{"id":38271,"date":"2025-02-13T04:00:57","date_gmt":"2025-02-12T20:00:57","guid":{"rendered":"https:\/\/17aitech.com\/?p=38271"},"modified":"2025-02-13T04:00:57","modified_gmt":"2025-02-12T20:00:57","slug":"8%e5%8d%a132b%e6%a8%a1%e5%9e%8b%e8%b6%85%e8%b6%8ao1%e9%a2%84%e8%a7%88%e7%89%88%e3%80%81deepseek-v3%ef%bc%8c%e6%99%ae%e6%9e%97%e6%96%af%e9%a1%bf%e3%80%81%e5%8c%97%e5%a4%a7%e6%8f%90%e5%87%ba%e5%b1%82","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=38271","title":{"rendered":"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-12-4\" target=\"_blank\">8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f<\/a><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png\"><\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"166\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<p>AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/p>\n<\/section>\n<\/blockquote>\n<p><strong>\u4e00\uff0e\u5f15\u8a00<\/strong><\/p>\n<p>\u63a8\u7406\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\uff0c\u5982 OpenAI \u7684 o1 \u7cfb\u5217\u3001Google \u7684 Gemini\u3001DeepSeek \u548c Qwen-QwQ \u7b49\uff0c\u901a\u8fc7\u6a21\u62df\u4eba\u7c7b\u63a8\u7406\u8fc7\u7a0b\uff0c\u5728\u591a\u4e2a\u4e13\u4e1a\u9886\u57df\u5df2\u8d85\u8d8a\u4eba\u7c7b\u4e13\u5bb6\uff0c\u5e76\u901a\u8fc7\u5ef6\u957f\u63a8\u7406\u65f6\u95f4\u63d0\u9ad8\u51c6\u786e\u6027\u3002\u63a8\u7406\u6a21\u578b\u7684\u6838\u5fc3\u6280\u672f\u5305\u62ec\u5f3a\u5316\u5b66\u4e60\uff08Reinforcement Learning\uff09\u548c\u63a8\u7406\u89c4\u6a21\uff08Inference scaling\uff09\u3002<\/p>\n<p>\u4e3b\u6d41\u7684\u5927\u6a21\u578b\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\uff0c\u5982 DPO\u3001PPO\u3001GRPO \u7b49\uff0c\u901a\u5e38\u9700\u8981\u5728\u5b8c\u6574\u7684\u601d\u7ef4\u94fe\u4e0a\u8fdb\u884c\u5fae\u8c03\uff0c\u9700\u8981\u9ad8\u8d28\u91cf\u6570\u636e\u3001\u7cbe\u786e\u7684\u5956\u52b1\u51fd\u6570\u3001\u5feb\u901f\u53cd\u9988\u548c\u5728\u7ebf\u8fed\u4ee3\u3001\u4ee5\u53ca\u5927\u91cf\u7684\u7b97\u529b\u3002\u5f53\u5904\u7406\u590d\u6742\u4efb\u52a1\uff0c\u5982\u9ad8\u7ea7\u6570\u5b66\u548c\u7f16\u7a0b\u95ee\u9898\u65f6\uff0c\u6a21\u578b\u9700\u8981\u66f4\u7ec6\u7c92\u5ea6\u7684\u641c\u7d22\u3001\u66f4\u7cbe\u786e\u7684\u63a8\u7406\u6b65\u9aa4\u548c\u66f4\u957f\u7684\u601d\u7ef4\u94fe\uff0c\u5bfc\u81f4\u72b6\u6001\u7a7a\u95f4\u548c\u7b56\u7565\u7a7a\u95f4\u7684\u89c4\u6a21\u6025\u5267\u6269\u5927\uff0c\u96be\u5ea6\u5927\u5e45\u4e0a\u5347\u3002<\/p>\n<p>Inference scaling \u7b56\u7565\uff0c\u4e0d\u4f9d\u8d56\u8bad\u7ec3\uff0c\u901a\u8fc7\u5ef6\u957f\u63a8\u7406\u65f6\u95f4\u8fdb\u4e00\u6b65\u63d0\u9ad8\u6a21\u578b\u7684 Reasoning \u80fd\u529b\u3002\u5e38\u89c1\u65b9\u6cd5\uff0c\u5982 Best-of-N \u6216\u8005\u8499\u7279\u5361\u6d1b\u6811\u641c\u7d22\uff08MCTS\uff09\uff0c\u5141\u8bb8 LLM \u540c\u65f6\u63a2\u7d22\u591a\u6761\u63a8\u7406\u8def\u5f84\uff0c\u6269\u5927\u641c\u7d22\u7a7a\u95f4\uff0c\u671d\u7740\u66f4\u6709\u5e0c\u671b\u7684\u65b9\u5411\u524d\u8fdb\u3002\u8fd9\u4e9b\u65b9\u6cd5\u8ba1\u7b97\u6210\u672c\u9ad8\uff0c\u7279\u522b\u662f\u6b65\u9aa4\u591a\u6216\u641c\u7d22\u7a7a\u95f4\u5927\u7684\u65f6\u5019\u3002\u91c7\u6837\u968f\u673a\u6027\u4f7f\u5f97\u786e\u5b9a\u6700\u4f73\u8def\u5f84\u56f0\u96be\uff0c\u4e14\u4f9d\u8d56\u624b\u52a8\u8bbe\u8ba1\u7684\u641c\u7d22\u7b56\u7565\u548c\u5956\u52b1\u51fd\u6570\uff0c\u9650\u5236\u4e86\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<p><strong>\u5728\u6b64\u80cc\u666f\u4e0b\uff0c\u666e\u6797\u65af\u987f\u5927\u5b66\u56e2\u961f\u8054\u5408\u5317\u4eac\u5927\u5b66\u56e2\u961f\u5408\u4f5c\u5f00\u53d1\u4e86\u540d\u4e3a ReasonFlux \u7684\u591a\u5c42\u6b21\uff08Hierarchical\uff09LLM \u63a8\u7406\u6846\u67b6\u3002<\/strong><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-d59bfd6400a8a5395ba8e30c4bc735cb.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-d59bfd6400a8a5395ba8e30c4bc735cb.png\"><\/a><\/p>\n<ul>\n<li>\n<p>\u6587\u7ae0\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2502.06772<\/p>\n<\/li>\n<li>\n<p>\u5f00\u6e90\u5730\u5740\uff1ahttps:\/\/github.com\/Gen-Verse\/ReasonFlux<\/p>\n<\/li>\n<\/ul>\n<p>\uff08\u8be5\u8bba\u6587\u4f5c\u8005\u7279\u522b\u58f0\u660e\uff1a\u672c\u5de5\u4f5c\u6ca1\u6709\u84b8\u998f\u6216\u7528\u4efb\u4f55\u65b9\u5f0f\u4f7f\u7528 DeepSeek R1\u3002\uff09<\/p>\n<p>\u57fa\u4e8e\u5c42\u6b21\u5316\u5f3a\u5316\u5b66\u4e60\uff08Hierachical Reinforcement Learning\uff09\u601d\u60f3\uff0cReasonFlux \u63d0\u51fa\u4e86\u4e00\u79cd\u66f4\u9ad8\u6548\u4e14\u901a\u7528\u7684\u5927\u6a21\u578b\u63a8\u7406\u8303\u5f0f\uff0c\u5b83\u5177\u6709\u4ee5\u4e0b\u7279\u70b9\uff1a<\/p>\n<ul>\n<li>\n<p><strong>\u601d\u7ef4\u6a21\u7248\uff1a<\/strong>ReasonFlux \u7684\u6838\u5fc3\u5728\u4e8e\u7ed3\u6784\u5316\u7684\u601d\u7ef4\u6a21\u677f\uff0c\u6bcf\u4e2a\u6a21\u7248\u62bd\u8c61\u4e86\u4e00\u4e2a\u6570\u5b66\u77e5\u8bc6\u70b9\u548c\u89e3\u9898\u6280\u5de7\u3002<strong>\u4ec5\u7528 500 \u4e2a\u901a\u7528\u7684\u601d\u7ef4\u6a21\u677f\u5e93<\/strong>\uff0c\u5c31\u53ef\u89e3\u51b3\u5404\u7c7b\u6570\u5b66\u96be\u9898\u3002<\/p>\n<\/li>\n<li>\n<p><strong>\u5c42\u6b21\u5316\u63a8\u7406\u548c\u5f3a\u53ef\u89e3\u91ca\u6027\uff1a<\/strong>ReasonFlux \u5229\u7528\u5c42\u6b21\u5316\u63a8\u7406\uff08Hierarchical Reasoning\uff09\u5c06\u601d\u7ef4\u6a21\u677f\u7ec4\u5408\u6210\u601d\u7ef4\u8f68\u8ff9\uff08Thought Template Trajectory\uff09\u3001\u518d\u5b9e\u4f8b\u5316\u5f97\u5230\u5b8c\u6574\u56de\u7b54\u3002\u6a21\u578b\u7684\u63a8\u7406\u8fc7\u7a0b\u4e0d\u518d\u662f \u201c\u9ed1\u76d2\u201d\uff0c\u800c\u662f\u6e05\u6670\u7684\u5c55\u73b0\u4e86\u63a8\u7406\u6b65\u9aa4\u548c\u4f9d\u636e\uff0c\u8fd9\u4e3a LLM \u7684\u53ef\u89e3\u91ca\u6027\u7814\u7a76\u63d0\u4f9b\u4e86\u65b0\u7684\u5de5\u5177\u548c\u89c6\u89d2\uff0c\u4e5f\u4e3a\u6a21\u578b\u7684\u8c03\u8bd5\u548c\u4f18\u5316\u63d0\u4f9b\u4e86\u4fbf\u5229\u3002\u4e0e DeepSeek-R1 \u548c OpenAI-o1 \u7b49\u6a21\u578b\u7684\u63a8\u7406\u65b9\u5f0f\u4e0d\u540c\uff0cReasonFlux \u5927\u5927\u538b\u7f29\u5e76\u51dd\u7ec3\u4e86\u63a8\u7406\u7684\u641c\u7d22\u7a7a\u95f4\uff0c\u63d0\u9ad8\u4e86\u5f3a\u5316\u5b66\u4e60\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u63d0\u9ad8\u4e86 inference scaling \u7684\u6548\u7387\u3002\u00a0<\/p>\n<\/li>\n<li>\n<p><strong>\u8f7b\u91cf\u7ea7\u7cfb\u7edf\uff1aReasonFlux \u4ec5 32B \u53c2\u6570\uff0c\u5f3a\u5316\u8bad\u7ec3\u53ea\u7528\u4e86 8 \u5757 NVIDIA A100-PCIE-80GB GPU\u3002<\/strong>\u5b83\u80fd\u901a\u8fc7\u81ea\u52a8\u6269\u5c55\u601d\u7ef4\u6a21\u677f\u6765\u63d0\u5347\u63a8\u7406\u80fd\u529b\uff0c\u66f4\u9ad8\u6548\u7075\u6d3b\u3002<\/p>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-daaa5896f356578743a26c24e8d4d8aa.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-daaa5896f356578743a26c24e8d4d8aa.png\"><\/a><\/section>\n<p>ReasonFlux-32B \u5728\u591a\u4e2a\u6570\u5b66\u63a8\u7406\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u4ec5\u4ec5\u7528\u4e86 500 \u4e2a\u57fa\u4e8e\u4e0d\u540c\u6570\u5b66\u77e5\u8bc6\u70b9\u7684\u601d\u7ef4\u6a21\u7248\uff0c\u5c31\u5c55\u73b0\u4e86\u5176\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\u548c\u8dfb\u8eab\u7b2c\u4e00\u68af\u961f\u7684\u5b9e\u529b\u3002<\/p>\n<section><strong>\u4e8c\uff0eReasonFlux\uff1a<\/strong><\/section>\n<section><strong>\u4e09\u5927\u5173\u952e\u6280\u672f\u6784\u5efa\u5927\u6a21\u578b\u63a8\u7406\u65b0\u6846\u67b6<\/strong><\/section>\n<section><\/section>\n<section>ReasonFlux \u7684\u6027\u80fd\u63d0\u5347\u5f97\u76ca\u4e8e\u5176\u4e09\u5927\u6838\u5fc3\u6280\u672f\uff1a<\/section>\n<section><\/section>\n<ol>\n<li>\n<section><strong>\u7ed3\u6784\u5316\u7684\u601d\u7ef4\u6a21\u677f\u62bd\u53d6\uff1a<\/strong>ReasonFlux \u5229\u7528\u5927\u8bed\u8a00\u6a21\u578b\u4ece\u4ee5\u5f80\u7684\u6570\u5b66\u95ee\u9898\u4e2d\u63d0\u53d6\u4e86\u4e00\u4e2a\u5305\u542b<strong>\u5927\u7ea6 500 \u4e2a\u7ed3\u6784\u5316\u601d\u7ef4\u6a21\u677f\u7684\u77e5\u8bc6\u5e93<\/strong>\u3002\u6bcf\u4e2a\u6a21\u677f\u90fd\u5305\u542b\u6807\u7b7e\u3001\u63cf\u8ff0\u3001\u9002\u7528\u8303\u56f4\u3001\u5e94\u7528\u6b65\u9aa4\u7b49\u4fe1\u606f\uff0c\u8fd9\u4e9b\u4fe1\u606f\u7ecf\u8fc7\u7ec4\u7ec7\u548c\u7ed3\u6784\u5316\u5904\u7406\uff0c\u4e3a LLM \u7684\u63a8\u7406\u63d0\u4f9b\u4e86\u5143\u77e5\u8bc6\u53c2\u8003\u3002\u8fd9\u4e9b\u6a21\u677f\u8986\u76d6\u4e86\u591a\u79cd\u6570\u5b66\u95ee\u9898\u7c7b\u578b\u548c\u89e3\u9898\u65b9\u6cd5\uff0c\u5982\u4e0d\u7b49\u5f0f\u6c42\u89e3\u3001\u4e09\u89d2\u51fd\u6570\u53d8\u6362\u3001\u6781\u503c\u5b9a\u7406\u7b49\uff0c\u662f ReasonFlux \u8fdb\u884c\u63a8\u7406\u7684\u57fa\u7840\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u591a\u5c42\u6b21\u5f3a\u5316\u5b66\u4e60\uff08Hierarchical RL\uff09<\/strong> \u2014\u00a0<strong>\u9009\u62e9\u6700\u4f18\u7684 Thought Template Trajectory\uff1a<\/strong>\u8be5\u7b97\u6cd5\u901a\u8fc7 Hierarchical Reinforcement Learning \u8bad\u7ec3\u4e00\u4e2a High-level \u7684 navigator\uff0c\u4f7f\u5176\u80fd\u591f\u5bf9\u8f93\u5165\u95ee\u9898\u8fdb\u884c\u62c6\u89e3\uff0c\u8f6c\u800c\u6c42\u89e3\u591a\u4e2a\u66f4\u7b80\u5355\u7684\u5b50\u95ee\u9898\uff0c\u6839\u636e\u5b50\u95ee\u9898\u7c7b\u578b\u4ece\u6a21\u677f\u5e93\u4e2d\u68c0\u7d22\u76f8\u5173\u7684\u601d\u7ef4\u6a21\u677f\uff0c\u5e76\u89c4\u5212\u51fa\u6700\u4f18\u7684 Thought Template Trajectory\u3002\u5b83\u53ef\u4ee5\u770b\u4f5c\u662f\u89e3\u51b3\u95ee\u9898\u7684 \u201c\u8def\u7ebf\u56fe\u201d\uff0c\u5b83\u7531\u4e00\u7cfb\u5217\u7684\u6a21\u677f\u7ec4\u5408\u800c\u6210\u3002\u8fd9\u79cd\u57fa\u4e8e Hierarchical RL \u7684\u4f18\u5316\u7b97\u6cd5\u901a\u8fc7\u5956\u52b1\u5728\u76f8\u4f3c\u95ee\u9898\u4e0a\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u63d0\u5347\u4e86\u63a8\u7406\u8f68\u8ff9\u7684\u9c81\u68d2\u6027\u548c\u6709\u6548\u6027\uff0c<strong>\u4f7f\u5f97 ReasonFlux \u80fd\u591f\u4e3e\u4e00\u53cd\u4e09\uff0c\u4e3a\u5404\u79cd\u6570\u5b66\u95ee\u9898\u751f\u6210\u6709\u6548\u7684\u601d\u7ef4\u6a21\u677f\u8f68\u8ff9\u3002<\/strong><\/section>\n<\/li>\n<li>\n<section><strong>\u65b0\u578b Inference Scaling \u7cfb\u7edf\uff1a<\/strong>\u8be5\u7cfb\u7edf\u5b9e\u73b0\u4e86\u7ed3\u6784\u5316\u6a21\u677f\u5e93\u548c inference LLM \u4e4b\u95f4\u7684\u591a\u8f6e\u4ea4\u4e92\u3002\u201cNavigator\u201d \u8d1f\u8d23\u89c4\u5212\u6a21\u677f\u8f68\u8ff9\u548c\u68c0\u7d22\u6a21\u677f\uff0cinference LLM \u8d1f\u8d23\u5c06\u6a21\u677f\u5b9e\u4f8b\u5316\u4e3a\u5177\u4f53\u7684\u63a8\u7406\u6b65\u9aa4\uff0c\u5e76\u901a\u8fc7\u5206\u6790\u4e2d\u95f4\u7ed3\u679c\u6765\u52a8\u6001\u8c03\u6574\u8f68\u8ff9\uff0c\u5b9e\u73b0\u9ad8\u6548\u7684\u63a8\u7406\u8fc7\u7a0b\u3002\u8fd9\u79cd\u4ea4\u4e92\u673a\u5236\u4f7f\u5f97 ReasonFlux \u80fd\u591f\u6839\u636e\u95ee\u9898\u7684\u5177\u4f53\u60c5\u51b5\u7075\u6d3b\u8c03\u6574\u63a8\u7406\u7b56\u7565\uff0c\u4ece\u800c\u63d0\u9ad8\u63a8\u7406\u7684\u51c6\u786e\u6027\u548c\u6548\u7387\u3002<\/section>\n<\/li>\n<\/ol>\n<section><\/section>\n<section><strong>\uff08a\uff09\u63a8\u7406\u793a\u4f8b\u5bf9\u6bd4\uff1a<\/strong><\/section>\n<section><\/section>\n<section>\u63a5\u4e0b\u6765\u6211\u4eec\u6765\u5206\u6790 ReasonFlux \u5728\u89e3\u51b3\u5b9e\u9645\u95ee\u9898\u4e0a\u76f8\u8f83\u4e8e o1-mini \u7684\u5bf9\u6bd4\u3002<\/section>\n<section><\/section>\n<section>\u6211\u4eec\u6765\u770b\u548c o1-mini \u7684\u5bf9\u6bd4<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-90450c40b5ff254aed5938ff4e59c377.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-90450c40b5ff254aed5938ff4e59c377.png\"><\/a><\/section>\n<section>\u5982\u4e0a\u56fe\u53ef\u77e5\uff0co1-mini \u5728\u9762\u5bf9\u8fd9\u9053\u96be\u9898\u65f6\uff0c<strong>\u5c1d\u8bd5\u4e86\u591a\u79cd\u7b56\u7565\uff0c\u4f46\u5747\u672a\u80fd\u627e\u5230\u6709\u6548\u7684\u7a81\u7834\u53e3\u3002<\/strong>\u5b83\u9996\u5148\u8bd5\u56fe\u901a\u8fc7\u5f15\u5165\u65b0\u53d8\u91cf\u548c\u5229\u7528\u5bf9\u79f0\u6027\u6765\u7b80\u5316\u65b9\u7a0b\u7ec4\uff0c\u4f46\u6536\u6548\u751a\u5fae\uff1b\u63a5\u7740\u53c8\u5c1d\u8bd5\u5047\u8bbe\u53d8\u91cf\u76f8\u7b49\u6765\u5bfb\u627e\u7279\u89e3\uff0c\u7ed3\u679c\u5374\u5f97\u51fa\u4e86\u77db\u76fe\uff1b\u968f\u540e\uff0c\u5b83\u8bd5\u56fe\u7528\u4e00\u4e2a\u53d8\u91cf\u8868\u793a\u5176\u4ed6\u53d8\u91cf\uff0c\u5e76\u5c1d\u8bd5\u5e73\u65b9\u53bb\u6839\u53f7\uff0c\u4f46\u590d\u6742\u7684\u8868\u8fbe\u5f0f\u4f7f\u5176\u671b\u800c\u5374\u6b65\uff1b\u6700\u540e\uff0c\u5b83\u751a\u81f3\u60f3\u5230\u4e86\u4e09\u89d2\u6362\u5143\uff0c\u4f46\u7531\u4e8e\u672a\u80fd\u6b63\u786e\u5e94\u7528\uff0c\u6700\u7ec8\u53ea\u80fd\u65e0\u5948\u5730\u653e\u5f03\u6c42\u89e3\u3002<\/section>\n<section><\/section>\n<section>\u76f8\u6bd4\u4e4b\u4e0b\uff0cReasonFlux \u7684\u89e3\u9898\u8fc7\u7a0b\u5982\u4e0b\uff1a<\/section>\n<section><\/section>\n<ol>\n<li>\n<section><strong>\u5206\u6790\u4e0e\u89c4\u5212\uff1a<\/strong>ReasonFlux \u9996\u5148\u5bf9\u9898\u76ee\u8fdb\u884c\u5206\u6790\uff0c\u786e\u5b9a\u4e86\u89e3\u9898\u7684\u4e3b\u8981\u6b65\u9aa4\uff1a\u521d\u6b65\u786e\u5b9a k \u503c\u7684\u8303\u56f4\u3001\u5229\u7528\u4e09\u89d2\u6362\u5143\u3001\u5316\u7b80\u65b9\u7a0b\u7ec4\u3001\u6c42\u89e3 \u03b8\u3001\u8ba1\u7b97\u76ee\u6807\u503c\u3002\u8fd9\u4e00\u6b65\u53cd\u6620\u4e86 ReasonFlux \u7684\u95ee\u9898\u5206\u6790\u548c\u89c4\u5212\u80fd\u529b\uff0c\u4e3a\u540e\u7eed\u89e3\u9898\u8fc7\u7a0b\u63d0\u4f9b\u4e86\u57fa\u7840\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u6a21\u677f\u5316\u63a8\u7406\uff1a<\/strong>ReasonFlux \u968f\u540e\u4f9d\u6b21\u5e94\u7528\u4e86 \u201c\u4e09\u89d2\u6362\u5143\u201d\u3001\u201c\u5316\u7b80\u65b9\u7a0b\u7ec4\u201d\u3001\u201c\u6c42\u89e3 \u03b8\u201d \u7b49\u6a21\u677f\uff0c\u5c06\u590d\u6742\u7684\u65b9\u7a0b\u7ec4\u9010\u6b65\u7b80\u5316\uff0c\u5e76\u6700\u7ec8\u6c42\u89e3\u51fa \u03b8 \u7684\u503c\u3002\u6bcf\u4e00\u6b65\u90fd\u4f9d\u636e\u6a21\u677f\u7684\u6307\u5bfc\uff0c\u65e8\u5728\u4fdd\u8bc1\u89e3\u9898\u8fc7\u7a0b\u7684\u51c6\u786e\u6027\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u9010\u6b65\u63a8\u5bfc\uff1a<\/strong>ReasonFlux \u6839\u636e\u6c42\u5f97\u7684\u89d2\u5ea6\u503c\uff0c\u8ba1\u7b97\u51fa (x, y, z) \u7684\u503c\uff0c\u5e76\u6700\u7ec8\u8ba1\u7b97\u51fa\u76ee\u6807\u503c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-0a68c7b990f62c453afa7a3546652598.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-0a68c7b990f62c453afa7a3546652598.png\"><\/a>\uff0c\u4ece\u800c\u5f97\u5230 (m=1, n=32, m+n=33)\u3002<strong>\u6574\u4e2a\u8fc7\u7a0b\u903b\u8f91\u6e05\u6670\uff0c\u6b65\u9aa4\u660e\u786e\uff0c\u5c55\u793a\u4e86 ReasonFlux \u7684\u89c4\u5212\u548c\u63a8\u7406\u80fd\u529b\u3002<\/strong><\/section>\n<\/li>\n<\/ol>\n<section><\/section>\n<section><strong>(b) \u65b0\u7684 inference scaling law\uff1a<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-9dd6b57e0ee7dec45781b8d7762ee8fa.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-9dd6b57e0ee7dec45781b8d7762ee8fa.png\"><\/a><\/section>\n<section>\u5982\u4e0a\u56fe\u6240\u793a\uff0c\u968f\u7740\u95ee\u9898\u590d\u6742\u5ea6\u7684\u589e\u52a0\uff0cReasonFlux \u6b63\u786e\u89e3\u7b54\u95ee\u9898\u65f6\u6240\u9700\u7684\u6a21\u677f\u6570\u91cf\u548c\u4ea4\u4e92\u8f6e\u6570\u4e5f\u76f8\u5e94\u589e\u52a0\u3002\u8fd9\u8868\u660e ReasonFlux \u80fd\u591f\u6839\u636e\u95ee\u9898\u7684\u96be\u5ea6\u52a8\u6001\u8c03\u6574\u63a8\u7406\u7b56\u7565\uff0c\u4f53\u73b0\u4e86\u5176\u4f18\u79c0\u7684\u81ea\u9002\u5e94\u80fd\u529b\u3002\u5e76\u4e14\u53ef\u4ee5\u89c2\u5bdf\u5230\uff0c\u4ea4\u4e92\u8f6e\u6570\u7684\u589e\u957f\u8d8b\u52bf\u7565\u9ad8\u4e8e\u6a21\u677f\u6570\u91cf\uff0c\u8fd9\u610f\u5473\u7740\u89c4\u5212\u80fd\u529b\u7684\u63d0\u5347\u5bf9\u89e3\u51b3\u590d\u6742\u95ee\u9898\u81f3\u5173\u91cd\u8981\u3002<\/section>\n<section><\/section>\n<section><strong>\u4e09\uff0e\u4e3b\u6d41\u63a8\u7406\u8303\u5f0f\u5bf9\u6bd4\uff1a<\/strong><\/section>\n<section><strong>ReasonFlux \u00a0vs Best-of-N &amp; MCTS<\/strong><\/section>\n<section><\/section>\n<section>\u76ee\u524d\uff0c\u63d0\u5347 LLM \u63a8\u7406\u6027\u80fd\u7684\u4e3b\u6d41\u65b9\u6cd5\u901a\u5e38\u4f9d\u8d56\u4e8e\u589e\u52a0\u6a21\u578b\u89c4\u6a21\u548c\u8ba1\u7b97\u8d44\u6e90\u3002\u4f8b\u5982\uff0c\u589e\u52a0\u6a21\u578b\u53c2\u6570\u91cf\u3001\u91c7\u7528 Best-of-N \u6216\u8499\u7279\u5361\u6d1b\u6811\u641c\u7d22 (MCTS) \u7b49\u65b9\u6cd5\u6765\u6269\u5927\u641c\u7d22\u7a7a\u95f4\u4ee5\u5bfb\u627e\u66f4\u4f18\u89e3\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u5f80\u5f80\u8ba1\u7b97\u6210\u672c\u8f83\u9ad8\uff0c\u4e14\u6a21\u578b\u7684\u63a8\u7406\u8fc7\u7a0b\u96be\u4ee5\u89e3\u91ca\u3002<\/section>\n<section><\/section>\n<section><strong>ReasonFlux \u91c7\u7528\u4e86\u4e00\u79cd\u4e0d\u540c\u7684\u65b9\u6cd5\uff0c\u901a\u8fc7\u6784\u5efa\u7ed3\u6784\u5316\u7684\u601d\u7ef4\u6a21\u677f\u5e93\u548c\u8bbe\u8ba1\u65b0\u7684\u5c42\u6b21\u5316\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\uff0c\u5b9e\u73b0\u4e86\u4e00\u79cd\u66f4\u9ad8\u6548\u548c\u53ef\u89e3\u91ca\u7684\u63a8\u7406\u65b9\u5f0f\u3002<\/strong><\/section>\n<section><\/section>\n<section>\u4f20\u7edf\u7684 Inference Scaling \u65b9\u6cd5\uff0c\u5982 Best-of-N \u548c MCTS\uff0c\u4e3b\u8981\u901a\u8fc7\u6269\u5927\u641c\u7d22\u7a7a\u95f4\u6765\u63d0\u9ad8\u51c6\u786e\u7387\u3002\u4f46\u968f\u7740\u95ee\u9898\u590d\u6742\u5ea6\u7684\u589e\u52a0\uff0c\u641c\u7d22\u7a7a\u95f4\u5448\u6307\u6570\u7ea7\u589e\u957f\uff0c\u5bfc\u81f4\u8ba1\u7b97\u6210\u672c\u663e\u8457\u4e0a\u5347\u3002<\/section>\n<section><\/section>\n<section>\u5728 ReasonFlux \u7684\u63a8\u7406\u8fc7\u7a0b\u4e2d\uff0cNavigator \u4e0e Inference LLM \u4e4b\u95f4\u5b58\u5728\u591a\u8f6e\u4ea4\u4e92\u3002Inference LLM \u6839\u636e Navigator \u7ed9\u51fa\u7684\u6a21\u677f\u8f68\u8ff9\u6267\u884c\u63a8\u7406\u6b65\u9aa4\u540e\uff0cNavigator \u4f1a\u5bf9\u6267\u884c\u7ed3\u679c\u8fdb\u884c\u8bc4\u4f30<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5263e92ce0c25061a4e1a24f108f68ce.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5263e92ce0c25061a4e1a24f108f68ce.png\"><\/a>\u3002\u5982\u516c\u5f0f<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5270607f21559afa05b1eaac9d9a79b7.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5270607f21559afa05b1eaac9d9a79b7.png\"><\/a>\u6240\u793a\uff0c\u6839\u636e\u8bc4\u4f30\u7ed3\u679c\uff0cNavigator \u4f1a\u52a8\u6001\u8c03\u6574\u6a21\u677f\u8f68\u8ff9<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-44a8cd27f12e6d1c39aa8def4724f293.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-44a8cd27f12e6d1c39aa8def4724f293.png\"><\/a>\uff0c\u4f8b\u5982\u4fee\u6539\u5f53\u524d\u6b65\u9aa4\u7684\u6a21\u677f\u3001\u6dfb\u52a0\u6216\u5220\u9664\u6b65\u9aa4\u7b49\u3002\u8fd9\u79cd\u8fed\u4ee3\u4f18\u5316\u7684\u673a\u5236\u4f7f\u5f97 ReasonFlux \u80fd\u591f\u6839\u636e\u95ee\u9898\u7684\u5177\u4f53\u60c5\u51b5\u7075\u6d3b\u8c03\u6574\u63a8\u7406\u7b56\u7565\uff0c\u4ece\u800c\u63d0\u9ad8\u63a8\u7406\u7684\u51c6\u786e\u6027\u548c\u6548\u7387\u3002<\/section>\n<section><\/section>\n<section><strong>ReasonFlux \u901a\u8fc7\u5f15\u5165\u7ed3\u6784\u5316\u7684\u601d\u7ef4\u6a21\u677f\uff0c\u5c06\u641c\u7d22\u7a7a\u95f4\u4ece \u201c\u539f\u59cb\u89e3\u7a7a\u95f4\u201d \u7f29\u5c0f\u5230 \u201c\u6a21\u677f\u7a7a\u95f4\u201d\uff0c\u4ece\u800c\u964d\u4f4e\u4e86\u641c\u7d22\u7684\u96be\u5ea6\u548c\u6210\u672c\u3002<\/strong>\u5982\u679c\u8bf4\u4f20\u7edf\u7684\u63a8\u7406\u8303\u5f0f\u662f \u201c\u5927\u6d77\u635e\u9488\u201d\uff0c\u90a3\u4e48 ReasonFlux \u5219\u662f \u201c\u6309\u56fe\u7d22\u9aa5\u201d\u3002\u8fd9\u4e9b\u6a21\u677f\u5e76\u975e\u7b80\u5355\u7684\u89c4\u5219\u5806\u780c\uff0c\u800c\u662f\u7ecf\u8fc7\u63d0\u70bc\u548c\u7ed3\u6784\u5316\u5904\u7406\u7684\u77e5\u8bc6\u6a21\u677f\uff0c\u5b83\u4eec\u5c06\u590d\u6742\u7684\u63a8\u7406\u8fc7\u7a0b\u5206\u89e3\u4e3a\u4e00\u7cfb\u5217\u53ef\u590d\u7528\u7684\u6b65\u9aa4\uff0c\u4ece\u800c\u63d0\u5347\u4e86\u63a8\u7406\u7684\u6548\u7387\u548c\u51c6\u786e\u7387\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-b86dad42bb037d47016f80c84acb3e28.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-b86dad42bb037d47016f80c84acb3e28.png\"><\/a><\/section>\n<section>\u5982\u4e0a\u56fe\u6240\u793a\uff0c\u968f\u7740\u95ee\u9898\u96be\u5ea6\u7684\u63d0\u5347\uff0cBest-of-N \u548c MCTS \u7684\u63a2\u7d22\u6210\u672c\uff08\u4f8b\u5982\u91c7\u6837\u8f68\u8ff9\u6570\u91cf\u548c\u8fed\u4ee3\u6b21\u6570\uff09\u663e\u8457\u589e\u52a0\uff0c\u800c ReasonFlux \u7684\u63a2\u7d22\u6210\u672c\uff08\u4ea4\u4e92\u8f6e\u6570\uff09\u5219\u4fdd\u6301\u5728\u8f83\u4f4e\u6c34\u5e73\u4e14\u76f8\u5bf9\u7a33\u5b9a\u3002\u8fd9\u8bf4\u660e ReasonFlux \u80fd\u591f\u66f4\u9ad8\u6548\u5730\u5229\u7528\u5df2\u6709\u7684\u77e5\u8bc6\u6a21\u677f\u6765\u89e3\u51b3\u95ee\u9898\uff0c\u800c\u4e0d\u9700\u8981\u50cf Best-of-N \u548c MCTS \u90a3\u6837\u8fdb\u884c\u5927\u91cf\u7684\u8bd5\u9519\u548c\u63a2\u7d22\u3002\u8fd9\u5f97\u76ca\u4e8e ReasonFlux \u7684\u7ed3\u6784\u5316\u6a21\u677f\u5e93\u548c\u6a21\u677f\u8f68\u8ff9\u89c4\u5212\u673a\u5236\uff0c\u4f7f\u5176\u80fd\u591f\u5728\u66f4\u5c0f\u7684\u641c\u7d22\u7a7a\u95f4\u5185\u627e\u5230\u6b63\u786e\u7684\u63a8\u7406\u8def\u5f84\u3002<\/section>\n<section><\/section>\n<section><strong>\u56db\uff0e\u8bad\u7ec3\u53ca\u63a8\u7406\u6846\u67b6\u4ecb\u7ecd<\/strong><\/section>\n<section><\/section>\n<section>\u4e0b\u56fe\u5c55\u793a\u4e86 ReasonFlux \u7684\u8bad\u7ec3\u6846\u67b6\uff0c\u5176\u6838\u5fc3\u5728\u4e8e\u5229\u7528\u7ed3\u6784\u5316\u7684\u601d\u7ef4\u6a21\u677f\u5e93\u548c\u57fa\u4e8e\u601d\u7ef4\u6a21\u677f\u8f68\u8ff9\u5956\u52b1\u7684\u5c42\u6b21\u5316\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\uff0c\u8bad\u7ec3\u51fa\u4e00\u4e2a\u80fd\u591f\u8fdb\u884c\u9ad8\u6548\u63a8\u7406\u7684\u5927\u6a21\u578b\u3002\u6574\u4e2a\u8bad\u7ec3\u8fc7\u7a0b\u53ef\u4ee5\u5206\u4e3a\u4e24\u4e2a\u4e3b\u8981\u9636\u6bb5\uff1a<strong>\u7ed3\u6784\u5316\u77e5\u8bc6\u5b66\u4e60\u548c\u601d\u7ef4\u6a21\u677f\u8f68\u8ff9\u4f18\u5316\u3002<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c4c817513047d6a953c7c15c196c2e01.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c4c817513047d6a953c7c15c196c2e01.png\"><\/a><\/section>\n<section>1. <strong>\u7ed3\u6784\u5316\u77e5\u8bc6\u5b66\u4e60\u9636\u6bb5\uff1a<\/strong>\u8fd9\u4e2a\u9636\u6bb5\u7684\u76ee\u6807\u662f\u8ba9\u6a21\u578b\u5b66\u4e60\u601d\u7ef4\u6a21\u677f\u5e93\u4e2d\u8574\u542b\u7684\u7ed3\u6784\u5316\u77e5\u8bc6\u3002\u8fd9\u4e9b\u7ed3\u6784\u5316\u7684 Thought template \u683c\u5f0f\u5982\u4e0b\u56fe\u6240\u793a\uff1a<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-9001f487b10da4c743fa59a3208ac198.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-9001f487b10da4c743fa59a3208ac198.png\"><\/a><\/section>\n<section>\u7136\u540e\uff0c\u6211\u4eec\u5229\u7528\u8fd9\u4e9b\u7ed3\u6784\u5316\u6a21\u677f\u6570\u636e\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-37e9ca93d68badf08625473df1bc31ad.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-37e9ca93d68badf08625473df1bc31ad.png\"><\/a>\u5bf9\u4e00\u4e2a\u57fa\u7840 LLM \u8fdb\u884c\u5fae\u8c03\uff0c\u5f97\u5230\u6a21\u578b<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-fe53aae3b130f430e044f6a0358332bf.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-fe53aae3b130f430e044f6a0358332bf.png\"><\/a>\u3002\u8bad\u7ec3\u7684\u76ee\u6807\u662f\u8ba9\u6a21\u578b\u80fd\u591f\u6839\u636e\u6a21\u677f\u7684\u540d\u79f0\u548c\u6807\u7b7e\uff0c\u751f\u6210\u5bf9\u5e94\u7684\u63cf\u8ff0\u548c\u9002\u7528\u8303\u56f4 \u3002\u901a\u8fc7\u8fd9\u4e2a\u9636\u6bb5\u7684\u8bad\u7ec3\uff0c\u6a21\u578b\u5b66\u4e60\u5230\u4e86\u6a21\u677f\u5e93\u4e2d\u8574\u542b\u7684\u4e30\u5bcc\u77e5\u8bc6\uff0c\u5e76\u5177\u5907\u4e86\u521d\u6b65\u7684\u6a21\u677f\u7406\u89e3\u548c\u5e94\u7528\u80fd\u529b\u3002<\/section>\n<section><\/section>\n<section>2.\u00a0<strong>\u6a21\u677f\u8f68\u8ff9\u4f18\u5316\u9636\u6bb5<\/strong><strong>\uff1a<\/strong>\u8fd9\u4e2a\u9636\u6bb5\u7684\u76ee\u6807\u662f\u8bad\u7ec3\u6a21\u578b\u751f\u6210\u6709\u6548\u7684\u6a21\u677f\u8f68\u8ff9\uff0c\u5373\u9488\u5bf9\u7279\u5b9a\u95ee\u9898\uff0c\u9009\u62e9\u5408\u9002\u7684\u6a21\u677f\u5e76\u8fdb\u884c\u6392\u5e8f\uff0c\u5f62\u6210\u89e3\u51b3\u95ee\u9898\u7684 \u201c\u8def\u7ebf\u56fe\u201d\u3002\u6211\u4eec\u5229\u7528\u65b0\u9896\u7684\u57fa\u4e8e<strong>\u00a0Thought Template Trajectory \u7684 Hierarchical RL<\/strong> \u7b97\u6cd5\u6765\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\u3002\u5728\u8fd9\u4e2a\u9636\u6bb5\uff0c\u6211\u4eec\u4f7f\u7528\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-4402493898a853d7ef1ff9cd861f98fd.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-4402493898a853d7ef1ff9cd861f98fd.png\"><\/a>\u6a21\u578b\u9488\u5bf9\u8f93\u5165\u95ee\u9898\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5f60dfc2810b888c4cc8eec48f211b01.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5f60dfc2810b888c4cc8eec48f211b01.png\"><\/a>\u751f\u6210\u591a\u4e2a\u5019\u9009\u7684 high-level \u601d\u7ef4\u6a21\u677f\u8f68\u8ff9<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-03d5b102690e3ee693f8943f8e4f2053.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-03d5b102690e3ee693f8943f8e4f2053.png\"><\/a>\u3002\u6bcf\u4e2a\u8f68\u8ff9\u7531\u4e00\u7cfb\u5217\u6b65\u9aa4\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-3f9292c9422c8e5dfed14bba9b216715.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-3f9292c9422c8e5dfed14bba9b216715.png\"><\/a>\u7ec4\u6210\uff0c\u6bcf\u4e2a\u6b65\u9aa4\u90fd\u5173\u8054\u5230\u4e00\u4e2a\u7279\u5b9a\u7684\u6a21\u677f\u3002\u4e3a\u4e86\u8bc4\u4f30\u8f68\u8ff9\u7684\u8d28\u91cf\uff0c\u6211\u4eec\u6784\u5efa\u4e86\u4e00\u7ec4\u4e0e\u8f93\u5165\u95ee\u9898\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c6827fe4190fb19884e7c3c25e11e635.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c6827fe4190fb19884e7c3c25e11e635.png\"><\/a>\u76f8\u4f3c\u7684\u95ee\u9898\u96c6<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f26e46a9e678244fdc04a9afeae78af6.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f26e46a9e678244fdc04a9afeae78af6.png\"><\/a>\u3002\u7136\u540e\uff0c\u6211\u4eec\u5229\u7528 inference LLM\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-51e07eca9b7d6b22c26ebf899be4b89d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-51e07eca9b7d6b22c26ebf899be4b89d.png\"><\/a>\u6839\u636e\u6a21\u677f\u8f68\u8ff9\u5bf9\u8fd9\u4e9b\u76f8\u4f3c\u95ee\u9898\u8fdb\u884c\u5177\u4f53\u7684\u89e3\u7b54\uff0c\u5e76\u8ba1\u7b97\u5e73\u5747\u51c6\u786e\u7387\u4f5c\u4e3a\u8f68\u8ff9\u7684\u5956\u52b1<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7b58bc33b6a655d27269fd13198a1f81.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7b58bc33b6a655d27269fd13198a1f81.png\"><\/a>\u3002\u57fa\u4e8e\u8fd9\u4e2a\u5956\u52b1\u4fe1\u53f7\uff0c\u6211\u4eec\u6784\u5efa\u4e86\u4f18\u5316\u6837\u672c\u5bf9<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7e7de22d39b34f972ffd70ff37e5e45e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7e7de22d39b34f972ffd70ff37e5e45e.png\"><\/a>\uff0c\u5176\u4e2d<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1cd38fb844dc3b7cd3b8fd951b40ae90.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1cd38fb844dc3b7cd3b8fd951b40ae90.png\"><\/a>\u3002\u7136\u540e\uff0c\u6211\u4eec\u5229\u7528\u8fd9\u4e9b\u6837\u672c\u5bf9\uff0c\u901a\u8fc7 DPO \u5bf9\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-b3e1438f4fe8f16ba17710e8931a485e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-b3e1438f4fe8f16ba17710e8931a485e.png\"><\/a>\u8fdb\u884c\u8fdb\u4e00\u6b65\u4f18\u5316\uff0c\u5f97\u5230\u6700\u7ec8\u7684 navigator \u6a21\u578b<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-bdf024d133171ffee34252ef1da89afb.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-bdf024d133171ffee34252ef1da89afb.png\"><\/a>\uff0c\u4e5f\u5c31\u662f\u6211\u4eec\u7684 ReasonFlux \u6a21\u578b\u3002\u00a0<\/section>\n<section><\/section>\n<section><strong>\u901a\u8fc7\u8fd9\u4e24\u4e2a\u9636\u6bb5\u7684\u8bad\u7ec3\uff0cReasonFlux \u6a21\u578b\u4e0d\u4ec5\u5b66\u4e60\u5230\u4e86\u7ed3\u6784\u5316\u7684\u6a21\u677f\u77e5\u8bc6\uff0c\u8fd8\u5b66\u4f1a\u4e86\u5982\u4f55\u9488\u5bf9\u7279\u5b9a\u95ee\u9898\u9009\u62e9\u548c\u7ec4\u5408\u6a21\u677f\uff0c\u5f62\u6210\u6709\u6548\u7684\u63a8\u7406\u8def\u5f84\u3002\u8fd9\u79cd\u80fd\u529b\u4f7f\u5f97 ReasonFlux \u80fd\u591f\u9ad8\u6548\u5730\u89e3\u51b3\u5404\u79cd\u590d\u6742\u7684\u6570\u5b66\u63a8\u7406\u95ee\u9898\u3002<\/strong><\/section>\n<section><\/section>\n<section>\u4e0b\u56fe\u662f ReasonFlux \u7684\u63a8\u7406\u6846\u67b6\u3002\u5176\u6838\u5fc3\u5728\u4e8e navigator\u3001inference LLM \u548c\u7ed3\u6784\u5316\u6a21\u677f\u5e93\u4e4b\u95f4\u7684\u591a\u8f6e\u4ea4\u4e92\u3002<strong>\u8fd9\u79cd\u4ea4\u4e92\u673a\u5236\u4f7f\u5f97 ReasonFlux \u80fd\u591f\u6839\u636e\u95ee\u9898\u7684\u5177\u4f53\u60c5\u51b5\u7075\u6d3b\u8c03\u6574\u63a8\u7406\u7b56\u7565\uff0c\u4ece\u800c\u63d0\u9ad8\u63a8\u7406\u7684\u51c6\u786e\u6027\u548c\u6548\u7387\u3002<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-83fe0ef7fe1f0ac03d9288db5cf6f88b.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-83fe0ef7fe1f0ac03d9288db5cf6f88b.png\"><\/a><\/section>\n<section>\u4ee5\u4e0b\u662f ReasonFlux \u7684\u63a8\u7406\u6d41\u7a0b\uff1a<\/section>\n<section><\/section>\n<section>1.\u00a0<strong>\u95ee\u9898\u62bd\u8c61\uff1a<\/strong>\u7ed9\u5b9a\u4e00\u4e2a\u8f93\u5165\u95ee\u9898<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-17c96193ba46eef556ac64b28ff258db.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-17c96193ba46eef556ac64b28ff258db.png\"><\/a>\uff0cReasonFlux\uff08\u5373 navigator<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-47e835863ddcb3b6490671f9e422da6b.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-47e835863ddcb3b6490671f9e422da6b.png\"><\/a>\uff09\u9996\u5148\u5bf9\u5176\u8fdb\u884c\u5206\u6790\uff0c\u5e76\u63d0\u53d6\u51fa\u95ee\u9898\u7684\u6838\u5fc3\u6570\u5b66\u6982\u5ff5\u548c\u5173\u7cfb\uff0c\u5f62\u6210\u4e00\u4e2a\u62bd\u8c61\u8868\u793a<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-6f4bae8a37fcd96aa86787365af11f1e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-6f4bae8a37fcd96aa86787365af11f1e.png\"><\/a>\u3002\u8fd9\u4e00\u6b65\u53ef\u4ee5\u7406\u89e3\u4e3a\u5bf9\u95ee\u9898\u8fdb\u884c \u201c\u964d\u7ef4\u201d\uff0c\u63d0\u53d6\u51fa\u95ee\u9898\u7684\u672c\u8d28\u7279\u5f81\u3002<\/section>\n<section>2.\u00a0<strong>\u8f68\u8ff9\u89c4\u5212\uff1a<\/strong>\u57fa\u4e8e\u95ee\u9898\u7684\u62bd\u8c61\u8868\u793a<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7de475e200eb0e52f4cac1224ba28eb0.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7de475e200eb0e52f4cac1224ba28eb0.png\"><\/a>\uff0cReasonFlux \u89c4\u5212\u51fa\u4e00\u4e2a\u6700\u4f18\u7684\u6a21\u677f\u8f68\u8ff9<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f68ff5177144fe67001ac5b0f93b5dd3.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f68ff5177144fe67001ac5b0f93b5dd3.png\"><\/a>\u3002\u8fd9\u4e2a\u8f68\u8ff9\u53ef\u4ee5\u770b\u4f5c\u662f\u89e3\u51b3\u95ee\u9898\u7684 \u201c\u8def\u7ebf\u56fe\u201d\uff0c\u5b83\u7531\u4e00\u7cfb\u5217\u6b65\u9aa4\u7ec4\u6210\uff0c\u6bcf\u4e2a\u6b65\u9aa4\u90fd\u5bf9\u5e94\u4e00\u4e2a\u7279\u5b9a\u7684\u6a21\u677f\u3002<\/section>\n<section>3.\u00a0<strong>\u6a21\u677f\u68c0\u7d22\uff1a<\/strong>\u6839\u636e\u8f68\u8ff9\u4e2d\u7684\u6bcf\u4e2a\u6b65\u9aa4\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-2e64473ac44b369165036ad52eee5fee.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-2e64473ac44b369165036ad52eee5fee.png\"><\/a>\u5173\u8054\u7684\u6a21\u677f\u540d\u79f0\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ba8aca993b13f15a4455645ec8ceafab.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ba8aca993b13f15a4455645ec8ceafab.png\"><\/a>\u548c\u6807\u7b7e<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c42591fa29210725c6a6ce19e4cf55ea.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c42591fa29210725c6a6ce19e4cf55ea.png\"><\/a>\uff0cReasonFlux \u4ece\u7ed3\u6784\u5316\u6a21\u677f\u5e93\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-476b22d6d97212316796fcc82f74fdc1.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-476b22d6d97212316796fcc82f74fdc1.png\"><\/a>\u4e2d\u68c0\u7d22\u51fa\u76f8\u5173\u7684\u6a21\u677f\u96c6\u5408\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-b918f6154822268fe9da19686bc15f13.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-b918f6154822268fe9da19686bc15f13.png\"><\/a>\u3002<\/section>\n<section>4.\u00a0<strong>\u6a21\u677f\u5b9e\u4f8b\u5316:<\/strong> Inference LLM\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ffae2dee71020f822787bd320091ef18.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ffae2dee71020f822787bd320091ef18.png\"><\/a>\u6839\u636e\u68c0\u7d22\u5230\u7684\u6a21\u677f\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-94f138b0946156e63b5eb0f5ebf2f3df.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-94f138b0946156e63b5eb0f5ebf2f3df.png\"><\/a>\u548c\u8f93\u5165\u95ee\u9898\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-21289c238fe3a84d73ba154b604b3ae4.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-21289c238fe3a84d73ba154b604b3ae4.png\"><\/a>\u7684\u5177\u4f53\u4fe1\u606f\uff0c\u5c06\u8f68\u8ff9\u4e2d\u7684\u6bcf\u4e2a\u6b65\u9aa4\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-e9f41259d0568b8f28d4e57ff926258b.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-e9f41259d0568b8f28d4e57ff926258b.png\"><\/a>\u5b9e\u4f8b\u5316\u4e3a\u5177\u4f53\u7684\u63a8\u7406\u6b65\u9aa4\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-504a9168af2130662ee3ae98a32d541d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-504a9168af2130662ee3ae98a32d541d.png\"><\/a>\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u53ef\u4ee5\u7406\u89e3\u4e3a\u5c06\u62bd\u8c61\u7684\u6a21\u677f\u5e94\u7528\u5230\u5177\u4f53\u7684\u95ee\u9898\u4e2d\u3002<\/section>\n<section>5.\u00a0<strong>\u8f68\u8ff9\u8c03\u6574:<\/strong> ReasonFlux \u4f1a\u8bc4\u4f30\u6bcf\u4e2a\u5b9e\u4f8b\u5316\u6b65\u9aa4\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5d49afa43c078abb2259536960792a86.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5d49afa43c078abb2259536960792a86.png\"><\/a>\u7684\u6267\u884c\u7ed3\u679c\uff0c\u5e76\u6839\u636e\u8bc4\u4f30\u7ed3\u679c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-eeafd3552162ada360eef2ff9103c579.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-eeafd3552162ada360eef2ff9103c579.png\"><\/a>\u52a8\u6001\u8c03\u6574\u6a21\u677f\u8f68\u8ff9\u3002\u4f8b\u5982\uff0c\u5982\u679c\u53d1\u73b0\u5f53\u524d\u6b65\u9aa4\u7684\u6a21\u677f\u4e0d\u9002\u7528\uff0cReasonFlux \u53ef\u80fd\u4f1a\u9009\u62e9\u53e6\u4e00\u4e2a\u6a21\u677f\uff0c\u6216\u8005\u6dfb\u52a0\u65b0\u7684\u6b65\u9aa4\u3002\u8fd9\u79cd\u8fed\u4ee3\u4f18\u5316\u7684\u673a\u5236\u4f7f\u5f97 ReasonFlux \u80fd\u591f\u6839\u636e\u95ee\u9898\u7684\u5177\u4f53\u60c5\u51b5\u7075\u6d3b\u8c03\u6574\u63a8\u7406\u7b56\u7565\uff0c\u4ece\u800c\u63d0\u9ad8\u63a8\u7406\u7684\u51c6\u786e\u6027\u548c\u6548\u7387\u3002<\/section>\n<section><\/section>\n<section><strong>\u901a\u8fc7\u8fd9\u79cd navigator \u5f15\u5bfc\u3001inference LLM \u6267\u884c\u3001\u6a21\u677f\u5e93\u652f\u6301\u3001\u52a8\u6001\u8c03\u6574\u8f68\u8ff9\u7684\u591a\u8f6e\u4ea4\u4e92\u673a\u5236\uff0cReasonFlux \u80fd\u591f\u9ad8\u6548\u5730\u89e3\u51b3\u5404\u79cd\u590d\u6742\u7684\u6570\u5b66\u63a8\u7406\u95ee\u9898\u3002\u8fd9\u79cd\u63a8\u7406\u6846\u67b6\u4e0d\u4ec5\u63d0\u9ad8\u4e86\u63a8\u7406\u7684\u51c6\u786e\u6027\u548c\u6548\u7387\uff0c\u8fd8\u589e\u5f3a\u4e86\u6a21\u578b\u7684\u53ef\u89e3\u91ca\u6027\uff0c\u56e0\u4e3a\u6211\u4eec\u53ef\u4ee5\u6e05\u6670\u5730\u8ffd\u8e2a\u6a21\u578b\u7684\u63a8\u7406\u8fc7\u7a0b\u548c\u4f9d\u636e\u3002<\/strong><\/section>\n<section><\/section>\n<section><strong>\u4e94\uff0e\u6570\u5b66\u63a8\u7406\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\uff1a<\/strong><\/section>\n<section><strong>\u5c0f\u6a21\u578b\u5ab2\u7f8e\u5927\u6a21\u578b\uff0c\u5c55\u73b0\u672a\u6765\u5e94\u7528\u6f5c\u529b<\/strong><\/section>\n<section><\/section>\n<section>ReasonFlux \u5728 MATH\u3001AIME 2024\u3001AMC 2023\u3001OlympiadBench \u548c Gaokao En 2023 \u7b49\u591a\u4e2a\u5177\u6709\u6311\u6218\u6027\u7684\u6570\u5b66\u63a8\u7406\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u4e86\u6d4b\u8bd5\uff0c\u5e76\u53d6\u5f97\u4e86\u826f\u597d\u7684\u7ed3\u679c\u3002<\/section>\n<section><\/section>\n<section>ReasonFlux-32B \u5728\u8fd9\u4e9b\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\u5904\u4e8e\u524d\u5217\uff0c\u4e0e\u5176\u4ed6\u5148\u8fdb\u6a21\u578b\u76f8\u6bd4\u5177\u6709\u7ade\u4e89\u529b\u3002\u5982\u4e0b\u8868\u6240\u793a\uff0c\u5728 MATH \u6570\u636e\u96c6\u4e0a\uff0cReasonFlux-32B \u7684\u51c6\u786e\u7387\u4e3a 91.2%\uff1b\u5728 AIME 2024 \u6570\u636e\u96c6\u4e0a\uff0cReasonFlux-32B \u7684\u51c6\u786e\u7387\u4e3a 56.7%\u3002\u8fd9\u4e9b\u7ed3\u679c\u8868\u660e ReasonFlux \u6846\u67b6\u5177\u6709\u6709\u6548\u6027\u3002\u66f4\u91cd\u8981\u7684\u662f\uff0c\u5b83\u8868\u660e\u8f83\u5c0f\u89c4\u6a21\u7684\u6a21\u578b\u901a\u8fc7\u4f18\u5316\u63a8\u7406\u6846\u67b6\uff0c\u53ef\u4ee5\u8fbe\u5230\u751a\u81f3\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\u8d85\u8d8a\u8f83\u5927\u6a21\u578b\u7684\u6027\u80fd\u3002<\/section>\n<section><\/section>\n<section>ReasonFlux \u8fd8\u53ef\u7528\u4e8e\u4e0d\u540c\u5927\u5c0f\uff081.5B, 7B \u548c 32B\uff09\u7684\u57fa\u7840\u6a21\u578b\uff0c\u5e76\u4e14\u90fd\u80fd\u83b7\u5f97\u5de8\u5e45\u7684\u63a8\u7406\u6548\u679c\u63d0\u5347\uff0c\u8db3\u89c1\u5176\u901a\u7528\u6027\u548c\u6cdb\u5316\u6027\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c4c459028334dcb773a449246cc1dc71.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c4c459028334dcb773a449246cc1dc71.png\"><\/a><\/section>\n<section>ReasonFlux \u7684\u6210\u529f\u4e0d\u4ec5\u9650\u4e8e\u6570\u5b66\u63a8\u7406\u9886\u57df\uff0c\u5176\u80cc\u540e\u7684\u6838\u5fc3\u601d\u60f3 \u2014\u2014 \u7ed3\u6784\u5316\u601d\u7ef4\u6a21\u677f\u548c\u6a21\u677f\u8f68\u8ff9 \u2014\u2014 \u5177\u6709\u5e7f\u6cdb\u7684\u5e94\u7528\u6f5c\u529b\u3002\u672a\u6765\uff0cReasonFlux \u6709\u6f5c\u529b\u88ab\u5e94\u7528\u4e8e\u66f4\u591a\u9886\u57df\uff0c\u5982\u4ee3\u7801\u751f\u6210\uff0c\u533b\u7597\u8bca\u65ad\uff0c\u5177\u8eab\u667a\u80fd\u7b49\u591a\u4e2a\u9886\u57df\u3002<\/section>\n<section><\/section>\n<section><strong>\u516d\uff0e\u4f5c\u8005\u4ecb\u7ecd<\/strong><\/section>\n<p>\u6768\u7075\uff1a\u5317\u5927\u5728\u8bfb\u535a\u58eb\uff0c\u666e\u6797\u65af\u987f\u9ad8\u7ea7\u7814\u7a76\u52a9\u7406\uff0c\u7814\u7a76\u9886\u57df\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u548c\u6269\u6563\u6a21\u578b\u3002<\/p>\n<p>\u4f59\u662d\u8fb0\uff1a\u65b0\u52a0\u5761\u56fd\u7acb\u5927\u5b66\u5728\u8bfb\u7855\u58eb\uff0c\u5317\u4eac\u5927\u5b66 PKU-DAIR \u5b9e\u9a8c\u5ba4\u79d1\u7814\u52a9\u7406\uff0c\u7814\u7a76\u9886\u57df\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u548c\u6269\u6563\u6a21\u578b\u3002<\/p>\n<p>\u5d14\u658c\u6559\u6388\uff1a\u5d14\u658c\u73b0\u4e3a\u5317\u4eac\u5927\u5b66\u8ba1\u7b97\u673a\u5b66\u9662\u535a\u96c5\u7279\u8058\u6559\u6388\u3001\u535a\u58eb\u751f\u5bfc\u5e08\uff0c\u62c5\u4efb\u8ba1\u7b97\u673a\u5b66\u9662\u526f\u9662\u957f\u3001\u6570\u636e\u79d1\u5b66\u4e0e\u5de5\u7a0b\u7814\u7a76\u6240\u6240\u957f\u3002\u4ed6\u7684\u7814\u7a76\u65b9\u5411\u5305\u62ec\u6570\u636e\u5e93\u7cfb\u7edf\u3001\u5927\u6570\u636e\u7ba1\u7406\u4e0e\u5206\u6790\u3001\u673a\u5668\u5b66\u4e60 \/ \u6df1\u5ea6\u5b66\u4e60\u7cfb\u7edf\u7b49\u3002<\/p>\n<p>\u738b\u68a6\u8fea\u6559\u6388\uff1a\u738b\u68a6\u8fea\u73b0\u4efb\u666e\u6797\u65af\u987f\u5927\u5b66\u7535\u5b50\u4e0e\u8ba1\u7b97\u673a\u5de5\u7a0b\u7cfb\u7ec8\u8eab\u6559\u6388\uff0c\u5e76\u521b\u7acb\u5e76\u62c5\u4efb\u666e\u6797\u65af\u987f\u5927\u5b66 \u201cAI for Accelerated Invention\u201d \u4e2d\u5fc3\u7684\u9996\u4efb\u4e3b\u4efb\u3002\u5979\u7684\u7814\u7a76\u9886\u57df\u6db5\u76d6\u5f3a\u5316\u5b66\u4e60\u3001\u53ef\u63a7\u5927\u6a21\u578b\u3001\u4f18\u5316\u5b66\u4e60\u7406\u8bba\u4ee5\u53ca AI for Science \u7b49\u591a\u4e2a\u65b9\u5411\u3002<\/p>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-12-4\" target=\"_blank\">8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:8\u536132B\u6a21\u578b\u8d85\u8d8ao1 [&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":[68,63,55,65],"class_list":["post-38271","post","type-post","status-publish","format-standard","hentry","category-news","tag-68","tag-63","tag-55","tag-65"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.9 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f - \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=38271\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f\",\"datePublished\":\"2025-02-12T20:00:57+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271\"},\"wordCount\":238,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png\",\"keywords\":[\"\u5927\u6a21\u578b\",\"\u6570\u5b66\u77e5\u8bc6\",\"\u673a\u5668\u5b66\u4e60\",\"\u6df1\u5ea6\u5b66\u4e60\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=38271#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=38271\",\"name\":\"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png\",\"datePublished\":\"2025-02-12T20:00:57+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=38271\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38271#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\",\"url\":\"https:\\\/\\\/17aitech.com\\\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/17aitech.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"url\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"contentUrl\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f - \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=38271","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=38271#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=38271"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f","datePublished":"2025-02-12T20:00:57+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=38271"},"wordCount":238,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=38271#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png","keywords":["\u5927\u6a21\u578b","\u6570\u5b66\u77e5\u8bc6","\u673a\u5668\u5b66\u4e60","\u6df1\u5ea6\u5b66\u4e60"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=38271#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=38271","url":"https:\/\/17aitech.com\/?p=38271","name":"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=38271#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=38271#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png","datePublished":"2025-02-12T20:00:57+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=38271#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=38271"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=38271#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ef74236aa230a0d4eb2096f271dbb5fc.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=38271#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"8\u536132B\u6a21\u578b\u8d85\u8d8ao1\u9884\u89c8\u7248\u3001DeepSeek V3\uff0c\u666e\u6797\u65af\u987f\u3001\u5317\u5927\u63d0\u51fa\u5c42\u6b21\u5316RL\u63a8\u7406\u65b0\u8303\u5f0f"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38271","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=38271"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38271\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38271"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38271"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38271"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}