6570\u636e\u96c6\u7684\u6df1\u5ea6\u5fae\u8c03\uff09\u6765\u63d0\u5347\u6a21\u578b\u7684\u57fa\u7840\u80fd\u529b\uff0c\u540c\u65f6\u901a\u8fc7\u589e\u52a0\u63a8\u7406\u8ba1\u7b97\uff08\u5982\u52a8\u6001\u6269\u5c55\u63a8\u7406\u6b65\u9aa4\u6216\u6267\u884c\u989d\u5916\u8ba1\u7b97\uff09\u6765\u8fdb\u4e00\u6b65\u589e\u5f3a\u5176\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u3002<\/section>\n<section><\/section>\n<section>\u8fd9\u79cd\u8054\u5408\u7b56\u7565\u7684\u4f7f\u7528\u80fd\u591f\u663e\u8457\u63d0\u5347 LLM \u5728\u6570\u5b66\u63a8\u5bfc\u3001\u4ee3\u7801\u751f\u6210\u7b49\uff0c\u4fc3\u8fdb\u4e86 LLM \u5728\u591a\u6b65\u63a8\u7406\u7684\u4efb\u52a1\u4e2d\u7684\u51c6\u786e\u6027\u548c\u53ef\u9760\u6027\u3002<\/section>\n<section><\/section>\n<section><strong>\u63a8\u7406\u6a21\u578b\u7684\u4e3b\u8981\u7c7b\u522b<\/strong><\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u6df1\u5165\u63a2\u7a76\u63a8\u7406\u6a21\u578b\u7684\u5f00\u53d1\u8fc7\u7a0b\uff0c\u4f5c\u8005\u5728\u7efc\u8ff0\u4e2d\u8fd8\u5217\u4e3e\u4e86\u4e00\u4e9b\u589e\u5f3a\u6a21\u578b\u63a8\u7406\u80fd\u529b\u7684\u4e3b\u8981\u65b9\u6cd5\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55<\/section>\n<\/li>\n<li>\n<section>\u7eaf\u5f3a\u5316\u5b66\u4e60<\/section>\n<\/li>\n<li>\n<section>\u5f3a\u5316\u5b66\u4e60\u548c\u76d1\u7763\u5fae\u8c03<\/section>\n<\/li>\n<li>\n<section>\u76d1\u7763\u5fae\u8c03\u548c\u6a21\u578b\u63d0\u70bc<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u4ece\u56fe\u4e2d\u80fd\u591f\u6e05\u6670\u770b\u5230\uff0c\u65e0\u8bba\u662f\u91c7\u7528\u7eaf\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5\uff0c\u8fd8\u662f\u5c06\u5f3a\u5316\u5b66\u4e60\u4e0e\u76d1\u7763\u5b66\u4e60\u76f8\u7ed3\u5408\uff0c\u4ea6\u6216\u662f\u5355\u7eaf\u8fd0\u7528\u76d1\u7763\u5b66\u4e60\u65b9\u6cd5\uff0c\u6a21\u578b\u901a\u5e38\u90fd\u4f1a\u751f\u6210\u8f83\u957f\u7684\u54cd\u5e94\u5185\u5bb9\uff0c\u5176\u4e2d\u6db5\u76d6\u4e86\u63a8\u7406\u7684\u4e2d\u95f4\u6b65\u9aa4\u4ee5\u53ca\u8be6\u7ec6\u89e3\u91ca\u3002\u7136\u800c\uff0c\u63a8\u7406\u6210\u672c\u662f\u4e0e\u54cd\u5e94\u957f\u5ea6\u5448\u6b63\u76f8\u5173\u7684\uff0c\u8fd9\u5c31\u610f\u5473\u7740\uff0c\u4e0a\u8ff0\u8fd9\u4e9b\u8bad\u7ec3\u65b9\u6cd5\u672c\u8d28\u4e0a\u4e0e\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u5bc6\u5207\u76f8\u5173\u3002<\/section>\n<section><\/section>\n<section>\u9488\u5bf9\u8fd9\u4e00\u95ee\u9898\uff0c\u8fd9\u7bc7\u7efc\u8ff0\u805a\u7126\u4e8e\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u7684\u7814\u7a76\uff0c\u7740\u91cd\u63a2\u8ba8\u4e86\u90a3\u4e9b\u660e\u786e\u8c03\u8282\u751f\u6210 token \u7684\u6280\u672f\uff0c\u5305\u62ec\u901a\u8fc7\u989d\u5916\u7684\u91c7\u6837\u7b56\u7565\u3001\u81ea\u6211\u6821\u6b63\u673a\u5236\u7b49\u65b9\u6cd5\u3002\u8fd9\u4e9b\u6280\u672f\u901a\u8fc7\u4e0d\u540c\u7684\u4f18\u5316\u65b9\u5f0f\uff0c\u76f4\u63a5\u4f5c\u7528\u4e8e\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u8fd9\u4e00\u5173\u952e\u7ef4\u5ea6\uff0c\u4ece\u800c\u663e\u8457\u63d0\u5347\u8ba1\u7b97\u6548\u7387\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-3d3f3aa7f935149ddd0e9e81fa524e8f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-3d3f3aa7f935149ddd0e9e81fa524e8f.png\"><\/a><\/section>\n<section><strong>\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u65b9\u6cd5<\/strong><\/section>\n<section><\/section>\n<section>\u5728\u8be5\u7efc\u8ff0\u91cc\uff0c\u5e7f\u6cdb\u6db5\u76d6\u4e86\u6d4b\u8bd5\u65f6\u6269\u5c55\u3001\u8bad\u7ec3\u65f6\u6269\u5c55\u3001\u63a8\u7406\u65f6\u6269\u5c55\u4ee5\u53ca\u63a8\u7406\u8ba1\u7b97\u65f6\u95f4\u6269\u5c55\u7b49\u591a\u4e2a\u65b9\u9762\u7684\u5185\u5bb9\u3002\u76ee\u524d\u800c\u8a00\uff0c\u6700\u7b80\u5355\u76f4\u63a5\u7684\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u65b9\u6cd5\u5373\u901a\u8fc7\u589e\u52a0\u63a8\u7406\u8fc7\u7a0b\u4e2d\u7684\u8ba1\u7b97\u8d44\u6e90\u6765\u6539\u5584 LLM \u7684\u63a8\u7406\u3002\u5176\u80cc\u540e\u6539\u5584\u63a8\u7406\u6548\u679c\u7684\u539f\u7406\uff0c\u53ef\u505a\u5982\u4e0b\u5f62\u8c61\u7c7b\u6bd4\uff1a\u5f53\u7ed9\u4e88\u4eba\u7c7b\u66f4\u591a\u601d\u8003\u65f6\u95f4\u65f6\uff0c\u4ed6\u4eec\u4f1a\u7ed9\u51fa\u66f4\u597d\u7684\u53cd\u5e94\uff0c\u540c\u7406\uff0cLLM \u53ef\u4ee5\u901a\u8fc7\u9f13\u52b1\u5728\u751f\u6210\u8fc7\u7a0b\u4e2d\u8fdb\u884c\u66f4\u591a\u300c\u601d\u8003\u300d\u7684\u6280\u672f\u6765\u6539\u8fdb\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-a4460b00bf9ede5ae36f1398caaff0ef.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-a4460b00bf9ede5ae36f1398caaff0ef.png\"><\/a><\/section>\n<section>\u63d0\u793a\u8bcd\u5de5\u7a0b\u4e5f\u662f\u4e00\u79cd\u65b9\u6cd5\uff0c\u5982\u601d\u7ef4\u94fe (CoT) \u63d0\u793a\uff0c\u5176\u4e2d \u201c\u9010\u6b65\u601d\u8003\u201d \u7b49\u77ed\u8bed\u6307\u5bfc\u6a21\u578b\u751f\u6210\u4e2d\u95f4\u63a8\u7406\u6b65\u9aa4\u3002\u8fd9\u63d0\u9ad8\u4e86\u590d\u6742\u95ee\u9898\u7684\u51c6\u786e\u6027\uff0c\u4f46\u5bf9\u4e8e\u7b80\u5355\u7684\u4e8b\u5b9e\u67e5\u8be2\u800c\u8a00\u662f\u4e0d\u5fc5\u8981\u7684\u3002\u5e76\u4e14\uff0c\u7531\u4e8e CoT \u63d0\u793a\u4f1a\u4fc3\u4f7f\u6a21\u578b\u751f\u6210\u66f4\u591a\u7684 tokens\uff0c\u8fd9\u5b9e\u9645\u4e0a\u4f1a\u589e\u52a0\u63a8\u7406\u6210\u672c\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-28ec170d3d41ad56888901a5af6029b2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-28ec170d3d41ad56888901a5af6029b2.png\"><\/a><\/section>\n<section>\u9664\u4e86\u4e0a\u8ff0\u63d0\u5230\u7684\u589e\u52a0\u8ba1\u7b97\u8d44\u6e90\u3001\u63d0\u793a\u8bcd\u5de5\u7a0b\u65b9\u6cd5\u5916\uff0c\u8fd8\u6709\u53e6\u4e00\u79cd\u65b9\u6cd5\u5373\u6d89\u53ca\u5230\u6295\u7968\u548c\u641c\u7d22\u7b56\u7565\uff0c\u4f8b\u5982\u591a\u6570\u6295\u7968\u6216\u6ce2\u675f\u641c\u7d22\uff0c\u8fd9\u4e9b\u7b56\u7565\u901a\u8fc7\u9009\u62e9\u6700\u4f73\u8f93\u51fa\u6765\u4f18\u5316\u54cd\u5e94\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-6a4179652a2fd17e325d2b08c86a146e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-6a4179652a2fd17e325d2b08c86a146e.png\"><\/a><\/section>\n<section><strong>1.\u300cs1\uff1a\u7b80\u5355\u6d4b\u8bd5 &#8211; \u65f6\u95f4\u89c4\u6a21\u5316\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1as1: Simple test-time scaling<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2501.19393<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u7814\u7a76\u524d\u6cbf\uff0c2025 \u5e74 1 \u6708\u7531\u65af\u5766\u798f\u5927\u5b66\u7684\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u7684\u300aSimple Test-Time Scaling\u300b\u5f15\u5165\u4e86\u300cwait\u300dtoken \u673a\u5236\u3002\u5177\u4f53\u800c\u8a00\uff0c\u5f53\u6211\u4eec\u5e0c\u671b\u6a21\u578b\u5728\u67d0\u4e2a\u95ee\u9898\u4e0a\u82b1\u8d39\u66f4\u591a\u6d4b\u8bd5\u9636\u6bb5\u7684\u8ba1\u7b97\u8d44\u6e90\uff0c\u4f1a\u6291\u5236\u601d\u8003\u7ed3\u675f\u8bcd\u4e2d\u5206\u9694\u7b26\u7684\u751f\u6210\uff0c\u800c\u662f\u5728\u6a21\u578b\u5f53\u524d\u7684\u63a8\u7406\u8fc7\u7a0b\u4e2d\u8ffd\u52a0\u300cwait\u300dtoken\uff0c\u4ee5\u9f13\u52b1\u6a21\u578b\u8fdb\u884c\u66f4\u591a\u63a2\u7d22\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-303dd875bee677326b2ebc10be78fd90.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-303dd875bee677326b2ebc10be78fd90.png\"><\/a><\/section>\n<section>\u5e76\u4e14\uff0c\u8be5\u7814\u7a76\u4e2d\u7684\u9884\u7b97\u5f3a\u5236\u6280\u672f\uff08budget forcing\uff09\u7684\u4e3b\u8981\u673a\u5236\u5305\u542b\u4ee5\u4e0b\u4e24\u4e2a\u65b9\u9762\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u5f3a\u5236\u7ed3\u675f\u63a8\u7406\uff1a\u5f53\u6a21\u578b\u751f\u6210\u8d85\u8fc7\u9884\u8bbe\u7684\u300c\u601d\u8003\u300dtoken \u6570\u91cf\u65f6\uff0c\u6a21\u578b\u7684\u601d\u8003\u8fc7\u7a0b\u88ab\u5f3a\u5236\u7ed3\u675f\u3002<\/section>\n<\/li>\n<li>\n<section>\u5ef6\u957f\u63a8\u7406\u65f6\u95f4\uff1a\u5982\u679c\u5e0c\u671b\u6a21\u578b\u5728\u89e3\u51b3\u95ee\u9898\u65f6\u82b1\u8d39\u66f4\u591a\u7684\u8ba1\u7b97\u9884\u7b97\uff0c\u5219\u4e0d\u751f\u6210\u300c\u7ed3\u675f\u601d\u8003\u300dtoken\uff0c\u800c\u662f\u5728\u5f53\u524d\u7684\u63a8\u7406\u7ed3\u679c\u4e0a\u9644\u52a0\u591a\u4e2a\u300cwait\u300dtoken\uff0c\u8fd9\u6837\u6a21\u578b\u53ef\u4ee5\u7ee7\u7eed\u601d\u8003\u3002<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-5efa09ade85d71f5ff9dbc2ab13d3f15.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-5efa09ade85d71f5ff9dbc2ab13d3f15.png\"><\/a><\/section>\n<section>\u53ef\u4ee5\u786e\u4fdd\u6a21\u578b\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u4f7f\u7528\u7279\u5b9a\u8ba1\u7b97\u9884\u7b97\u7684\u63a7\u5236\u673a\u5236\u3002\u901a\u8fc7\u63a7\u5236\u6a21\u578b\u7684\u63a8\u7406\u65f6\u95f4\u6216\u6b65\u9aa4\u6570\uff0c\u652f\u6301\u66f4\u6709\u6548\u7684\u63a8\u7406\u548c\u9519\u8bef\u66f4\u6b63\u3002<\/section>\n<section><\/section>\n<section>Test-Time Scaling \u6280\u672f\u7684\u6838\u5fc3\u5728\u4e8e\u901a\u8fc7\u52a8\u6001\u8c03\u914d\u8ba1\u7b97\u8d44\u6e90\u8fdb\u884c\u300c\u4e34\u573a\u7279\u8bad\u300d\uff0c\u4f7f\u5f97\u6a21\u578b\u5728\u9047\u5230\u5177\u4f53\u95ee\u9898\u65f6\u80fd\u591f\u5c55\u73b0\u51fa\u60ca\u4eba\u7684\u63a8\u7406\u80fd\u529b\u3002\u8fd9\u79cd\u6280\u672f\u7a81\u7834\u89e3\u51b3\u4e86\u4f20\u7edf AI \u6a21\u578b\u8bad\u7ec3\u6210\u672c\u9ad8\u3001\u80fd\u6e90\u6d88\u8017\u5927\u3001\u521b\u65b0\u95e8\u69db\u9ad8\u7b49\u95ee\u9898\u3002<\/section>\n<section><\/section>\n<section>\u8be6\u89c1\u673a\u5668\u4e4b\u5fc3\u62a5\u9053\u300a<a data-itemshowtype=\"0\" data-linktype=\"2\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzA3MzI4MjgzMw==&amp;mid=2650953548&amp;idx=2&amp;sn=49e2bbe04e672580dbf9c1b2b7cd68df&amp;scene=21#wechat_redirect\" target=\"_blank\">\u8bad\u7ec3 1000 \u6837\u672c\u5c31\u80fd\u8d85\u8d8a o1\uff0c\u674e\u98de\u98de\u7b49\u4eba\u753b\u51fa AI \u6269\u5c55\u65b0\u66f2\u7ebf<\/a>\u300b\u3002<\/section>\n<section><\/section>\n<section><strong>2.\u300c\u6d4b\u8bd5\u504f\u597d\u4f18\u5316\u6846\u67b6\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aTest-Time Preference Optimization: On-the-Fly Alignment via Iterative Textual Feedback<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2501.12895<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u200cTest-Time Preference Optimization\uff08TPO\uff09\u200c\u662f\u4e00\u79cd\u65b0\u7684\u6846\u67b6\uff0c\u5176\u6838\u5fc3\u76ee\u6807\u5728\u4e8e\u4f7f LLM \u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u5feb\u901f\u5bf9\u9f50\u4eba\u7c7b\u504f\u597d\uff0c\u800c\u65e0\u9700\u91cd\u65b0\u8bad\u7ec3\u6a21\u578b\u53c2\u6570\u3002\u5728\u6bcf\u6b21\u8fed\u4ee3\u65f6\uff0c\u6a21\u578b\u4f1a\u6309\u4ee5\u4e0b\u6b65\u9aa4\u6267\u884c\u64cd\u4f5c\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u9488\u5bf9\u7ed9\u5b9a\u7684\u63d0\u793a\u751f\u6210\u591a\u4e2a\u54cd\u5e94<\/section>\n<\/li>\n<li>\n<section>\u8fd0\u7528\u5956\u52b1\u6a21\u578b\u5bf9\u54cd\u5e94\u8fdb\u884c\u8bc4\u5206\uff0c\u4ee5\u9009\u62e9\u5f97\u5206\u6700\u9ad8\u548c\u6700\u4f4e\u7684\u54cd\u5e94\u4f5c\u4e3a \u201c\u9009\u5b9a\u201d \u548c \u201c\u62d2\u7edd\u201d \u54cd\u5e94<\/section>\n<\/li>\n<li>\n<section>\u63d0\u793a\u6a21\u578b\u6bd4\u8f83\u548c\u6279\u8bc4 \u201c\u9009\u5b9a\u201d \u548c \u201c\u62d2\u7edd\u201d \u54cd\u5e94<\/section>\n<\/li>\n<li>\n<section>\u901a\u8fc7\u5c06\u6279\u8bc4\u8f6c\u6362\u4e3a\u6587\u672c\u5efa\u8bae\u6765\u66f4\u65b0\u539f\u59cb\u6a21\u578b\u54cd\u5e94\uff0c\u4ece\u800c\u4f18\u5316\u8f93\u51fa<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-53f95bc7fa3667976f394672259a993f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-53f95bc7fa3667976f394672259a993f.png\"><\/a><\/section>\n<section>\u901a\u8fc7\u6267\u884c\u4e0a\u8ff0\u8fed\u4ee3\u6b65\u9aa4\uff0c\u6a21\u578b\u80fd\u591f\u4e0d\u65ad\u6c72\u53d6\u6bcf\u6b21\u8fed\u4ee3\u7684\u7ecf\u9a8c\u6559\u8bad\uff0c\u9010\u6b65\u4fee\u6b63\u539f\u59cb\u54cd\u5e94\u4e2d\u4e0d\u7b26\u5408\u4eba\u7c7b\u504f\u597d\u7684\u90e8\u5206\uff0c\u4f7f LLM \u7684\u8f93\u51fa\u66f4\u52a0\u8d34\u5408\u4eba\u7c7b\u7684\u671f\u671b\u548c\u9700\u6c42\u3002<\/section>\n<section><\/section>\n<section><strong>3.\u300c\u601d\u7eea\u4e07\u5343\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aThoughts Are All Over the Place:On the Under thinking of o1-Like LLMs<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2501.18585<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u7814\u7a76\u56e2\u961f\u53d1\u73b0\u4e86\u4e00\u79cd\u79f0\u4e3a\u300cunderthinking\uff08\u601d\u8003\u4e0d\u8db3\uff09\u300d\u7684\u73b0\u8c61\uff0c\u5373\u63a8\u7406\u6a21\u578b\u9891\u7e41\u5728\u63a8\u7406\u8def\u5f84\u4e4b\u95f4\u5207\u6362\uff0c\u800c\u4e0d\u662f\u5b8c\u5168\u4e13\u6ce8\u4e8e\u63a2\u7d22\u6709\u5e0c\u671b\u7684\u8def\u5f84\uff0c\u8fd9\u964d\u4f4e\u4e86\u89e3\u51b3\u95ee\u9898\u7684\u51c6\u786e\u6027\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-7491e95b332afecbebcd4c4114119746.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-7491e95b332afecbebcd4c4114119746.png\"><\/a><\/section>\n<section>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u300c\u601d\u8003\u4e0d\u8db3\u300d\u95ee\u9898\uff0c\u7814\u7a76\u4eba\u5458\u5f15\u5165\u4e86\u4e00\u79cd\u79f0\u4e3a\u601d\u7ef4\u8f6c\u6362\u60e9\u7f5a (TIP) \u7684\u7b56\u7565\uff0c\u8be5\u7b56\u7565\u4fee\u6539\u4e86\u601d\u7ef4\u8f6c\u6362 token \u7684\u903b\u8f91\uff0c\u4ee5\u963b\u6b62\u8fc7\u65e9\u7684\u63a8\u7406\u8def\u5f84\u8f6c\u6362\u3002\u4e14\u5b9e\u9a8c\u7814\u7a76\u8bc1\u660e\u8be5\u65b9\u6cd5\u4e0d\u9700\u8981\u6a21\u578b\u5fae\u8c03\uff0c\u5e76\u4e14\u5728\u5b9e\u9a8c\u4e2d\u63d0\u9ad8\u4e86\u5728\u591a\u4e2a\u9ad8\u96be\u5ea6\u6d4b\u8bd5\u96c6\u4e0a\u7684\u51c6\u786e\u6027\u3002<\/section>\n<section><\/section>\n<section><strong>4.\u300c\u7528\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6362\u53d6\u5bf9\u6297\u9c81\u68d2\u6027\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aTrading Inference-Time Compute for Adversarial Robustness<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2501.18841<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u8be5\u7814\u7a76\u663e\u793a\u5728\u4f17\u591a\u60c5\u51b5\u4e0b\uff0c\u5ef6\u957f\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u80fd\u591f\u63d0\u9ad8\u63a8\u7406 LLM \u7684\u5bf9\u6297\u9c81\u68d2\u6027\uff0c\u8fdb\u800c\u964d\u4f4e\u6210\u529f\u653b\u51fb\u7684\u6982\u7387\u3002\u4e0e\u4f20\u7edf\u7684\u5bf9\u6297\u8bad\u7ec3\u4e0d\u540c\uff0c\u8fd9\u79cd\u65b9\u6cd5\u4e0d\u9700\u8981\u4efb\u4f55\u7279\u6b8a\u8bad\u7ec3\uff0c\u4e5f\u4e0d\u9700\u8981\u4e8b\u5148\u4e86\u89e3\u7279\u5b9a\u7684\u653b\u51fb\u7c7b\u578b\uff0c\u5177\u6709\u663e\u8457\u7684\u4fbf\u6377\u6027\u548c\u901a\u7528\u6027\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-e435b31fe17f0f1271c6623429dc4da0.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-e435b31fe17f0f1271c6623429dc4da0.png\"><\/a><\/section>\n<section>\u7136\u800c\uff0c\u8fd9\u79cd\u65b9\u6cd5\u4e5f\u5e76\u975e\u5728\u6240\u6709\u573a\u666f\u4e2d\u90fd\u80fd\u53d1\u6325\u7406\u60f3\u6548\u679c\u3002\u4f8b\u5982\uff0c\u5728\u6d89\u53ca\u7b56\u7565\u6a21\u7cca\u6027\u6216\u8005\u53ef\u88ab\u5229\u7528\u6f0f\u6d1e\u7684\u73af\u5883\u91cc\uff0c\u8be5\u65b9\u6cd5\u6539\u8fdb\u662f\u6709\u9650\u7684\u3002\u6b64\u5916\uff0c\u63a8\u7406\u6539\u8fdb\u7684\u9c81\u68d2\u6027\u589e\u52a0\u53ef\u80fd\u4f1a\u88ab\u300cThink Less\u300d\u548c\u300cNerd Sniping\u300d\u7b49\u65b0\u653b\u51fb\u7b56\u7565\u6240\u524a\u5f31\u3002<\/section>\n<section><\/section>\n<section><strong>5.\u300c\u8054\u60f3\u601d\u7ef4\u94fe\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aCoAT: Chain-of-Associated-Thoughts Framework for Enhancing Large Language Models Reasoning<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2502.02390<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>Chain-of-Associated-Thoughts\uff08CoAT\uff09\u6846\u67b6\u200c\u662f\u4e00\u79cd\u7528\u4e8e\u589e\u5f3a\u5927\u578b\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u80fd\u529b\u7684\u65b0\u65b9\u6cd5\uff0c\u8be5\u6846\u67b6\u5de7\u5999\u5730\u5c06\u8499\u7279\u5361\u6d1b\u6811\u641c\u7d22\uff08MCTS\uff09\u7b97\u6cd5\u4e0e\u5173\u8054\u8bb0\u5fc6\u673a\u5236\u76f8\u7ed3\u5408\uff0c\u901a\u8fc7\u7ed3\u6784\u5316\u63a2\u7d22\u4e0e\u81ea\u9002\u5e94\u5b66\u4e60\uff0c\u6709\u6548\u6269\u5c55\u4e86\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u641c\u7d22\u7a7a\u95f4\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-8a8d23438ede7def9f5f885ce842d3ba.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-8a8d23438ede7def9f5f885ce842d3ba.png\"><\/a><\/section>\n<section>CoAT \u6846\u67b6\u7684\u6838\u5fc3\u5de5\u4f5c\u539f\u7406\u5982\u4e0b\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8499\u7279\u5361\u6d1b\u6811\u641c\u7d22\uff08MCTS\uff09\u200c\uff1aMCTS \u7b97\u6cd5\u7528\u4e8e\u7ed3\u6784\u5316\u63a2\u7d22\uff0c\u5e2e\u52a9\u6a21\u578b\u5728\u51b3\u7b56\u8fc7\u7a0b\u4e2d\u8fdb\u884c\u591a\u6b65\u9aa4\u7684\u63a8\u7406\u3002<\/section>\n<\/li>\n<li>\n<section>\u5173\u8054\u8bb0\u5fc6\u673a\u5236\u200c\uff1a\u4e00\u79cd\u7528\u4e8e\u96c6\u6210\u65b0\u7684\u5173\u952e\u4fe1\u606f\u7684\u52a8\u6001\u673a\u5236\uff0c\u80fd\u591f\u6839\u636e\u4e0a\u4e0b\u6587\u548c\u63a8\u7406\u8fc7\u7a0b\u4e2d\u7684\u9700\u8981\uff0c\u4e0d\u65ad\u66f4\u65b0\u548c\u8865\u5145\u76f8\u5173\u4fe1\u606f\uff0c\u4ece\u800c\u589e\u5f3a\u6a21\u578b\u7684\u81ea\u9002\u5e94\u5b66\u4e60\u80fd\u529b\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u901a\u8fc7\u5c06 MCTS \u7684\u7ed3\u6784\u5316\u63a2\u7d22\u4e0e\u5173\u8054\u8bb0\u5fc6\u7684\u52a8\u6001\u5b66\u4e60\u7d27\u5bc6\u7ed3\u5408\uff0cCoAT \u663e\u8457\u6269\u5c55\u4e86\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u641c\u7d22\u7a7a\u95f4\uff0c\u63d0\u5347\u4e86\u5176\u5728\u590d\u6742\u63a8\u7406\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u200c\u3002<\/section>\n<section><\/section>\n<section><strong>6.\u300c\u81ea\u6211\u56de\u6eaf\u4ee5\u4fc3\u8fdb\u63a8\u7406\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aStep Back to Leap Forward: Self-Backtracking for Boosting Reasoning of Language Models<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740:\u00a0https:\/\/www.arxiv.org\/pdf\/2502.04404<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u6765\u81ea\u5357\u4eac\u5927\u5b66\u7684\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u4e86\u4e00\u79cd\u901a\u8fc7\u81ea\u6211\u56de\u6eaf\uff08self-back tracking\uff09\u6765\u63d0\u5347\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u80fd\u529b\u7684\u65b9\u6cd5\uff0c\u8fd9\u79cd\u65b9\u6cd5\u7684\u6838\u5fc3\u601d\u60f3\u662f\u8ba9\u8bed\u8a00\u6a21\u578b\u5728\u89e3\u51b3\u5177\u4f53\u95ee\u9898\u65f6\uff0c\u5148\u4ece\u5b8f\u89c2\u89d2\u5ea6\u8fdb\u884c\u62bd\u8c61\uff0c\u7136\u540e\u518d\u56de\u5230\u7ec6\u8282\u5c42\u9762\u8fdb\u884c\u5177\u4f53\u63a8\u7406\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-49157a97c611c954f346a3b265b07ba0.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-49157a97c611c954f346a3b265b07ba0.png\"><\/a><\/section>\n<section>\u8be5\u65b9\u6cd5\u5305\u542b\u4e24\u4e2a\u4e3b\u8981\u6b65\u9aa4\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u62bd\u8c61\u9636\u6bb5\u200c\uff1a\u6a21\u578b\u9996\u5148\u88ab\u8981\u6c42\u56de\u7b54\u4e00\u4e2a\u66f4\u4e00\u822c\u7684\u95ee\u9898\uff0c\u8fd9\u4e2a\u95ee\u9898\u662f\u5bf9\u539f\u59cb\u5177\u4f53\u95ee\u9898\u7684\u62bd\u8c61\u3002<\/section>\n<\/li>\n<li>\n<section>\u63a8\u7406\u9636\u6bb5\u200c\uff1a\u57fa\u4e8e\u62bd\u8c61\u9636\u6bb5\u5f97\u5230\u7684\u4e00\u822c\u7ed3\u8bba\uff0c\u6a21\u578b\u518d\u56de\u5230\u5177\u4f53\u95ee\u9898\u4e0a\u8fdb\u884c\u63a8\u7406\uff0c\u4ece\u800c\u5f97\u51fa\u6700\u7ec8\u7684\u7b54\u6848\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5b9e\u9a8c\u7ed3\u679c\u663e\u793a\uff0c\u5728 STEM\uff08\u79d1\u5b66\u3001\u6280\u672f\u3001\u5de5\u7a0b\u548c\u6570\u5b66\uff09\u95ee\u9898\u3001\u77e5\u8bc6\u95ee\u7b54\u4ee5\u53ca\u591a\u8df3\u63a8\u7406\u4efb\u52a1\u7b49\u573a\u666f\u4e2d\uff0c\u8fd9\u79cd\u81ea\u6211\u56de\u6eaf\u65b9\u6cd5\u7684\u6548\u679c\u663e\u8457\u4f18\u4e8e\u4f20\u7edf\u7684\u601d\u7ef4\u94fe\uff08CoT\uff09\u65b9\u6cd5\u3002<\/section>\n<section><\/section>\n<section><strong>7.\u300c\u57fa\u4e8e\u6df1\u5ea6\u5faa\u73af\u9690\u7a7a\u95f4\u63a8\u7406\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aScaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2502.05171<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u63d0\u5347\u6a21\u578b\u63a8\u7406\u80fd\u529b\u7684\u63a2\u7d22\u4e2d\uff0c\u7814\u7a76\u4eba\u5458\u5e76\u672a\u9075\u5faa\u751f\u6210\u66f4\u591a token \u4ee5\u6539\u8fdb\u63a8\u7406\u7684\u5e38\u89c4\u601d\u8def\uff0c\u800c\u662f\u63d0\u51fa\u4e86\u4e00\u4e2a\u901a\u8fc7\u5728\u6f5c\u5728\u7a7a\u95f4\u4e2d\u8fed\u4ee3\u5faa\u73af\u6df1\u5ea6\u5757\u6765\u6269\u5c55\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u7684\u6a21\u578b\u3002\u8fd9\u4e2a\u6df1\u5ea6\u5757\u7684\u529f\u80fd\u7c7b\u4f3c\u4e8e RNN \u4e2d\u7684\u9690\u85cf\u72b6\u6001\uff0c\u5b83\u5141\u8bb8\u6a21\u578b\u6539\u8fdb\u5176\u63a8\u7406\u800c\u4e0d\u9700\u8981\u66f4\u957f\u7684 token \u8f93\u51fa\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-8215a48338f6ceb91445f97c0c4c44c4.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-8215a48338f6ceb91445f97c0c4c44c4.png\"><\/a><\/section>\n<section>\u7814\u7a76\u56e2\u961f\u8fd8\u8fdb\u4e00\u6b65\u5c06\u4e00\u4e2a\u6982\u5ff5\u9a8c\u8bc1\u6a21\u578b\u8fdb\u884c\u4e86\u6269\u5c55\uff0c\u4f7f\u5176\u5177\u5907 35 \u4ebf\u53c2\u6570\u548c 8000 \u4ebf token\u3002\u5728\u63a8\u7406\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u8be5\u6269\u5c55\u540e\u7684\u6a21\u578b\u5c55\u73b0\u51fa\u4e86\u663e\u8457\u7684\u6027\u80fd\u63d0\u5347\uff0c\u90e8\u5206\u60c5\u51b5\u4e0b\u5176\u8ba1\u7b97\u8d1f\u8f7d\u6548\u679c\u751a\u81f3\u53ef\u4e0e\u62e5\u6709 500 \u4ebf\u53c2\u6570\u7684\u6a21\u578b\u76f8\u5ab2\u7f8e\u3002<\/section>\n<section><\/section>\n<section>\u4e0d\u8fc7\uff0c\u4e00\u4e2a\u5173\u952e\u77ed\u677f\u5728\u4e8e\u8be5\u6a21\u578b\u7f3a\u5c11\u6e05\u6670\u660e\u786e\u7684\u63a8\u7406\u6b65\u9aa4\u3002\u5bf9\u4e8e\u7528\u6237\u548c\u7814\u7a76\u4eba\u5458\u800c\u8a00\uff0c\u6e05\u6670\u7684\u63a8\u7406\u6b65\u9aa4\u80fd\u591f\u6781\u5927\u5730\u63d0\u5347\u6a21\u578b\u7684\u4eba\u7c7b\u53ef\u89e3\u91ca\u6027\u3002\u800c\u8fd9\u6070\u6070\u662f\u601d\u7ef4\u94fe\u65b9\u6cd5\u7684\u4e00\u9879\u4e3b\u8981\u4f18\u52bf\u6240\u5728\u3002<\/section>\n<section><\/section>\n<section><strong>8.\u300c1B LLM \u53ef\u4ee5\u8d85\u8d8a 405B LLM \u5417\uff1f\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aCan 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2502.06703<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u63d0\u5347\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u63a8\u7406\u80fd\u529b\u7684\u76f8\u5173\u7814\u7a76\u4e2d\uff0c\u8bb8\u591a\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u6280\u672f\u4f9d\u8d56\u4e8e\u91c7\u6837\uff0c\u8fd9\u9700\u8981\u8fc7\u7a0b\u5956\u52b1\u6a21\u578b (PRM) \u6765\u9009\u62e9\u6700\u4f73\u89e3\u51b3\u65b9\u6848\u3002\u8fd9\u7bc7\u7814\u7a76\u6210\u679c\u5c31\u5bf9\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u4e0e PRM\u3001\u95ee\u9898\u96be\u5ea6\u4e4b\u95f4\u7684\u76f8\u4e92\u4f5c\u7528\u5c55\u5f00\u4e86\u7cfb\u7edf\u5206\u6790\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-f08e299c964aed72eef07b4a208b709e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-f08e299c964aed72eef07b4a208b709e.png\"><\/a><\/section>\n<section>\u7814\u7a76\u4eba\u5458\u5f00\u53d1\u4e86\u4e00\u79cd\u8ba1\u7b97\u4f18\u5316\u6269\u5c55\u7b56\u7565\uff0c\u8be5\u7b56\u7565\u53ef\u9002\u5e94 PRM\u3001\u7b56\u7565\u6a21\u578b\u548c\u4efb\u52a1\u590d\u6742\u6027\u7684\u9009\u62e9\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u901a\u8fc7\u6b63\u786e\u7684\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u65b9\u6cd5\uff0c1B \u53c2\u6570\u6a21\u578b\u53ef\u4ee5\u80dc\u8fc7\u7f3a\u4e4f\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u7684 405B Llama 3 \u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section>\u540c\u6837\u5730\uff0c\u7814\u7a76\u8fd8\u5c55\u793a\u4e86\u914d\u5907\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u6280\u672f\u7684 7B \u6a21\u578b\u4e0d\u4ec5\u80fd\u591f\u8d85\u8d8a DeepSeek-R1\uff0c\u8fd8\u80fd\u4fdd\u6301\u66f4\u9ad8\u7684\u6548\u7387\u3002\u8fd9\u4e9b\u53d1\u73b0\u51f8\u663e\u4e86\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u5bf9\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u6027\u80fd\u63d0\u5347\u7684\u663e\u8457\u4f5c\u7528\uff0c\u5176\u4e2d\u5177\u6709\u6b63\u786e\u63a8\u7406\u8ba1\u7b97\u9884\u7b97\u7684\u5c0f\u578b LLM \u53ef\u4ee5\u80dc\u8fc7\u66f4\u5927\u7684\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section><strong>9.\u300c\u91cd\u65b0\u601d\u8003\u8ba1\u7b97\u6700\u4f18\u6d4b\u8bd5\u65f6\u95f4\u6269\u5c55\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aLearning to Reason from Feedback at Test-Time<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740:\u00a0https:\/\/arxiv.org\/pdf\/2502.15771<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u533a\u522b\u4e8e\u524d\u51e0\u79cd\u6210\u679c\u7684\u601d\u8def\uff0c\u8be5\u7814\u7a76\u7684\u6838\u5fc3\u5728\u4e8e\u901a\u8fc7\u63a8\u7406\u65f6\u52a8\u6001\u8c03\u6574\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u6743\u91cd\u53c2\u6570\uff0c\u4f7f\u5176\u80fd\u591f\u4ece\u9519\u8bef\u4e2d\u5b66\u4e60\uff0c\u800c\u65e0\u9700\u5c06\u5931\u8d25\u7684\u5c1d\u8bd5\u5b58\u50a8\u5728\u63d0\u793a\uff08prompt\uff09\u4e2d\uff0c\u4ece\u800c\u907f\u514d\u4e86\u9ad8\u6602\u7684\u6210\u672c\u3002\u4e0e\u4f20\u7edf\u7684\u901a\u8fc7\u5c06\u5148\u524d\u7684\u5c1d\u8bd5\u6dfb\u52a0\u5230\u4e0a\u4e0b\u6587\u4e2d\u8fdb\u884c\u987a\u5e8f\u4fee\u6b63\u6216\u76f2\u76ee\u751f\u6210\u65b0\u7b54\u6848\u7684\u5e76\u884c\u91c7\u6837\u4e0d\u540c\uff0c\u8be5\u65b9\u6cd5\u5728\u63a8\u7406\u65f6\u76f4\u63a5\u66f4\u65b0\u6a21\u578b\u7684\u6743\u91cd\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-4578375b77aee80a1565112445687993.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-4578375b77aee80a1565112445687993.png\"><\/a><\/section>\n<section>\u4e3a\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\uff0c\u7814\u7a76\u8005\u63d0\u51fa\u4e86 OpTune\uff0c\u8fd9\u662f\u4e00\u79cd\u5c0f\u578b\u3001\u53ef\u8bad\u7ec3\u7684\u4f18\u5316\u5668\uff0c\u80fd\u591f\u6839\u636e\u6a21\u578b\u5728\u5148\u524d\u5c1d\u8bd5\u4e2d\u72af\u7684\u9519\u8bef\u66f4\u65b0\u6a21\u578b\u7684\u6743\u91cd\u3002\u8fd9\u610f\u5473\u7740\u6a21\u578b\u4f1a\u8bb0\u4f4f\u5b83\u505a\u9519\u4e86\u4ec0\u4e48\uff0c\u800c\u65e0\u9700\u5728\u63d0\u793a \/ \u4e0a\u4e0b\u6587\u4e2d\u4fdd\u7559\u9519\u8bef\u7684\u7b54\u6848\u3002\u8fd9\u79cd\u65b9\u6cd5\u4e0d\u4ec5\u63d0\u9ad8\u4e86\u6a21\u578b\u7684\u81ea\u6211\u4fee\u6b63\u80fd\u529b\uff0c\u8fd8\u663e\u8457\u964d\u4f4e\u4e86\u63a8\u7406\u8fc7\u7a0b\u4e2d\u7684\u8ba1\u7b97\u548c\u5b58\u50a8\u5f00\u9500\u3002<\/section>\n<section><\/section>\n<section><strong>10.\u300c\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u5728\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u63a8\u7406\u548c\u89c4\u5212\u4e2d\u7684\u5e94\u7528\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aInference-Time Computations for LLM Reasoning and Planning: A Benchmark and Insights<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/www.arxiv.org\/pdf\/2502.12521<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u9488\u5bf9\u63a8\u7406\u548c\u89c4\u5212\u4efb\u52a1\u7684\u5404\u79cd\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u6280\u672f\u8fdb\u884c\u4e86\u57fa\u51c6\u6d4b\u8bd5\uff0c\u8be5\u6587\u7ae0\u91cd\u70b9\u5206\u6790\u4e86\u5b83\u4eec\u5728\u8ba1\u7b97\u6210\u672c\u548c\u6027\u80fd\u4e4b\u95f4\u7684\u6743\u8861\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-1fc5ccf580da3aac6cbe0599759d582d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-1fc5ccf580da3aac6cbe0599759d582d.png\"><\/a><\/section>\n<section>\u5728\u5177\u4f53\u7684\u5b9e\u9a8c\u91cc\uff0c\u7814\u7a76\u4eba\u5458\u8bc4\u4f30\u4e86\u591a\u79cd\u6280\u672f\uff0c\u4f8b\u5982\u601d\u7ef4\u94fe\u3001\u601d\u7ef4\u6811\u548c\u63a8\u7406\u89c4\u5212\uff0c\u6d89\u53ca\u7b97\u672f\u3001\u903b\u8f91\u3001\u5e38\u8bc6\u3001\u7b97\u6cd5\u63a8\u7406\u548c\u89c4\u5212\u7b49 11 \u4e2a\u4efb\u52a1\uff0c\u7ec6\u81f4\u5730\u8986\u76d6\u4e86\u591a\u4e2a\u9886\u57df\u7684\u63a8\u7406\u548c\u89c4\u5212\u573a\u666f\u3002<\/section>\n<section><\/section>\n<section>\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u867d\u7136\u6269\u5c55\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u53ef\u4ee5\u6539\u5584\u63a8\u7406\uff0c\u4f46\u6ca1\u6709\u4e00\u79cd\u6280\u672f\u5728\u6240\u6709\u4efb\u52a1\u4e2d\u59cb\u7ec8\u4f18\u4e8e\u5176\u4ed6\u6280\u672f\u3002\u8fd9\u4e5f\u4fa7\u9762\u8bf4\u660e\u4e86\u76ee\u524d\u5728\u4e0d\u540c\u7684\u63a8\u7406\u548c\u89c4\u5212\u4efb\u52a1\u4e2d\uff0c\u8fd8\u9700\u8981\u6839\u636e\u5177\u4f53\u60c5\u51b5\u7075\u6d3b\u9009\u62e9\u5408\u9002\u7684\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u6280\u672f\uff0c\u4ee5\u5728\u8ba1\u7b97\u6210\u672c\u548c\u6027\u80fd\u4e4b\u95f4\u627e\u5230\u6700\u4f73\u5e73\u8861\u3002<\/section>\n<section><\/section>\n<section><strong>11.\u300c\u5185\u90e8\u601d\u7ef4 Transformer\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aInner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740:\u00a0https:\/\/arxiv.org\/pdf\/2502.13842<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u9762\u5bf9\u9700\u8981\u5904\u7406\u590d\u6742\u63a8\u7406\u7684\u5173\u952e token \u65f6\uff0c\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u5728\u53c2\u6570\u7ea6\u675f\u4e0b\u4f1a\u51fa\u73b0\u56fa\u6709\u7684\u6027\u80fd\u74f6\u9888\u3002\u9488\u5bf9\u8fd9\u4e2a\u95ee\u9898\uff0c\u5185\u90e8\u601d\u7ef4 Transformer (The Inner Thinking Transformer\uff0cITT) \u7684\u89e3\u51b3\u601d\u8def\u662f\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u52a8\u6001\u5206\u914d\u66f4\u591a\u8ba1\u7b97\u8d44\u6e90\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-0d94210d5b02c0c56d6584a2fe390adc.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-0d94210d5b02c0c56d6584a2fe390adc.png\"><\/a><\/section>\n<section>\u5177\u4f53\u800c\u8a00\uff0cITT \u901a\u8fc7\u81ea\u9002\u5e94\u8def\u7531 token\uff08Adaptive Token Routing\uff09\u52a8\u6001\u5206\u914d\u8ba1\u7b97\u8d44\u6e90\uff0c\u901a\u8fc7\u6b8b\u5dee\u601d\u8003\u8fde\u63a5\u8fed\u4ee3\u5730\u4f18\u5316\u8868\u5f81\uff0c\u8ba9\u8fd9\u4e9b\u56f0\u96be\u7684 token \u591a\u6b21\u901a\u8fc7\u540c\u4e00\u5c42\u8fdb\u884c\u989d\u5916\u5904\u7406\uff0c\u4ece\u800c\u4f7f\u5f97 ITT \u80fd\u591f\u5728\u4e0d\u589e\u52a0\u53c2\u6570\u7684\u60c5\u51b5\u4e0b\u5bf9\u5173\u952e\u7684 token \u8fdb\u884c\u66f4\u6df1\u5165\u7684\u5904\u7406\u3002<\/section>\n<section><\/section>\n<section><strong>12.\u300c\u7a81\u7834\u6027\u6846\u67b6 S*\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aScaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2408.03314<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u4f20\u7edf\u7684\u4ee3\u7801\u751f\u6210\u6a21\u578b\u5f80\u5f80\u4f9d\u8d56\u4e8e\u66f4\u5927\u7684\u53c2\u6570\u91cf\u6765\u63d0\u5347\u6027\u80fd\u3002\u4f46\u662f\uff0c\u65b0\u63d0\u51fa\u7684 S * \u6846\u67b6\u901a\u8fc7\u4e00\u79cd\u5de7\u5999\u7684\u65b9\u5f0f\uff0c\u8ba9\u5c0f\u6a21\u578b\u4e5f\u80fd\u53d1\u6325\u51fa\u5927\u6a21\u578b\u7684\u5b9e\u529b\u3002\u8fd9\u4e2a\u6846\u67b6\u4e3b\u8981\u5305\u542b\u4e24\u4e2a\u5173\u952e\u521b\u65b0\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u6df7\u5408\u6269\u5c55\u7b56\u7565\uff1a\u7ed3\u5408\u4e86\u5e76\u884c\u91c7\u6837\u548c\u987a\u5e8f\u8c03\u8bd5\uff0c\u5927\u5e45\u63d0\u5347\u4e86\u4ee3\u7801\u751f\u6210\u7684\u8986\u76d6\u7387<\/section>\n<\/li>\n<li>\n<section>\u81ea\u9002\u5e94\u9009\u62e9\u673a\u5236\uff1a\u901a\u8fc7\u667a\u80fd\u751f\u6210\u6d4b\u8bd5\u7528\u4f8b\u6765\u533a\u5206\u4e0d\u540c\u7684\u4ee3\u7801\u65b9\u6848\uff0c\u5e76\u901a\u8fc7\u5b9e\u9645\u6267\u884c\u7ed3\u679c\u6765\u9009\u62e9\u6700\u4f73\u7b54\u6848<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-4c441289636a7164dd92bf8d6764ebc5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-4c441289636a7164dd92bf8d6764ebc5.png\"><\/a><\/section>\n<section>S* \u662f\u4e00\u4e2a\u4e24\u9636\u6bb5\u6df7\u5408\u6d4b\u8bd5\u65f6\u7f29\u653e\u6846\u67b6\uff0c\u7531\u751f\u6210\u548c\u9009\u62e9\u9636\u6bb5\u7ec4\u6210\uff0c\u5982\u4e0b\u56fe\u6240\u793a\u3002\u5b83\u901a\u8fc7\u8fed\u4ee3\u8c03\u8bd5\u6269\u5c55\u5e76\u884c\u91c7\u6837\u4e0e\u987a\u5e8f\u91c7\u6837\u4ee5\u63d0\u9ad8\u8986\u76d6\u7387\uff0c\u5e76\u5728\u9009\u62e9\u8fc7\u7a0b\u4e2d\u91c7\u7528\u81ea\u9002\u5e94\u8f93\u5165\u5408\u6210\u6765\u589e\u5f3a\u9009\u62e9\u51c6\u786e\u6027\uff0c\u5728\u6574\u4e2a\u8fc7\u7a0b\u4e2d\u5229\u7528\u6267\u884c\u7ed3\u679c\u3002\u4e0d\u540c\u9636\u6bb5\u7684\u6548\u679c\u793a\u4f8b\u53ef\u4ee5\u5728\u4e0b\u56fe\u4e2d\u627e\u5230\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-a92e0cefbea5df424e3a54ecdeb0d57d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-a92e0cefbea5df424e3a54ecdeb0d57d.png\"><\/a><\/section>\n<ul>\n<li>\n<section>\u9636\u6bb5 1 \uff1a\u5728\u751f\u6210\u9636\u6bb5\uff0c\u6a21\u578b S* \u751f\u6210\u591a\u4e2a\u4ee3\u7801\u89e3\u51b3\u65b9\u6848\uff0c\u5e76\u4f7f\u7528\u95ee\u9898\u63d0\u793a\u8bcd\u4e2d\u63d0\u4f9b\u7684\u6267\u884c\u7ed3\u679c\u548c\u6d4b\u8bd5\u7528\u4f8b\u8fed\u4ee3\u7ec6\u5316\u5b83\u4eec\u3002(1) \u6a21\u578b\u751f\u6210\u591a\u4e2a\u5019\u9009\u89e3\u51b3\u65b9\u6848\u3002(2) \u6bcf\u4e2a\u89e3\u51b3\u65b9\u6848\u90fd\u5728\u516c\u5171\u6d4b\u8bd5\u7528\u4f8b\uff08\u9884\u5b9a\u4e49\u7684\u8f93\u5165\u8f93\u51fa\u5bf9\uff09\u4e0a\u6267\u884c\u3002(3) \u5982\u679c\u89e3\u51b3\u65b9\u6848\u5931\u8d25\uff08\u8f93\u51fa\u4e0d\u6b63\u786e\u6216\u5d29\u6e83\uff09\uff0c\u6a21\u578b\u4f1a\u5206\u6790\u6267\u884c\u7ed3\u679c\uff08\u9519\u8bef\u3001\u8f93\u51fa\uff09\u5e76\u4fee\u6539\u4ee3\u7801\u4ee5\u6539\u8fdb\u5b83\u3002(4) \u6b64\u6539\u8fdb\u8fc7\u7a0b\u4e0d\u65ad\u8fed\u4ee3\uff0c\u76f4\u5230\u6a21\u578b\u627e\u5230\u901a\u8fc7\u6d4b\u8bd5\u7528\u4f8b\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/section>\n<\/li>\n<li>\n<section>\u9636\u6bb5 2\uff1a\u5728\u9009\u62e9\u9636\u6bb5\uff0cS* \u5728\u751f\u6210 N \u4e2a\u5019\u9009\u89e3\u51b3\u65b9\u6848\u540e\uff0c\u4e0b\u4e00\u6b65\u662f\u8bc6\u522b\u6700\u4f73\u89e3\u51b3\u65b9\u6848\u3002(1) \u6a21\u578b\u6bd4\u8f83\u4e24\u4e2a\u90fd\u901a\u8fc7\u516c\u5f00\u6d4b\u8bd5\u7684\u89e3\u51b3\u65b9\u6848\u3002(2) \u751f\u6210\u7684\u6d4b\u8bd5\u7528\u4f8b\uff0c\u5b83\u4f7f\u7528\u5408\u6210\u7684\u6d4b\u8bd5\u7528\u4f8b\u6765\u6307\u5bfc\u9009\u62e9\u3002(3) \u5c06\u65b0\u7684\u6d4b\u8bd5\u8f93\u5165\u5e76\u5728\u5176\u4e0a\u8fd0\u884c\u4e24\u4e2a\u89e3\u51b3\u65b9\u6848\u3002(4) \u5982\u679c\u4e00\u4e2a\u89e3\u51b3\u65b9\u6848\u4ea7\u751f\u6b63\u786e\u7684\u8f93\u51fa\u800c\u53e6\u4e00\u4e2a\u5931\u8d25\uff0c\u5219\u6a21\u578b\u4f1a\u9009\u62e9\u66f4\u597d\u7684\u89e3\u51b3\u65b9\u6848\u3002(5) \u5982\u679c\u4e24\u79cd\u89e3\u51b3\u65b9\u6848\u7684\u8868\u73b0\u76f8\u540c\uff0c\u6a21\u578b\u5c06\u968f\u673a\u9009\u62e9\u5176\u4e2d\u4e00\u4e2a\u3002<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-e871999036eda7fd959bbd89b094af59.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-e871999036eda7fd959bbd89b094af59.png\"><\/a><\/section>\n<section>S* \u6846\u67b6\u7684\u6548\u679c\u975e\u5e38\u4e0d\u9519\uff0c\u5176\u5728\u6a21\u578b\u6027\u80fd\u63d0\u5347\u65b9\u9762\u5c55\u73b0\u4e86\u663e\u8457\u4f18\u52bf\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u5c0f\u6a21\u578b\u9006\u88ad\uff1a\u5728 S* \u6846\u67b6\u52a0\u6301\u4e0b\uff0cQwen2.5-7B \u6a21\u578b\u7684\u6027\u80fd\u8868\u73b0\u8d85\u8d8a\u5176\u539f\u751f Qwen2.5-32B \u7248\u672c\uff0c\u5b9e\u73b0\u4e86 10.7% \u7684\u6027\u80fd\u8dc3\u5347\uff0c\u5145\u5206\u5c55\u73b0\u4e86\u5c0f\u6a21\u578b\u5728\u4f18\u5316\u6846\u67b6\u4e0b\u7684\u5de8\u5927\u6f5c\u529b\u3002<\/section>\n<\/li>\n<li>\n<section>\u6027\u80fd\u7a81\u7834\uff1aGPT-4o-mini \u6a21\u578b\u5728\u96c6\u6210 S* \u6846\u67b6\u540e\uff0c\u6027\u80fd\u8868\u73b0\u8d85\u8d8a\u4e86 o1-preview \u7248\u672c\uff0c\u63d0\u5347\u5e45\u5ea6\u8fbe\u5230 3.7%\uff0c\u6210\u529f\u7a81\u7834\u4e86\u539f\u6709\u6027\u80fd\u5929\u82b1\u677f\u3002<\/section>\n<\/li>\n<li>\n<section>\u9876\u5c16\u8ffd\u5e73\uff1a\u901a\u8fc7 S \u6846\u67b6\u7684\u4f18\u5316\uff0cDeepSeek-R1-Distill-Qwen-32B \u6a21\u578b\u7684\u6027\u80fd\u8fbe\u5230 85.7%\uff0c\u4e0e\u5f53\u524d\u4e1a\u754c\u9886\u5148\u7684 o1-high \u6a21\u578b\uff0888.5%\uff09\u4ec5\u76f8\u5dee 2.8 \u4e2a\u767e\u5206\u70b9\uff0c\u5c55\u73b0\u51fa\u6781\u5f3a\u7684\u7ade\u4e89\u529b\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u8fd9\u4e9b\u6210\u679c\u5145\u5206\u8bc1\u660e\u4e86 S * \u6846\u67b6\u5728\u6a21\u578b\u6027\u80fd\u4f18\u5316\u65b9\u9762\u7684\u5353\u8d8a\u80fd\u529b\uff0c\u4e3a AI \u6a21\u578b\u7684\u53d1\u5c55\u5f00\u8f9f\u4e86\u65b0\u7684\u53ef\u80fd\u6027\u3002<\/section>\n<section><\/section>\n<section><strong>13.\u300c\u8349\u7a3f\u94fe Chain of Draft\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aChain of Draft: Thinking Faster by Writing Less<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740:\u00a0https:\/\/arxiv.org\/pdf\/2502.18600<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u7814\u7a76\u4eba\u5458\u89c2\u5bdf\u5230\uff0c\u601d\u7ef4\u94fe (CoT) \u4e4b\u7c7b\u7684\u6280\u672f\u5728\u5927\u8bed\u8a00\u6a21\u578b (LLM) \u63a8\u7406\u4efb\u52a1\u4e0a\u901a\u5e38\u4f1a\u751f\u6210\u5197\u957f\u7684\u5206\u6b65\u89e3\u91ca\uff0c\u4f46\u4eba\u7c7b\u901a\u5e38\u4f9d\u8d56\u4e8e\u4ec5\u6355\u83b7\u57fa\u672c\u4fe1\u606f\u7684\u7b80\u6d01\u8349\u7a3f\u3002<\/section>\n<section><\/section>\n<section>\u53d7\u6b64\u542f\u53d1\uff0c\u4ed6\u4eec\u63d0\u51fa\u4e86 Chain of Draft (CoD)\uff0c\u8fd9\u662f\u4e00\u79cd\u901a\u8fc7\u751f\u6210\u6700\u5c11\u4f46\u4fe1\u606f\u4e30\u5bcc\u7684\u4e2d\u95f4\u6b65\u9aa4\u6765\u51cf\u5c11\u5197\u957f\u7684\u63d0\u793a\u7b56\u7565\u3002\u56e0\u6b64\u4ece\u672c\u8d28\u4e0a\u8bb2\uff0c\u5b83\u662f\u4e00\u79cd\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u65b9\u6cd5\uff0c\u901a\u8fc7\u751f\u6210\u66f4\u5c11\u7684 token \u6765\u63d0\u9ad8\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u7684\u6548\u7387\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-dfab5235e9c4a9ba54a6c79b8b6cf6f2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-dfab5235e9c4a9ba54a6c79b8b6cf6f2.png\"><\/a><\/section>\n<section>\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cCoD \u7684\u63d0\u793a\u957f\u5ea6\u51e0\u4e4e\u4e0e\u6807\u51c6\u63d0\u793a\u76f8\u5f53\uff0c\u4f46\u5176\u51c6\u786e\u6027\u5374\u4e0e\u601d\u7ef4\u94fe\uff08CoT\uff09\u63d0\u793a\u4e0d\u76f8\u4e0a\u4e0b\u3002\u63a8\u7406\u6a21\u578b\u7684\u4e00\u5927\u4f18\u52bf\u5728\u4e8e\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u9605\u8bfb\u63a8\u7406\u8fc7\u7a0b\u6765\u5b66\u4e60\u5e76\u66f4\u597d\u5730\u8bc4\u4f30\u548c\u4fe1\u4efb\u6a21\u578b\u7684\u54cd\u5e94\u3002\u867d\u7136 CoD \u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u524a\u5f31\u4e86\u8fd9\u4e00\u4f18\u52bf\uff0c\u4f46\u5b83\u5728\u65e0\u9700\u5197\u957f\u4e2d\u95f4\u6b65\u9aa4\u7684\u573a\u666f\u4e2d\u5c55\u73b0\u51fa\u5de8\u5927\u6f5c\u529b \u2014\u2014 \u5b83\u4e0d\u4ec5\u80fd\u591f\u663e\u8457\u52a0\u5feb\u751f\u6210\u901f\u5ea6\uff0c\u8fd8\u80fd\u4fdd\u6301 CoT \u7684\u51c6\u786e\u6027\u3002\u56e0\u6b64 CoD \u4e3a\u9ad8\u6548\u63a8\u7406\u63d0\u4f9b\u4e86\u4e00\u79cd\u66f4\u5177\u5b9e\u7528\u6027\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/section>\n<section><\/section>\n<section>\u8be6\u89c1\u673a\u5668\u4e4b\u5fc3\u62a5\u9053\u300a<a data-itemshowtype=\"0\" data-linktype=\"2\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzA3MzI4MjgzMw==&amp;mid=2650958987&amp;idx=3&amp;sn=97e8ee4b2ad7a92cb151a6807af6f4fa&amp;scene=21#wechat_redirect\" target=\"_blank\">\u5168\u65b0 CoD \u98a0\u8986\u63a8\u7406\u8303\u5f0f\uff0c\u51c6\u786e\u7387\u63a5\u8fd1\u4f46 token \u6d88\u8017\u6210\u500d\u964d\u4f4e<\/a>\u300b\u3002<\/section>\n<section><\/section>\n<section><strong>14.\u300c\u66f4\u597d\u7684\u53cd\u9988\u548c\u7f16\u8f91\u6a21\u578b Better Feedback and Edit Models\u300d<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aDedicated Feedback and Edit Models Empower Inference-Time Scaling for Open-Ended General-Domain Tasks<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u5730\u5740: https:\/\/arxiv.org\/pdf\/2503.04378<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u8bb8\u591a\u6269\u5c55\u63a8\u7406\u65f6\u95f4\u63a8\u7406\u7684\u6280\u672f\u4f9d\u8d56\u4e8e\u5177\u6709\u53ef\u9a8c\u8bc1\u7b54\u6848\u7684\u4efb\u52a1\uff08\u5982\u53ef\u4ee5\u68c0\u67e5\u7684\u6570\u5b66\u548c\u4ee3\u7801\uff09\uff0c\u8fd9\u4f7f\u5f97\u5b83\u4eec\u5f88\u96be\u5e94\u7528\u4e8e\u5199\u4f5c\u548c\u4e00\u822c\u95ee\u9898\u89e3\u51b3\u7b49\u5f00\u653e\u5f0f\u4efb\u52a1\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u89e3\u51b3\u53ef\u9a8c\u8bc1\u7b54\u6848\u7684\u8fd9\u79cd\u9650\u5236\uff0c\u7814\u7a76\u4eba\u5458\u5f00\u53d1\u4e86\u4e00\u4e2a\u7cfb\u7edf\uff0c\u5176\u4e2d\u4e00\u4e2a\u6a21\u578b\u751f\u6210\u521d\u59cb\u54cd\u5e94\uff0c\u53e6\u4e00\u4e2a\u6a21\u578b\u63d0\u4f9b\u53cd\u9988\u300c\u53cd\u9988\u6a21\u578b\u300d\uff0c\u7b2c\u4e09\u4e2a\u6a21\u578b\u6839\u636e\u8be5\u53cd\u9988\u6539\u8fdb\u54cd\u5e94\u300c\u7f16\u8f91\u6a21\u578b\u300d\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-538df82ef5ed54c91298ed18300e24a4.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-538df82ef5ed54c91298ed18300e24a4.png\"><\/a><\/section>\n<section>\u4e3a\u4e86\u786e\u4fdd\u8fd9\u4e9b\u6a21\u578b\u7684\u6709\u6548\u6027\uff0c\u7814\u7a76\u4eba\u5458\u4f7f\u7528\u5927\u91cf\u4eba\u5de5\u6ce8\u91ca\u7684\u54cd\u5e94\u548c\u53cd\u9988\u6570\u636e\u96c6\u6765\u8bad\u7ec3\u8fd9\u4e9b\u4e13\u95e8\u7684 \u201c\u53cd\u9988\u201d \u548c \u201c\u7f16\u8f91\u201d \u6a21\u578b\u3002\u7136\u540e\uff0c\u8fd9\u4e9b\u6a21\u578b\u901a\u8fc7\u5728\u63a8\u7406\u65f6\u95f4\u5185\u751f\u6210\u66f4\u597d\u7684\u53cd\u9988\u548c\u8fdb\u884c\u66f4\u6709\u6548\u7684\u7f16\u8f91\u6765\u5e2e\u52a9\u6539\u8fdb\u54cd\u5e94\u3002<\/section>\n<section><\/section>\n<section><strong>\u603b\u7ed3<\/strong><\/section>\n<section><\/section>\n<section>\u53ef\u4ee5\u770b\u5230\uff0c\u73b0\u6709\u7684\u8bb8\u591a\u63a8\u7406\u6a21\u578b\u6280\u672f\u5305\u62ec\u7b80\u5355\u7684\u57fa\u4e8e token \u7684\u5e72\u9884\u63aa\u65bd\u5230\u590d\u6742\u7684\u57fa\u4e8e\u641c\u7d22\u548c\u4f18\u5316\u7b56\u7565\uff0c\u5b83\u4eec\u7684\u6838\u5fc3\u76ee\u6807\u662f\u589e\u52a0\u63a8\u7406\u8ba1\u7b97\u91cf\uff0c\u751a\u81f3\u53ef\u4ee5\u8ba9\u76f8\u5bf9\u8f83\u5c0f\u7684\u6a21\u578b\u5b9e\u73b0\u663e\u8457\u7684\u6539\u8fdb\u3002\u8fd9\u8868\u660e\u63a8\u7406\u7b56\u7565\u53ef\u4ee5\u5e2e\u52a9\u7f29\u5c0f\u8f83\u5c0f\u3001\u66f4\u5177\u6210\u672c\u6548\u76ca\u7684\u6a21\u578b\u4e0e\u8f83\u5927\u6a21\u578b\u4e4b\u95f4\u7684\u6027\u80fd\u5dee\u8ddd\u3002<\/section>\n<section><\/section>\n<section><strong>\u6210\u672c\u8b66\u544a<\/strong><\/section>\n<section><\/section>\n<section>\u5c3d\u7ba1\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u80fd\u591f\u663e\u8457\u63d0\u5347\u6a21\u578b\u6027\u80fd\uff0c\u4f46\u5176\u5e26\u6765\u7684\u63a8\u7406\u6210\u672c\u589e\u52a0\u662f\u4e00\u4e2a\u4e0d\u5bb9\u5ffd\u89c6\u7684\u95ee\u9898\u3002\u56e0\u6b64\uff0c\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c<strong>\u5f00\u53d1\u8005\u9700\u8981\u5728\u300c\u4f7f\u7528\u5177\u5907\u5927\u91cf\u63a8\u7406\u6269\u5c55\u7684\u5c0f\u6a21\u578b\u300d\u4e0e\u300c\u8bad\u7ec3\u66f4\u5927\u6a21\u578b\u4f46\u51cf\u5c11\u6216\u4e0d\u8fdb\u884c\u63a8\u7406\u6269\u5c55\u300d\u4e4b\u95f4\u8fdb\u884c\u6743\u8861<\/strong>\u3002\u8fd9\u79cd\u6743\u8861\u5fc5\u987b\u57fa\u4e8e\u6570\u5b66\u8ba1\u7b97\uff0c\u5e76\u7ed3\u5408\u6a21\u578b\u7684\u5b9e\u9645\u4f7f\u7528\u91cf\u6765\u8bc4\u4f30\u3002<\/section>\n<section><\/section>\n<section>\u4e3e\u4f8b\u6765\u8bf4\uff0c\u4f7f\u7528\u91cd\u5ea6\u63a8\u7406\u65f6\u95f4\u7f29\u653e\u7684 o1 \u6a21\u578b\u5b9e\u9645\u4e0a\u4ecd\u7136\u6bd4\u53ef\u80fd\u4e0d\u4f7f\u7528\u63a8\u7406\u65f6\u95f4\u7f29\u653e\u7684\u66f4\u5927\u7684 GPT-4.5 \u6a21\u578b\u7a0d\u5fae\u4fbf\u5b9c\u4e00\u4e9b\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-46d1e9dd1bb20b57bf4b156ed2327709.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-46d1e9dd1bb20b57bf4b156ed2327709.png\"><\/a><\/section>\n<section>\u7136\u800c\uff0c\u63a8\u7406\u65f6\u95f4\u6269\u5c55\u5e76\u975e\u4e07\u80fd\u4e4b\u7b56\u3002\u5c3d\u7ba1\u50cf\u8499\u7279\u5361\u7f57\u6811\u641c\u7d22\u3001\u81ea\u6211\u56de\u6eaf\u548c\u52a8\u6001\u6df1\u5ea6\u7f29\u653e\u7b49\u6280\u672f\uff0c\u867d\u80fd\u663e\u8457\u63d0\u5347\u63a8\u7406\u6027\u80fd\uff0c\u4f46\u6548\u679c\u4ecd\u53d6\u51b3\u4e8e\u5177\u4f53\u4efb\u52a1\u53ca\u5176\u96be\u5ea6\u3002\u6b63\u5982\u65e9\u671f\u4e00\u7bc7\u8bba\u6587\u6240\u6307\u51fa\u7684\uff0c<strong>\u6ca1\u6709\u54ea\u79cd\u63a8\u7406\u65f6\u8ba1\u7b97\u6269\u5c55\u6280\u672f\u80fd\u5728\u6240\u6709\u4efb\u52a1\u4e2d\u90fd\u8868\u73b0\u6700\u4f73<\/strong>\u3002<\/section>\n<section><\/section>\n<section>\u6b64\u5916\uff0c\u8bb8\u591a\u6b64\u7c7b\u65b9\u6cd5\u4e3a\u63d0\u5347\u63a8\u7406\u80fd\u529b\u800c\u727a\u7272\u4e86\u54cd\u5e94\u5ef6\u8fdf\uff0c\u800c\u8f83\u6162\u7684\u54cd\u5e94\u5ef6\u8fdf\u53ef\u80fd\u4f1a\u5f71\u54cd\u7528\u6237\u4f53\u9a8c\uff0c\u751a\u81f3\u8ba9\u90e8\u5206\u7528\u6237\u611f\u5230\u538c\u70e6\u3002<\/section>\n<section><\/section>\n<section><strong>\u5c55\u671b\u672a\u6765<\/strong><\/section>\n<section><\/section>\n<section>\u57fa\u4e8e\u4e0a\u8ff0\u7814\u7a76\u6210\u679c\uff0c\u672a\u6765\u5927\u5bb6\u53ef\u80fd\u5c06\u770b\u5230\u66f4\u591a\u300c\u56f4\u7ed5\u901a\u8fc7\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u8fdb\u884c\u63a8\u7406\u300d\u7814\u7a76\u7684\u4e24\u4e2a\u4e3b\u8981\u5206\u652f\u7684\u8bba\u6587\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u7eaf\u7cb9\u4ee5\u5f00\u53d1\u8d85\u8d8a\u57fa\u51c6\u7684\u6700\u4f73\u6a21\u578b\u4e3a\u4e2d\u5fc3\u7684\u7814\u7a76\u3002<\/section>\n<\/li>\n<li>\n<section>\u5173\u6ce8\u5728\u4e0d\u540c\u63a8\u7406\u4efb\u52a1\u4e4b\u95f4\u5e73\u8861\u6210\u672c\u548c\u6027\u80fd\u6743\u8861\u7684\u7814\u7a76\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u63a8\u7406\u65f6\u8ba1\u7b97\u6269\u5c55\u7684\u4f18\u52bf\u5728\u4e8e\uff0c\u5b83\u53ef\u4ee5\u5e94\u7528\u4e8e\u4efb\u4f55\u73b0\u6709\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u4f7f\u5176\u5728\u7279\u5b9a\u4efb\u52a1\u4e0a\u8868\u73b0\u5f97\u66f4\u4e3a\u51fa\u8272\u3002<\/section>\n<section><\/section>\n<section>\u5728 DeepSeek R1 \u53d1\u5e03\u4e4b\u540e\uff0c\u884c\u4e1a\u51fa\u73b0\u4e86\u4e00\u4e2a\u5f15\u4eba\u6ce8\u76ee\u7684\u8d8b\u52bf\uff0c\u5373\u300c\u6309\u9700\u601d\u8003\u300d\u3002<strong>\u5404\u5bb6\u516c\u53f8\u7eb7\u7eb7\u7ade\u76f8\u4e3a\u5176\u4ea7\u54c1\u589e\u6dfb\u63a8\u7406\u529f\u80fd\uff0c\u4ee5\u63d0\u5347\u6a21\u578b\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u7684\u8868\u73b0<\/strong>\u3002<\/section>\n<section><\/section>\n<section>\u4f5c\u8005\u8868\u793a\uff1a\u8fd8\u6709\u4e00\u4e2a\u503c\u5f97\u5173\u6ce8\u7684\u53d1\u5c55\u52a8\u6001\uff0c\u5927\u591a\u6570\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u63d0\u4f9b\u5546\u5f00\u59cb\u4e3a\u7528\u6237\u63d0\u4f9b\u542f\u7528\u6216\u7981\u7528\u300c\u601d\u8003\u300d\u529f\u80fd\u7684\u9009\u9879\u3002\u76ee\u524d\u8be5\u673a\u5236\u5e76\u672a\u516c\u5f00\uff0c\u4e0d\u8fc7\u5b83\u53ef\u80fd\u4e0e\u5177\u5907\u56de\u62e8\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u7684\u662f\u540c\u4e00\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section>\u603b\u4f53\u800c\u8a00\uff0c\u65e0\u8bba\u662f\u501f\u52a9\u63a8\u7406\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\uff0c\u8fd8\u662f\u8bad\u7ec3\u65f6\u95f4\u8ba1\u7b97\u6269\u5c55\u6765\u6dfb\u52a0\u63a8\u7406\u80fd\u529b\uff0c\u8fd9\u4e00\u8d8b\u52bf\u90fd\u662f 2025 \u5e74\u5927\u8bed\u8a00\u6a21\u578b\u53d1\u5c55\u5411\u524d\u8fc8\u51fa\u7684\u91cd\u8981\u4e00\u6b65\u3002<\/section>\n<section><\/section>\n<section>\u6700\u540e\uff0c\u53ef\u4ee5\u9884\u8ba1\u968f\u7740\u65f6\u95f4\u7684\u63a8\u79fb\u63a8\u7406\u5c06\u4e0d\u518d\u88ab\u89c6\u4f5c\u53ef\u9009\u6216\u7279\u6b8a\u7684\u529f\u80fd\uff0c\u800c\u662f\u4f1a\u6210\u4e3a\u4e00\u79cd\u6807\u51c6\u914d\u7f6e\uff0c\u5c31\u5982\u540c\u5982\u4eca\u6307\u4ee4\u5fae\u8c03\u6216\u57fa\u4e8e\u4eba\u7c7b\u53cd\u9988\u7684\u5f3a\u5316\u5b66\u4e60\uff08RLHF\uff09\u8c03\u6574\u6a21\u578b\u5df2\u6210\u4e3a\u539f\u59cb\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u5e38\u89c4\u64cd\u4f5c\u4e00\u6837\u3002<\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-04-01-9\" target=\"_blank\">DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:DeepSeek-R1 [&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":[71,68,81],"class_list":["post-39832","post","type-post","status-publish","format-standard","hentry","category-news","tag-rag","tag-68","tag-81"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587 - \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=39832\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=39832\",\"url\":\"https:\/\/17aitech.com\/?p=39832\",\"name\":\"DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=39832#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=39832#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png\",\"datePublished\":\"2025-04-01T17:02:42+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=39832#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=39832\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=39832#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=39832#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587\"}]},{\"@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":"DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587 - \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=39832","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=39832","url":"https:\/\/17aitech.com\/?p=39832","name":"DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=39832#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=39832#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png","datePublished":"2025-04-01T17:02:42+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=39832#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=39832"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=39832#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/04\/frc-692faac364a01ff4a8fad2ad6de52765.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=39832#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"DeepSeek-R1\u4e4b\u540e\u63a8\u7406\u6a21\u578b\u53d1\u5c55\u5982\u4f55\uff1fRaschka\u957f\u6587\u68b3\u7406\u540eR1\u65f6\u4ee314\u7bc7\u91cd\u8981\u8bba\u6587"}]},{"@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\/39832","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=39832"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/39832\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=39832"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=39832"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=39832"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}