{"id":4572,"date":"2025-11-28T15:53:37","date_gmt":"2025-11-28T07:53:37","guid":{"rendered":"https:\/\/www.ihub1.com\/?p=4572"},"modified":"2025-11-28T15:55:15","modified_gmt":"2025-11-28T07:55:15","slug":"deepseek%e5%86%8d%e7%a0%b4%e8%b0%b7%e6%ad%8copenai%e5%9e%84%e6%96%ad%ef%bc%9a%e5%bc%80%e6%ba%90imo%e6%95%b0%e5%ad%a6%e9%87%91%e7%89%8c%e5%a4%a7%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"https:\/\/www.ihub1.com\/index.php\/2025\/11\/28\/deepseek%e5%86%8d%e7%a0%b4%e8%b0%b7%e6%ad%8copenai%e5%9e%84%e6%96%ad%ef%bc%9a%e5%bc%80%e6%ba%90imo%e6%95%b0%e5%ad%a6%e9%87%91%e7%89%8c%e5%a4%a7%e6%a8%a1%e5%9e%8b\/","title":{"rendered":"DeepSeek\u518d\u7834\u8c37\u6b4cOpenAI\u5784\u65ad\uff1a\u5f00\u6e90IMO\u6570\u5b66\u91d1\u724c\u5927\u6a21\u578b"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-post\" data-elementor-id=\"4572\" class=\"elementor elementor-4572\">\n\t\t\t\t<div class=\"elementor-element elementor-element-cbfa331 e-flex e-con-boxed e-con e-parent\" data-id=\"cbfa331\" data-element_type=\"container\" data-e-type=\"container\">\n\t\t\t\t\t<div class=\"e-con-inner\">\n\t\t\t\t<div class=\"elementor-element elementor-element-95c0f21 elementor-widget elementor-widget-text-editor\" data-id=\"95c0f21\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\t<p>AI \u754c\u638c\u7ba1\u5f00\u6e90\u7684\u795e\u2014\u2014 DeepSeek \u56de\u6765\u4e86\uff01<\/p><p>\u521a\u521a\uff0cDeepSeek \u5f00\u6e90\u4e86\u5168\u65b0\u7684\u6570\u5b66\u6a21\u578b<strong>DeepSeekMath-V2<\/strong>\uff0c\u4e13\u6ce8\u4e8e\u53ef\u81ea\u9a8c\u8bc1\u7684\u6570\u5b66\u63a8\u7406\u3002<\/p><p>DeepSeekMath-V2 \u4e0d\u4ec5\u5728 IMO 2025 \u548c CMO 2024 \u4e2d\u53d6\u5f97\u91d1\u724c\u7ea7\u5206\u6570\uff0c\u800c\u4e14\u8fd8\u5728 Putnam 2024 \u4e2d\uff0c\u5f97\u5206 118\/120\uff0c\u8d85\u8fc7\u4e86\u4eba\u7c7b\u6700\u9ad8\u5206 90\u3002<\/p><p>\u4e0e\u6b64\u540c\u65f6\uff0cDeepSeekMath-V2 \u5728\u6240\u6709 CNML \u7ea7\u522b\u95ee\u9898\u7c7b\u522b\uff08\u4ee3\u6570\u3001\u51e0\u4f55\u3001\u6570\u8bba\u3001\u7ec4\u5408\u5b66\u3001\u4e0d\u7b49\u5f0f\uff09\u4e0a\u5747\u4f18\u4e8e GPT-5-Thinking-High \u548c Gemini 2.5-Pro\u3002<\/p><div id=\"id_imagebox_0\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_0\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d1_1024.jpg\" data-height=\"642\" data-width=\"1742\" \/><\/div><\/div><p>\u4e0d\u4ec5\u6027\u80fd\u65e0\u654c\uff0c\u7f51\u53cb\u8868\u793a\u8fd9\u8fd8\u662f\u7b2c\u4e00\u4e2a\u5f00\u6e90\u7684 IMO \u91d1\u724c\u6a21\u578b\u3002<\/p><div id=\"id_imagebox_1\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_1\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d2_1024.jpg\" data-height=\"258\" data-width=\"1190\" \/><\/div><\/div><p>\u8fd9\u4e0b\uff0c\u8c37\u6b4c\u548c OpenAI \u8981\u5750\u4e0d\u4f4f\u4e86\uff01<\/p><p>\u7279\u522b\u662f OpenAI\uff0c\u672c\u6765\u5c31\u6253\u7b97\u653e\u51fa IMO \u91d1\u724c\u6a21\u578b\u6765\u5e94\u5bf9\u8c37\u6b4c Gemini 3 Pro \u7684\u51b2\u51fb\uff0c\u73b0\u5728\u88ab DeepSeek \u62a2\u5148\u4e00\u6b65\u3002<\/p><p>\uff08\u9cb8\u9c7c\u56de\u6765\u4e86\uff01\uff09<\/p><p>\u503c\u5f97\u4e00\u63d0\u662f\uff0c\u8fd9\u7bc7\u8bba\u6587\u7684\u4e00\u4f5c<strong>\u90b5\u667a\u5b8f<\/strong>\u4e5f\u662f\u4e4b\u524d DeepSeekMath 7B \u7684\u4e00\u4f5c\uff0c\u5728\u90a3\u7bc7\u8bba\u6587\u4e2d\uff0c\u4ed6\u4eec\u63d0\u51fa\u4e86\u8457\u540d\u7684<strong>GRPO<\/strong>\u3002<\/p><p>\u6700\u5f3a\u5f00\u6e90 IMO \u91d1\u724c\u6a21\u578b<\/p><p>\u603b\u7684\u6765\u8bf4\uff0cDeepSeekMath-V2 \u662f\u4e00\u4e2a\u65e8\u5728\u5b9e\u73b0<strong>\u81ea\u9a8c\u8bc1\u6570\u5b66\u63a8\u7406<\/strong>\uff08Self-verification\uff09\u7684\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08685B\uff09\u3002<\/p><p>\u5b83\u7684\u6838\u5fc3\u5728\u4e8e\u5f00\u53d1\u548c\u5229\u7528\u5f3a\u5927\u7684<strong>\u8bc1\u660e\u9a8c\u8bc1<\/strong>\u80fd\u529b\u6765\u6307\u5bfc\u548c\u4f18\u5316<strong>\u8bc1\u660e\u751f\u6210<\/strong>\uff0c\u4ece\u800c\u514b\u670d\u4f20\u7edf\u4e0a\u4f9d\u8d56\u6700\u7ec8\u7b54\u6848\u4f5c\u4e3a\u5956\u52b1\u7684\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u65b9\u6cd5\u7684\u5c40\u9650\u6027\u3002<\/p><p>\u4f20\u7edf\u7528\u4e8e\u6570\u5b66\u63a8\u7406\u7684\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u65b9\u6cd5\u5b58\u5728\u6839\u672c\u6027\u9650\u5236\uff1a<\/p><p>\u6700\u7ec8\u7b54\u6848\u5956\u52b1\u7684\u4e0d\u53ef\u9760\u6027\uff1a\u5c06 LLM \u5956\u52b1\u57fa\u4e8e\u6700\u7ec8\u7b54\u6848\u7684\u6b63\u786e\u6027\uff0c\u5e76\u4e0d\u80fd\u4fdd\u8bc1\u63a8\u7406\u8fc7\u7a0b\u7684\u6b63\u786e\u6027\u6216\u903b\u8f91\u7684\u4e25\u8c28\u6027\uff0c\u6a21\u578b\u53ef\u80fd\u901a\u8fc7\u9519\u8bef\u7684\u903b\u8f91\u5f97\u51fa\u6b63\u786e\u7b54\u6848 \u3002<\/p><p>\u5bf9\u5b9a\u7406\u8bc1\u660e\u4efb\u52a1\u7684\u5c40\u9650\u6027\uff1a\u8bb8\u591a\u6570\u5b66\u4efb\u52a1\uff08\u5982\u5b9a\u7406\u8bc1\u660e\uff09\u4e0d\u8981\u6c42\u6570\u503c\u7b54\u6848\uff0c\u800c\u662f\u9700\u8981\u4e25\u683c\u7684\u6b65\u9aa4\u63a8\u5bfc\u548c\u903b\u8f91\u4e25\u8c28\u6027\uff0c\u4f7f\u5f97\u57fa\u4e8e\u6700\u7ec8\u7b54\u6848\u7684\u5956\u52b1\u673a\u5236\u4e0d\u9002\u7528\u3002<\/p><p>\u7f3a\u4e4f\u5185\u90e8\u9a8c\u8bc1\u80fd\u529b\uff1a\u7ecf\u8fc7\u4f20\u7edf\u65b9\u6cd5\u8bad\u7ec3\u7684 LLMs \u7f3a\u4e4f\u9a8c\u8bc1\u81ea\u8eab\u8bc1\u660e\u6709\u6548\u6027\u7684\u80fd\u529b\uff0c\u7ecf\u5e38\u8868\u73b0\u51fa\u9ad8\u5047\u9633\u6027\u7387\uff08\u5373\u8ba4\u4e3a\u9519\u8bef\u7684\u8bc1\u660e\u662f\u6709\u6548\u7684 ) \u3002<\/p><p>DeepSeekMath-V2 \u91c7\u7528<strong>\u8fed\u4ee3\u7684\u5f3a\u5316\u5b66\u4e60\u5faa\u73af<\/strong>\uff0c\u4ea4\u66ff\u4f18\u5316\u8bc1\u660e\u9a8c\u8bc1\u5668\u548c\u8bc1\u660e\u751f\u6210\u5668\uff0c\u4ee5\u5b9e\u73b0\u53ef\u81ea\u9a8c\u8bc1\u7684\u6570\u5b66\u63a8\u7406\u3002<\/p><p>\u8bc1\u660e\u9a8c\u8bc1<\/p><p><strong>\u8bad\u7ec3\u9a8c\u8bc1\u5668<\/strong><\/p><p>\u7814\u7a76\u9996\u5148\u8bad\u7ec3\u4e00\u4e2a\u51c6\u786e\u4e14\u5fe0\u5b9e\u7684 LLM-based \u9a8c\u8bc1\u5668\uff0c\u4f7f\u5176\u80fd\u591f\u6839\u636e\u4eba\u7c7b\u4e13\u5bb6\u7684\u6807\u51c6\u8bc6\u522b\u8bc1\u660e\u4e2d\u7684\u95ee\u9898\u5e76\u8bc4\u5206\u3002<\/p><p>\u5177\u4f53\u6765\u8bf4\uff0c\u9a8c\u8bc1\u5668\u9488\u5bf9\u7ed9\u5b9a\u7684\u6570\u5b66\u95ee\u9898\u4e0e\u8bc1\u660e\uff0c\u8f93\u51fa\u4e00\u4e2a\u8bc1\u660e\u5206\u6790\uff0c\u8be5\u5206\u6790\u9996\u5148\u603b\u7ed3\u8bc6\u522b\u51fa\u7684\u95ee\u9898\uff08\u5982\u679c\u6709\uff09\uff0c\u7136\u540e\u57fa\u4e8e\u4e09\u4e2a\u7ea7\u522b\u5206\u914d\u4e00\u4e2a\u5206\u6570\uff1a<\/p><p>1\uff1a\u5b8c\u5168\u6b63\u786e\u3001\u4e25\u8c28\u4e14\u903b\u8f91\u6e05\u6670\u7684\u8bc1\u660e\u30020.5\uff1a\u603b\u4f53\u903b\u8f91\u5408\u7406\uff0c\u4f46\u6709\u5fae\u5c0f\u9519\u8bef\u6216\u7ec6\u8282\u9057\u6f0f\u7684\u8bc1\u660e\u30020\uff1a\u5305\u542b\u81f4\u547d\u903b\u8f91\u9519\u8bef\u6216\u5173\u952e\u6f0f\u6d1e\u7684\u6839\u672c\u6027\u9519\u8bef\u7684\u8bc1\u660e\u3002<\/p><p>\u9a8c\u8bc1\u5668\u7684\u8bad\u7ec3\u5206\u4e3a\u6570\u636e\u6784\u5efa\uff08\u51b7\u542f\u52a8\uff09\u548c\u6a21\u578b\u5f3a\u5316\u5b66\u4e60\u76ee\u6807\u4e24\u4e2a\u5173\u952e\u9636\u6bb5\u3002<\/p><p>\u5728\u6570\u636e\u6784\u5efa\u9636\u6bb5\uff0c\u7814\u7a76\u9996\u5148\u4ece AoPS \u7ade\u8d5b\u4e2d\u6536\u96c6\u4e86 1.75 \u4e07\u4e2a\u8981\u6c42\u8bc1\u660e\u7684\u5965\u8d5b\u7ea7\u522b\u6570\u5b66\u95ee\u9898\u3002<\/p><p>\u968f\u540e\uff0c\u5229\u7528\u73b0\u6709\u6a21\u578b\uff08DeepSeek-V3.2-Exp-Thinking\uff09\u751f\u6210\u5927\u91cf\u7684\u5019\u9009\u8bc1\u660e\uff0c\u5e76\u901a\u8fc7\u591a\u8f6e\u8fed\u4ee3\u6765\u63d0\u5347\u8bc1\u660e\u7684\u4e25\u8c28\u6027\uff0c\u6700\u7ec8\u8bf7\u6570\u5b66\u4e13\u5bb6\u4eba\u5de5\u5bf9\u8fd9\u4e9b\u8bc1\u660e\u8fdb\u884c\u8bc4\u5206\uff0c\u5206\u6570\u5206\u4e3a 1\u30010.5 \u548c 0 \u4e09\u4e2a\u7ea7\u522b\uff0c\u4ece\u800c\u521b\u5efa\u4e86\u521d\u59cb\u7684 RL \u9a8c\u8bc1\u6570\u636e\u96c6\u3002<\/p><p>\u8fdb\u5165\u5f3a\u5316\u5b66\u4e60\u76ee\u6807\u9636\u6bb5\uff0c\u7814\u7a76\u4f7f\u7528\u4e0a\u8ff0\u6570\u636e\u96c6\u5bf9\u57fa\u7840\u6a21\u578b\uff08DeepSeek-V3.2-Exp-SFT\uff09\u8fdb\u884c\u8bad\u7ec3\uff0c\u4f7f\u5176\u80fd\u591f\u8f93\u51fa\u8bc1\u660e\u5206\u6790\u603b\u7ed3\u548c\u6700\u7ec8\u5206\u6570\u3002<\/p><p>\u5956\u52b1\u51fd\u6570\u4e00\u65b9\u9762\u901a\u8fc7\u683c\u5f0f\u5956\u52b1\u5f3a\u5236\u6a21\u578b\u8f93\u51fa\u683c\u5f0f\u5305\u542b\u95ee\u9898\u603b\u7ed3\u548c\u5206\u6570\uff0c\u53e6\u4e00\u65b9\u9762\u901a\u8fc7\u5206\u6570\u5956\u52b1\u6fc0\u52b1\u6a21\u578b\u9884\u6d4b\u7684\u5206\u6570\u4e0e\u4e13\u5bb6\u6807\u6ce8\u7684\u5206\u6570\u9ad8\u5ea6\u4e00\u81f4\uff0c\u4ece\u800c\u4f7f\u9a8c\u8bc1\u5668\u5177\u5907\u6a21\u4eff\u4eba\u7c7b\u4e13\u5bb6\u8bc4\u4f30\u4e25\u8c28\u6027\u7684\u80fd\u529b\u3002<\/p><div id=\"id_imagebox_2\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_2\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d3_1024.jpg\" data-height=\"112\" data-width=\"894\" \/><\/div><\/div><p><strong>\u5f15\u5165\u5143\u9a8c\u8bc1 ( Meta-Verification )<\/strong><\/p><p>\u4e3a\u89e3\u51b3\u521d\u6b65\u8bad\u7ec3\u7684\u9a8c\u8bc1\u5668\u53ef\u80fd\u56e0\u5e7b\u89c9\uff08hallucinating\uff09\u4e0d\u5b58\u5728\u7684\u95ee\u9898\u800c\u83b7\u5f97\u6b63\u786e\u4f4e\u5206\uff0c\u4ece\u800c\u635f\u5bb3\u5176\u5bf9\u9519\u8bef\u8bc6\u522b\u7684\u5fe0\u5b9e\u6027\uff08faithfulness\uff09\u7684\u95ee\u9898\uff0c\u7814\u7a76\u56e2\u961f\u5f15\u5165\u4e86\u5143\u9a8c\u8bc1\uff08Meta-Verification\uff09\u673a\u5236\u3002<\/p><p>\u5143\u9a8c\u8bc1\u4f5c\u4e3a\u4e00\u4e2a\u4e8c\u7ea7\u8bc4\u4f30\u8fc7\u7a0b\uff0c\u65e8\u5728\u5ba1\u67e5\u9a8c\u8bc1\u5668\u751f\u6210\u7684\u8bc1\u660e\u5206\u6790\uff08Proof Analysis\uff09\uff0c\u8bc4\u4f30\u5176\u4e2d\u8bc6\u522b\u51fa\u7684\u95ee\u9898\u662f\u5426\u771f\u5b9e\u5b58\u5728\uff0c\u4ee5\u53ca\u8fd9\u4e9b\u95ee\u9898\u662f\u5426\u5728\u903b\u8f91\u4e0a\u5408\u7406\u5730\u652f\u6301\u4e86\u5176\u9884\u6d4b\u7684\u8bc1\u660e\u5206\u6570\u3002<\/p><p>\u4e3a\u4e86\u8bad\u7ec3\u5143\u9a8c\u8bc1\u5668\uff0c\u7814\u7a76\u9996\u5148\u8ba9\u6570\u5b66\u4e13\u5bb6\u6839\u636e\u4e13\u95e8\u7684\u5143\u9a8c\u8bc1\u6807\u51c6\u5bf9\u9a8c\u8bc1\u5668\u8f93\u51fa\u7684\u5206\u6790\u8d28\u91cf\u8fdb\u884c\u8bc4\u5206\uff0c\u521b\u5efa\u4e86\u5143\u9a8c\u8bc1\u6570\u636e\u96c6\u3002<\/p><p>\u968f\u540e\uff0c\u7814\u7a76\u8bad\u7ec3\u4e86\u4e00\u4e2a\u4e13\u95e8\u7684\u5143\u9a8c\u8bc1\u5668\uff0c\u8be5\u5143\u9a8c\u8bc1\u5668\u751f\u6210\u5bf9\u9a8c\u8bc1\u5206\u6790\u672c\u8eab\u7684\u95ee\u9898\u603b\u7ed3\uff0c\u5e76\u5206\u914d\u4e00\u4e2a\u8d28\u91cf\u5206\u6570\uff0c\u4ee5\u8861\u91cf\u539f\u9a8c\u8bc1\u5668\u5206\u6790\u7684\u51c6\u786e\u6027\u548c\u5408\u7406\u6027\u3002<\/p><p>\u5143\u9a8c\u8bc1\u5668\u7684\u5f3a\u5316\u5b66\u4e60\u76ee\u6807\u7ed3\u6784\u4e0e\u9a8c\u8bc1\u5668\u8bad\u7ec3\u7c7b\u4f3c\uff0c\u540c\u6837\u5305\u542b\u683c\u5f0f\u5956\u52b1\u548c\u5206\u6570\u5956\u52b1\u3002<\/p><p>\u63a5\u4e0b\u6765\u662f\u589e\u5f3a\u9a8c\u8bc1\u5668\u8bad\u7ec3\uff0c\u7814\u7a76\u5229\u7528\u8bad\u7ec3\u597d\u7684\u5143\u9a8c\u8bc1\u5668\uff0c\u5c06\u5143\u9a8c\u8bc1\u7684\u8d28\u91cf\u5206\u6570\u96c6\u6210\u5230\u9a8c\u8bc1\u5668\u7684\u5956\u52b1\u51fd\u6570\u4e2d\uff0c\u4ee5\u589e\u5f3a\u9a8c\u8bc1\u5668\u7684\u5fe0\u5b9e\u6027\u3002<\/p><p>\u6700\u7ec8\uff0c\u4f7f\u7528\u539f\u9a8c\u8bc1\u6570\u636e\u96c6\u548c\u5143\u9a8c\u8bc1\u6570\u636e\u96c6\u5171\u540c\u8bad\u7ec3\u589e\u5f3a\u540e\u7684\u9a8c\u8bc1\u5668\uff0c\u4f7f\u8be5\u6a21\u578b\u80fd\u591f\u540c\u65f6\u6267\u884c\u8bc1\u660e\u9a8c\u8bc1\u548c\u5143\u9a8c\u8bc1\u4e24\u9879\u4efb\u52a1\u3002<\/p><p>\u5728\u539f\u9a8c\u8bc1\u6570\u636e\u96c6\u7684\u4e00\u4e2a\u9a8c\u8bc1\u5b50\u96c6\u4e0a\uff0c\u7ecf\u9a8c\u8bc1\u5668\u8bc4\u4f30\u7684\u8bc1\u660e\u5206\u6790\u7684\u5e73\u5747\u8d28\u91cf\u5206\u6570\u4ece 0.85 \u63d0\u5347\u5230\u4e86<strong>0.96<\/strong>\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u8bc1\u660e\u5206\u6570\u9884\u6d4b\u7684\u51c6\u786e\u6027\u4e0d\u53d8\uff0c\u6709\u529b\u8bc1\u660e\u4e86\u5143\u9a8c\u8bc1\u673a\u5236\u80fd\u6709\u6548\u63d0\u9ad8\u9a8c\u8bc1\u5668\u8bc6\u522b\u95ee\u9898\u7684\u5fe0\u5b9e\u5ea6\u3002<\/p><p>\u8bc1\u660e\u751f\u6210<\/p><p>\u63a5\u4e0b\u6765\uff0c\u7814\u7a76\u7528\u8bad\u7ec3\u597d\u7684\u9a8c\u8bc1\u5668\u4f5c\u4e3a\u5956\u52b1\u6a21\u578b\u6765\u8bad\u7ec3\u8bc1\u660e\u751f\u6210\u5668\uff0c\u5e76\u8fdb\u4e00\u6b65\u901a\u8fc7 &#8221; \u81ea\u9a8c\u8bc1 &#8221; \u673a\u5236\uff0c\u8ba9\u751f\u6210\u5668\u5b66\u4f1a\u4e25\u683c\u5730\u81ea\u6211\u5ba1\u67e5\u548c\u4fee\u6b63\uff0c\u4ece\u800c\u63d0\u9ad8\u8bc1\u660e\u8d28\u91cf\u3002<\/p><p>\u5177\u4f53\u6765\u8bf4\uff0c\u7814\u7a76\u8bad\u7ec3\u751f\u6210\u5668\u4ee5\u6700\u5927\u5316\u9a8c\u8bc1\u5668\u8d4b\u7ed9\u5176\u751f\u6210\u7684\u8bc1\u660e\u7684\u5206\u6570\u3002<\/p><p>\u5728\u8bad\u7ec3\u4e2d\uff0c\u751f\u6210\u5668\u88ab\u8981\u6c42\u5728\u751f\u6210\u8bc1\u660e\u540e\uff0c\u7d27\u8ddf\u7740\u8fdb\u884c\u81ea\u6211\u5206\u6790\u3002\u5956\u52b1\u51fd\u6570\u6fc0\u52b1\u51c6\u786e\u7684\u81ea\u6211\u8bc4\u4f30\u548c\u6b63\u786e\u6027\u3002<\/p><p>\u6700\u7ec8\u5956\u52b1\u662f\u5bf9\u8bc1\u660e\u8d28\u91cf\u548c\u81ea\u6211\u8bc4\u4f30\u8d28\u91cf\u7684\u52a0\u6743\u7ec4\u5408\u3002<\/p><p>\u7531\u6b64\uff0c\u81ea\u6211\u8bc4\u4f30\u5956\u52b1\u4e0d\u4ec5\u5956\u52b1\u81ea\u8bc4\u5206\u7684\u51c6\u786e\u6027\uff0c\u8fd8\u5956\u52b1\u81ea\u6211\u5206\u6790\u7684\u5fe0\u5b9e\u6027\u3002<\/p><p>\u8fd9\u79cd\u5956\u52b1\u7ed3\u6784\u6fc0\u52b1\u751f\u6210\u5668\uff1a\u5fe0\u5b9e\u5730\u627f\u8ba4\u9519\u8bef\uff08\u800c\u975e\u76f2\u76ee\u81ea\u4fe1\uff09\u3002\u83b7\u5f97\u9ad8\u5956\u52b1\u7684\u6700\u4f73\u7b56\u7565\u662f\u5728\u6700\u7ec8\u8f93\u51fa\u524d\uff0c\u79ef\u6781\u8bc6\u522b\u5e76\u89e3\u51b3\u81ea\u8eab\u8bc1\u660e\u4e2d\u5b58\u5728\u7684\u95ee\u9898\uff0c\u4ece\u800c\u5b9e\u73b0\u81ea\u6211\u8fed\u4ee3\u5b8c\u5584\u3002<\/p><p>\u8bc1\u660e\u9a8c\u8bc1\u4e0e\u751f\u6210\u4e4b\u95f4\u7684\u534f\u540c\u4f5c\u7528<\/p><p>\u63a5\u4e0b\u6765\uff0c\u7814\u7a76\u5229\u7528\u9a8c\u8bc1\u5668\u548c\u751f\u6210\u5668\u7684\u534f\u540c\u4f5c\u7528\uff0c\u901a\u8fc7\u89c4\u6a21\u5316\u7684\u8ba1\u7b97\u548c\u5143\u9a8c\u8bc1\u673a\u5236\uff0c\u5efa\u7acb\u4e86\u4e00\u4e2a\u5b8c\u5168\u81ea\u52a8\u5316\u7684\u6570\u636e\u6807\u6ce8\u6d41\u7a0b\uff0c\u4ece\u800c\u6301\u7eed\u63d0\u5347\u9a8c\u8bc1\u5668\u7684\u80fd\u529b\uff0c\u5e76\u6700\u7ec8\u53d6\u4ee3\u8017\u65f6\u7684\u4eba\u5de5\u6807\u6ce8\u3002<\/p><p>\u7136\u800c\uff0c\u968f\u7740\u95ee\u9898\u96be\u5ea6\u589e\u52a0\uff0c\u4eba\u5de5\u6807\u6ce8\u8017\u65f6\u4e14\u6548\u7387\u4f4e\u4e0b\u3002\u7531\u6b64\u7814\u7a76\u63d0\u51fa\u4e86\u4e00\u5957\u81ea\u52a8\u5316\u6807\u6ce8\u7684\u65b9\u6cd5\uff1a<\/p><p>\u9996\u5148\uff0c\u901a\u8fc7\u751f\u6210 n \u4e2a\u72ec\u7acb\u7684\u9a8c\u8bc1\u5206\u6790\uff0c\u63d0\u9ad8\u5728\u6709\u7f3a\u9677\u8bc1\u660e\u4e2d\u6355\u83b7\u771f\u5b9e\u95ee\u9898\u7684\u6982\u7387\u3002\u63a5\u4e0b\u6765\uff0c\u5229\u7528\u5143\u9a8c\u8bc1\u5668\u751f\u6210 m \u4e2a\u8bc4\u4f30\uff0c\u5bf9\u62a5\u544a\u95ee\u9898\u7684\u5206\u6790\u8fdb\u884c\u6709\u6548\u6027\u786e\u8ba4\uff0c\u786e\u4fdd\u8bc6\u522b\u51fa\u7684\u95ee\u9898\u662f\u771f\u5b9e\u7684\uff08\u5143\u9a8c\u8bc1\u6bd4\u4ece\u96f6\u8bc6\u522b\u95ee\u9898\u66f4\u9ad8\u6548\uff09\u3002<\/p><p>\u5177\u4f53\u7684\u6807\u6ce8\u6d41\u7a0b\u5982\u4e0b\uff1a<\/p><p>\u5206\u6570\u5224\u5b9a\uff1a\u68c0\u67e5\u6240\u6709\u5206\u6790\u4e2d\u5206\u914d\u7684\u6700\u4f4e\u5206\u6570\u3002\u53ea\u6709\u5f53\u81f3\u5c11\u6709 k \u4e2a\u5206\u6790\u88ab\u5143\u9a8c\u8bc1\u786e\u8ba4\u4e3a\u6709\u6548\u65f6\uff0c\u8be5\u6700\u4f4e\u5206\u624d\u88ab\u8d4b\u7ed9\u8be5\u8bc1\u660e\uff1b\u5426\u5219\u6807\u8bb0\u4e3a 1 \u5206\uff08\u65e0\u5408\u6cd5\u7f3a\u9677\uff09\u3002<\/p><p>\u53d6\u4ee3\u4eba\u5de5\uff1a \u6700\u7ec8\uff0c\u8fd9\u79cd\u5b8c\u5168\u81ea\u52a8\u5316\u7684\u6d41\u7a0b\u5728\u540e\u7eed\u8bad\u7ec3\u8fed\u4ee3\u4e2d\u5f7b\u5e95\u53d6\u4ee3\u4e86\u4eba\u5de5\u6807\u6ce8\uff0c\u5e76\u4e14\u8d28\u91cf\u68c0\u67e5\u8bc1\u5b9e\u5176\u6807\u6ce8\u7ed3\u679c\u4e0e\u4e13\u5bb6\u5224\u65ad\u9ad8\u5ea6\u4e00\u81f4\u3002<\/p><p>\u7531\u6b64\uff0c\u4e0a\u8ff0\u6d41\u7a0b\u5728\u6700\u7ec8\u7684\u8bad\u7ec3\u8fed\u4ee3\u4e2d\u5f7b\u5e95\u53d6\u4ee3\u4e86\u8017\u65f6\u7684\u4eba\u529b\u6807\u6ce8\uff0c\u5b9e\u73b0\u4e86\u9a8c\u8bc1\u548c\u751f\u6210\u7684\u534f\u540c\u5faa\u73af\uff0c\u4fdd\u8bc1\u4e86\u6a21\u578b\u80fd\u529b\u7684\u6301\u7eed\u7a81\u7834\u3002<\/p><p>\u5b9e\u9a8c\u7ed3\u679c<\/p><p>\u7814\u7a76\u91c7\u7528 GRPO \u8fdb\u884c\u5f3a\u5316\u5b66\u4e60\uff0c\u8fed\u4ee3\u5730\u4f18\u5316\u8bc1\u660e\u9a8c\u8bc1\u548c\u751f\u6210\u80fd\u529b\u3002<\/p><p>\u5728\u6bcf\u6b21\u8fed\u4ee3\u4e2d\uff0c\u7814\u7a76\u9996\u5148\u4f18\u5316\u8bc1\u660e\u9a8c\u8bc1\u3002\u7136\u540e\uff0c\u8bc1\u660e\u751f\u6210\u5668\u4f1a\u4ece\u9a8c\u8bc1\u5668\u7684 checkpoint \u521d\u59cb\u5316\uff0c\u5e76\u9488\u5bf9\u8bc1\u660e\u751f\u6210\u8fdb\u884c\u4f18\u5316\u3002<\/p><p>\u4ece\u7b2c\u4e8c\u6b21\u8fed\u4ee3\u5f00\u59cb\uff0c\u8bc1\u660e\u9a8c\u8bc1\u5668\u4f1a\u4f7f\u7528\u4e00\u4e2a\u68c0\u67e5\u70b9\u8fdb\u884c\u521d\u59cb\u5316\uff0c\u8be5\u68c0\u67e5\u70b9\u901a\u8fc7<strong>\u62d2\u7edd\u5fae\u8c03\uff08rejection fine-tuning\uff09<\/strong>\u5de9\u56fa\u4e86\u524d\u4e00\u6b21\u8fed\u4ee3\u4e2d\u7684\u9a8c\u8bc1\u548c\u751f\u6210\u80fd\u529b\u3002<\/p><p>\u7814\u7a76\u9996\u5148\u8bc4\u4f30\u4e86\u6a21\u578b\u672a\u7ecf\u8fed\u4ee3\u5b8c\u5584\u7684\u5355\u6b21\u751f\u6210\u6b63\u786e\u8bc1\u660e\u7684\u80fd\u529b\u3002<\/p><p>\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5728\u6240\u6709 CNML \u7ea7\u522b\u95ee\u9898\u7c7b\u522b\u4e2d\u2014\u2014\u5305\u62ec\u4ee3\u6570\u3001\u51e0\u4f55\u3001\u6570\u8bba\u3001\u7ec4\u5408\u5b66\u548c\u4e0d\u7b49\u5f0f\u2014\u2014 DeepSeekMath-V2 \u6301\u7eed\u4f18\u4e8e GPT-5-Thinking-High \u548c Gemini 2.5-Pro\u3002<\/p><div id=\"id_imagebox_3\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_3\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d5_1024.jpg\" data-height=\"384\" data-width=\"1058\" \/><\/div><\/div><p>\u4e3a\u4e86\u63a2\u7a76\u6269\u5c55\u4e0a\u4e0b\u6587\u548c\u81ea\u9a8c\u8bc1\u5982\u4f55\u63d0\u9ad8\u8bc1\u660e\u8d28\u91cf\uff0c\u7814\u7a76\u53c8\u8bc4\u4f30\u4e86\u5e26\u6709\u81ea\u9a8c\u8bc1\u7684\u987a\u5e8f\u7cbe\u70bc\u65b9\u6cd5\u3002<\/p><div id=\"id_imagebox_4\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_4\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d6_1024.jpg\" data-height=\"364\" data-width=\"544\" \/><\/div><\/div><p>\u7814\u7a76\u8868\u660e\uff0c\u81ea\u9009\u7684\u6700\u4f73\u8bc1\u660e\u6bd4\u7ebf\u7a0b\u5e73\u5747\u5f97\u5206\u83b7\u5f97\u4e86\u663e\u8457\u66f4\u9ad8\u7684\u9a8c\u8bc1\u5206\u6570\uff0c\u8fd9\u8bc1\u660e\u751f\u6210\u5668\u5177\u5907\u51c6\u786e\u8bc4\u4f30\u8bc1\u660e\u8d28\u91cf\u7684\u80fd\u529b\u3002<\/p><p>\u6b64\u5916\uff0c\u968f\u7740\u6700\u5927\u987a\u5e8f\u5c1d\u8bd5\u6b21\u6570\u7684\u589e\u52a0\uff0cPass@1 \u5927\u5e45\u63d0\u5347\uff0c\u8868\u660e\u81ea\u9a8c\u8bc1\u6709\u6548\u5730\u6307\u5bfc\u4e86\u8fed\u4ee3\u6539\u8fdb\u3002<\/p><p>\u8fd9\u4e9b\u7ed3\u679c\u8868\u660e\uff0c\u751f\u6210\u5668\u80fd\u591f\u53ef\u9760\u5730\u533a\u5206\u9ad8\u8d28\u91cf\u548c\u6709\u7f3a\u9677\u7684\u8bc1\u660e\uff0c\u5e76\u5229\u7528\u8fd9\u79cd\u81ea\u6211\u610f\u8bc6\u7cfb\u7edf\u5730\u6539\u8fdb\u5176\u6570\u5b66\u63a8\u7406\u3002<\/p><p>\u6700\u540e\uff0c\u4e3a\u4e86\u89e3\u51b3\u6700\u5177\u5907\u6311\u6218\u6027\u7684\u95ee\u9898\uff0c\u7814\u7a76\u91c7\u7528\u4e86\u9ad8\u8ba1\u7b97\u91cf\u641c\u7d22\u7b56\u7565\uff0c\u8be5\u7b56\u7565\u901a\u8fc7\u5e76\u884c\u751f\u6210\u63a2\u7d22\u591a\u6837\u5316\u7684\u8bc1\u660e\u8def\u5f84\uff0c\u5e76\u7ed3\u5408\u89c4\u6a21\u5316\u7684\uff0864 \u6b21\uff09\u9a8c\u8bc1\u6765\u7cbe\u786e\u8bc6\u522b\u7ec6\u5fae\u9519\u8bef\u3002<\/p><p>\u6a21\u578b\u8fed\u4ee3\u5730\u4ece\u5305\u542b\u95ee\u9898\u5206\u6790\u7684\u5019\u9009\u8bc1\u660e\u6c60\u4e2d\u7cbe\u70bc\u51fa\u6700\u4f18\u8bc1\u660e\uff0c\u76f4\u5230\u8bc1\u660e\u901a\u8fc7\u6240\u6709\u9a8c\u8bc1\u3002<\/p><p>\u6700\u7ec8\uff0c\u8fd9\u79cd\u65b9\u6cd5\u4f7f\u5176\u5728 Putnam \u7ade\u8d5b\u4e2d\u4ee5 118\/120 \u7684\u5206\u6570\u8d85\u8d8a\u4eba\u7c7b\u6700\u9ad8\u5206 90 \u5206\uff0c\u5c55\u73b0\u4e86\u5728\u9a8c\u8bc1\u5668\u6307\u5bfc\u4e0b\uff0cAI \u89e3\u51b3\u590d\u6742\u95ee\u9898\u7684\u5f3a\u5927\u80fd\u529b\u3002<\/p><div id=\"id_imagebox_5\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_5\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d7_1024.jpg\" data-height=\"234\" data-width=\"564\" \/><\/div><\/div><p>One more thing<\/p><p>\u5982\u5f00\u5934\u6240\u8bf4\uff0c\u8fd9\u7bc7\u8bba\u6587\u7684\u4e00\u4f5c<strong>\u90b5\u667a\u5b8f<\/strong>\u4e5f\u662f DeepSeek \u4e4b\u524d\u6570\u5b66\u6a21\u578b DeepSeekMath 7B \u7684\u4e00\u4f5c\u3002<\/p><div id=\"id_imagebox_6\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_6\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d8_1024.jpg\" data-height=\"200\" data-width=\"200\" \/><\/div><\/div><p>\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u4e5f\u5c31\u662f\u5728 DeepSeekMath 7B \u7684\u8bba\u6587\u4e2d\uff0c\u4ed6\u548c\u56e2\u961f\u63d0\u51fa\u4e86\u7ecf\u5178\u7684<strong>GRPO<\/strong>\uff08Group Relative Policy Optimization \uff09\u3002\u540c\u6837\u7684\uff0c\u4ed6\u4e5f\u662f DeepSeek-R1 \u7684\u6838\u5fc3\u8d21\u732e\u8005\u3002<\/p><div id=\"id_imagebox_7\" class=\"img_box \"><div class=\"content_img_div perview_img_div\"><img decoding=\"async\" id=\"img_7\" class=\"opacity_0\" src=\"https:\/\/zkres1.myzaker.com\/202511\/6929011fb15ec039c94526d9_1024.jpg\" data-height=\"538\" data-width=\"1602\" \/><\/div><\/div><p>\u90b5\u667a\u5b8f\u76ee\u524d\u662f DeepSeek \u4ece\u4e8b\u5927\u6a21\u578b\u63a8\u7406\u7814\u7a76\u7684\u7814\u7a76\u5458\uff0c\u4ed6\u672c\u79d1\u6bd5\u4e1a\u4e8e\u5317\u4eac\u822a\u7a7a\u822a\u5929\u5927\u5b66\uff0c\u535a\u58eb\u6bd5\u4e1a\u4e8e\u6e05\u534e\uff0c\u5e08\u4ece\u9ec4\u6c11\u70c8\u6559\u6388\u3002<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>henry \u53d1\u81ea \u51f9\u975e\u5bfa<\/p>\n","protected":false},"author":1,"featured_media":4574,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-4572","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/posts\/4572","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/comments?post=4572"}],"version-history":[{"count":7,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/posts\/4572\/revisions"}],"predecessor-version":[{"id":4580,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/posts\/4572\/revisions\/4580"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/media\/4574"}],"wp:attachment":[{"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/media?parent=4572"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/categories?post=4572"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ihub1.com\/index.php\/wp-json\/wp\/v2\/tags?post=4572"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}