{"id":38235,"date":"2025-02-12T10:00:55","date_gmt":"2025-02-12T02:00:55","guid":{"rendered":"https:\/\/17aitech.com\/?p=38235"},"modified":"2025-02-12T10:00:55","modified_gmt":"2025-02-12T02:00:55","slug":"%e5%bc%80%e6%ba%9022%e4%b8%87%e6%9d%a1deepseek-r1%e7%9a%84%e9%ab%98%e8%b4%a8%e9%87%8f%e6%95%b0%e6%8d%ae%ef%bc%81%e4%bd%a0%e4%b9%9f%e8%83%bd%e5%a4%8d%e7%8e%b0deepseek%e4%ba%86","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=38235","title":{"rendered":"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-11-5\" target=\"_blank\">\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86<\/a><\/p>\n<section>DeepSeek \u5728\u6d77\u5185\u5916\u6405\u8d77\u7684\u60ca\u6d9b\u5de8\u6d6a\uff0c\u4f59\u6ce2\u4ecd\u5728\u6c79\u6d8c\u3002<\/section>\n<section><\/section>\n<section>\u5f53\u4e2d\u56fd\u5927\u6a21\u578b\u6495\u5f00\u7845\u8c37\u7684\u9632\u7ebf\u4e4b\u540e\uff0c\u5728\u9884\u8bbe\u4e2d\u603b\u662f\u843d\u540e\u534a\u62cd\u7684\u4e2d\u56fd AI \u519b\u56e2\uff0c\u8fd9\u6b21\u7adf\u5b8c\u6210\u4e86\u4e00\u6b21\u53cd\u5411\u6280\u672f\u8f93\u51fa\uff0c\u5f15\u53d1\u4e86\u5168\u7403\u8303\u56f4\u5185\u590d\u73b0 DeepSeek \u7684\u70ed\u6f6e\u3002<\/section>\n<section><\/section>\n<section>DeepSeek-R1 \u867d\u7136\u5f00\u6e90\uff0c\u4f46\u4e5f\u6ca1\u6709\u5b8c\u5168\u5f00\u6e90\uff0c\u8bad\u7ec3\u6570\u636e\u3001\u8bad\u7ec3\u811a\u672c\u7b49\u5173\u952e\u4fe1\u606f\u5e76\u672a\u5b8c\u5168\u516c\u5e03\u3002<\/section>\n<section><\/section>\n<section>\u4e0d\u8fc7\uff0c\u6709\u6280\u672f\u62a5\u544a\uff0c\u76f8\u5f53\u4e8e\u62e5\u6709\u7740\u590d\u73b0 R1 \u7684\u6307\u5bfc\u65b9\u9488\uff0c\u5df2\u7ecf\u6709\u4e0d\u5c11\u6267\u884c\u529b\u5f3a\u7684\u56e2\u961f\u7528\u5c0f\u6a21\u578b\u89c1\u8bc1\u300c<a data-itemshowtype=\"0\" data-linktype=\"2\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzA3MzI4MjgzMw==&amp;mid=2650953087&amp;idx=1&amp;sn=f8b35941ec2625c30870a16bf90386e6&amp;scene=21#wechat_redirect\" target=\"_blank\">aha moment<\/a>\u300d\u4e86\u3002\u5728\u6d69\u6d69\u8361\u8361\u7684\u590d\u523b\u5927\u519b\u4e2d\uff0c\u6700\u4ee4\u4eba\u77a9\u76ee\u7684\uff0c\u5f53\u6570 Hugging Face \u9886\u8854\u7684 Open R1 \u9879\u76ee\u3002<\/section>\n<section><\/section>\n<section>Open R1 \u5ba3\u79f0\u8981\u505a\u5230\u5b8c\u5168\u5f00\u653e\u590d\u73b0 DeepSeek-R1\uff0c\u8865\u9f50 DeepSeek \u6240\u6709\u672a\u516c\u5f00\u7684\u6280\u672f\u7ec6\u8282\u3002Open R1 \u9879\u76ee\u521a\u521a\u542f\u52a8\u4e0d\u8fc7\u51e0\u5468\uff0c\u4ed6\u4eec\u5df2\u7ecf\u5b8c\u6210\u4e86\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>GRPO \u5b9e\u73b0<\/section>\n<\/li>\n<li>\n<section>\u8bad\u7ec3\u4e0e\u8bc4\u4f30\u4ee3\u7801<\/section>\n<\/li>\n<li>\n<section>\u7528\u4e8e\u5408\u6210\u6570\u636e\u7684\u751f\u6210\u5668<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png\"><\/a><\/section>\n<section>\u9879\u76ee\u5730\u5740\uff1ahttps:\/\/github.com\/huggingface\/open-r1<\/section>\n<section><\/section>\n<section>\u5728\u5f00\u6e90\u793e\u533a\u7684\u4f17\u5fd7\u6210\u57ce\uff0cOpen R1 \u66f4\u662f\u52a8\u4f5c\u8fc5\u901f\u3002\u4eca\u5929\uff0c\u4ed6\u4eec\u53d1\u5e03\u4e86 OpenR1-Math-220k \u6570\u636e\u96c6\uff0c\u53c8\u8865\u5168\u4e86\u4e00\u5757 DeepSeek R1\u300c\u788e\u7247\u300d\u2014\u2014 \u5408\u6210\u6570\u636e\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7c1df4bd2000edf480c0058a15078989.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7c1df4bd2000edf480c0058a15078989.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 OpenR1-Math-220k \u6570\u636e\u96c6\u6982\u89c8<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u6570\u636e\u96c6\u94fe\u63a5\uff1ahttps:\/\/huggingface.co\/datasets\/open-r1\/OpenR1-Math-220k<\/section>\n<section><\/section>\n<section>DeepSeek R1 \u7684\u4e00\u4e2a\u91cd\u8981\u4f18\u52bf\u5728\u4e8e\u5b83\u80fd\u591f\u5c06\u9ad8\u7ea7\u63a8\u7406\u80fd\u529b\u8fc1\u79fb\u5230\u8f83\u5c0f\u7684\u6a21\u578b\u4e2d\u3002DeepSeek \u56e2\u961f\u751f\u6210\u4e86 60 \u4e07\u6761\u63a8\u7406\u6570\u636e\uff0c\u5728 Qwen \u548c Llama \u7b49\u5f00\u6e90\u6a21\u578b\u4e0a\u8bc1\u660e\u4e86\u8fd9\u79cd\u8fc1\u79fb\u80fd\u529b\u3002\u5373\u4f7f\u4e0d\u4f7f\u7528<mark data-type=\"tech_methods\" data-id=\"ee1a8f69-3170-4ddf-b2b6-47d91c844425\">\u5f3a\u5316\u5b66\u4e60<\/mark>\uff0c\u76f4\u63a5\u4ece R1 \u6a21\u578b\u8fdb\u884c\u8fc1\u79fb\u4e5f\u80fd\u5b9e\u73b0\u5f3a\u5927\u7684\u63a8\u7406\u6027\u80fd\u3002<\/section>\n<section><\/section>\n<section>\u7136\u800c\uff0c\u8fd9\u4e9b\u5408\u6210\u6570\u636e\u4ec5 DeepSeek \u53ef\u89c1\uff0c\u672a\u5bf9\u5176\u4ed6\u56e2\u961f\u5f00\u653e\u9605\u8bfb\u6743\u9650\u3002<\/section>\n<section><\/section>\n<section>OpenR1-Math-220k \u6570\u636e\u96c6\u5c31\u662f\u6765\u8865\u4e0a\u8fd9\u5757\u7a7a\u7f3a\u7684\u3002\u5177\u4f53\u800c\u8a00\uff0cOpen R1 \u56e2\u961f\u4f7f\u7528 DeepSeek R1 \u751f\u6210\u4e86 80 \u4e07\u6761\u63a8\u7406\u8f68\u8ff9\uff0c\u7ecf\u8fc7\u7b5b\u9009\u548c\u9a8c\u8bc1\u540e\u5f97\u5230\u4e86 22 \u4e07\u6761\u9ad8\u8d28\u91cf\u6570\u636e\u3002<\/section>\n<section><\/section>\n<section>\u8fd9\u4e9b\u6570\u636e\u53ef\u4ee5\u7528\u6765\u652f\u6301\u66f4\u5c0f\u7684\u6a21\u578b\uff0c\u6765\u8fbe\u5230\u5ab2\u7f8e DeepSeek R1 \u7684\u6548\u679c\u3002\u6bd4\u5982<strong>\u5728 OpenR1-Math-220k \u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u51fa\u6765\u7684 Qwen-7B-Math-Instruct\uff0c\u8fbe\u5230\u4e86\u4e0e DeepSeek-Distill-Qwen-7B \u76f8\u5f53\u7684\u6027\u80fd<\/strong>\u3002<\/section>\n<section><\/section>\n<section>\u4e0b\u9762\u5c31\u8ba9\u6211\u4eec\u6765\u770b\u770b OpenR1-Math-220k \u6570\u636e\u96c6\u7684\u7279\u70b9\uff0c\u4ee5\u53ca\u5b83\u662f\u600e\u4e48\u8bde\u751f\u7684\uff1a<\/section>\n<section><\/section>\n<section><strong>\u81ea\u52a8\u8fc7\u6ee4\u6b63\u786e\u7b54\u6848<\/strong><\/section>\n<section><strong>\u672c\u5730\u9ad8\u6548\u751f\u6210<\/strong><\/section>\n<section><\/section>\n<section>\u76ee\u524d\uff0c\u5f00\u6e90\u793e\u533a\u5df2\u7ecf\u53d1\u5e03\u4e86 OpenThoughts-114k\u3001Bespoke-Stratos-17k\u3001Dolphin-R1 \u548c LIMO \u7b49\u591a\u4e2a\u63a8\u7406\u6570\u636e\u96c6\u3002<\/section>\n<section><\/section>\n<section>\u7136\u800c\uff0c\u4e3a\u4e86\u8fdb\u884c\u63a8\u7406\u516c\u4f17\u516c\u5f00\u53d1\u5e03\uff0c\u8fd9\u4fc3\u4f7f\u793e\u533a\u72ec\u7acb\u91cd\u5efa\u7c7b\u4f3c\u7684\u6570\u636e\u96c6\u3002\u53e6\u5916\uff0c\u793e\u533a\u5df2\u7ecf\u53d1\u5e03\u4e86\u591a\u4e2a\u5f00\u653e\u6570\u636e\u96c6\uff0c\u5305\u62ec OpenThoughts-114k\u3001Bespoke-Stratos-17k\u3001Dolphin-R1 \u548c LIMO\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u6574\u7406 OpenR1-Math-220k\uff0cOpen R1 \u8fd8\u548c\u5f00\u53d1\u4e86\u5e7f\u53d7\u6b22\u8fce\u7684 NuminaMath-CoT \u6570\u636e\u96c6\u5f00\u53d1\u56e2\u961f Numina \u8fdb\u884c\u4e86\u5408\u4f5c\u3002<\/section>\n<section><\/section>\n<section>\u90a3\u4e48\uff0c\u4e0e\u73b0\u6709\u6570\u636e\u96c6\u76f8\u6bd4\uff0cOpen R1 \u7684\u6570\u636e\u96c6\u6709\u4ec0\u4e48\u65b0\u7279\u70b9\u5462\uff1f<\/section>\n<section><\/section>\n<ul>\n<li>\n<section><strong>80 \u4e07\u6761 R1 \u63a8\u7406\u8f68\u8ff9<\/strong>\uff1a\u4f7f\u7528 DeepSeek R1 \u4e3a 40 \u4e07\u4e2a\u95ee\u9898\u5404\u751f\u6210\u4e86\u4e24\u4e2a\u7b54\u6848\uff0c\u6700\u7ec8\u7ecf\u8fc7\u7b5b\u9009\u540e\u4fdd\u7559\u4e86 22 \u4e07\u4e2a\u5177\u6709\u6b63\u786e\u63a8\u7406\u8f68\u8ff9\u7684\u95ee\u9898\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u672c\u5730\u8fd0\u884c 512 \u4e2a H100<\/strong>\uff1a\u6ca1\u6709\u4f9d\u8d56 API\uff0c\u800c\u662f\u5728\u8ba1\u7b97\u96c6\u7fa4\u4e0a\u5229\u7528 vLLM \u548c SGLang \u672c\u5730\u8fd0\u884c\u751f\u6210\u4efb\u52a1\uff0c\u6bcf\u5929\u53ef\u4ee5\u751f\u6210 18 \u4e07\u6761\u63a8\u7406\u8f68\u8ff9\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u57fa\u4e8e NuminaMath 1.5<\/strong>\uff1a\u4e13\u6ce8\u4e8e\u6570\u5b66\u63a8\u7406\u516c\u5f0f\uff0c\u4e3a NuminaMath 1.5\uff08NuminaMath-CoT \u6570\u636e\u96c6\u7684\u6539\u8fdb\u7248\u672c\uff09\u4e2d\u7684\u95ee\u9898\u751f\u6210\u7b54\u6848\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u81ea\u52a8\u8fc7\u6ee4<\/strong>\uff1aOpen R1 \u56e2\u961f\u901a\u8fc7\u6570\u5b66\u9a8c\u8bc1\uff0c\u53ea\u4fdd\u7559\u81f3\u5c11\u6709\u4e00\u4e2a\u6b63\u786e\u7b54\u6848\u7684\u95ee\u9898\uff0c\u8fd8\u8ba9 Llama3.3-70B-Instruct \u4f5c\u4e3a\u300c\u5224\u5b98\u300d\uff0c\u4ee5\u7b5b\u9009\u51fa\u66f4\u591a\u6b63\u786e\u7684\u6837\u672c\uff0c\u7279\u522b\u662f\u90a3\u4e9b\u56e0\u683c\u5f0f\u9519\u8bef\u800c\u65e0\u6cd5\u901a\u8fc7\u57fa\u4e8e\u89c4\u5219\u7684\u89e3\u6790\u5668\u9a8c\u8bc1\u7684\u7b54\u6848\u3002<\/section>\n<\/li>\n<li>\n<section>\u5728 OpenR1-Math-220k \u8bad\u7ec3\u51fa\u6765\u7684 Qwen-7B-Math-Instruct\uff0c\u8fbe\u5230\u4e86\u4e0e DeepSeek-Distill-Qwen-7B \u76f8\u5f53\u7684\u6027\u80fd\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u6570\u636e\u96c6\u5206\u4e3a\u4e24\u4e2a\u90e8\u5206\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>default\uff0894k \u95ee\u9898\uff09\uff1a\u8fd9\u90e8\u5206\u6570\u636e\u5728\u7ecf\u8fc7\u76d1\u7763\u5fae\u8c03\uff08SFT\uff09\u540e\u8868\u73b0\u6700\u4f73\u3002<\/section>\n<\/li>\n<li>\n<section>extended\uff08131k \u95ee\u9898\uff09\uff1a\u8fd9\u90e8\u5206\u6570\u636e\u5305\u542b\u989d\u5916\u7684 NuminaMath 1.5 \u6570\u636e\u6e90\uff0c\u4f8b\u5982 cn_k12\uff0c\u63d0\u4f9b\u4e86\u66f4\u591a\u7684\u63a8\u7406\u516c\u5f0f\u3002\uff08\u7814\u7a76\u53d1\u73b0\u8fd9\u4e2a\u5b50\u96c6\u5728\u7ecf\u8fc7\u76d1\u7763\u5fae\u8c03\u540e\u7684\u6027\u80fd\u4f4e\u4e8e\u9ed8\u8ba4\u6570\u636e\u96c6\uff0c\u53ef\u80fd\u662f\u56e0\u4e3a cn_k12 \u4e2d\u7684\u95ee\u9898\u76f8\u5bf9\u7b80\u5355\u3002\uff09<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-722adec3fe3ec1e7389d19a92c9da822.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-722adec3fe3ec1e7389d19a92c9da822.png\"><\/a><\/section>\n<section>Open R1 \u56e2\u961f\u8868\u793a\uff0c\u5e0c\u671b\u8fd9\u4e2a\u53ef\u6269\u5c55\u7684\u3001\u9ad8\u8d28\u91cf\u7684\u63a8\u7406\u6570\u636e\u751f\u6210\u8fc7\u7a0b\uff0c\u80fd\u591f\u542f\u53d1\u4ee3\u7801\u751f\u6210\u7b49\u6570\u5b66\u4e4b\u5916\u7684\u9886\u57df\u3002<\/section>\n<section><\/section>\n<section><strong>\u6570\u636e\u751f\u6210<\/strong><\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u6784\u5efa\u6570\u636e\u96c6\uff0cOpenR1 \u56e2\u961f\u8ba9 DeepSeek R1 \u4e3a\u6765\u81ea NuminaMath 1.5 \u7684 40 \u4e07\u4e2a\u95ee\u9898\u751f\u6210\u7b54\u6848\u3002\u4ed6\u4eec\u9075\u5faa\u4e86 DeepSeek \u6280\u672f\u62a5\u544a\u4e2d\u63a8\u8350\u7684<mark data-type=\"concepts\" data-id=\"2e982b73-88e2-41e8-a430-f7ae5a9af4bf\">\u53c2\u6570<\/mark>\u8bbe\u7f6e\uff0c\u5e76\u5728\u63d0\u793a\u8bcd\u524d\u6dfb\u52a0\u4e86\u4ee5\u4e0b\u6307\u4ee4\uff1a<\/section>\n<section>\n<pre data-lang=\"php\"><section><code>Please reason step by step, and put your final answer within boxed{}.<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u786e\u4fdd\u751f\u6210\u8fc7\u7a0b\u7684\u9ad8\u6548\u6027\uff0c\u56e2\u961f\u5c06\u6bcf\u6b21\u751f\u6210\u7684 tokens \u9650\u5236\u8bbe\u7f6e\u4e3a 16k\u3002\u7ecf\u8fc7\u5206\u6790\u53d1\u73b0\uff0c\u53ea\u6709 75% \u7684\u95ee\u9898\u80fd\u591f\u5728 8k tokens \u5185\u89e3\u51b3\uff0c\u800c\u5927\u591a\u6570\u5269\u4f59\u95ee\u9898\u9700\u8981\u5b8c\u6574\u7684 16k tokens\u3002<\/section>\n<section><\/section>\n<section>\u6700\u521d\uff0c\u4ed6\u4eec\u4f7f\u7528 vLLM \u8fdb\u884c\u63a8\u7406\uff0c\u6bcf\u4e2a H100 \u8282\u70b9\u6bcf\u79d2\u53ef\u4ee5\u751f\u6210 15 \u4e2a\u7b54\u6848\uff0c\u5e76\u4e14\u76f8\u5173\u751f\u6210\u811a\u672c\u5df2\u5206\u4eab\u5728 OpenR1 \u4ed3\u5e93\u4e2d\u3002\u6700\u8fd1\uff0c\u4ed6\u4eec\u53c8\u5f00\u59cb\u5c1d<strong>\u8bd5\u4f7f\u7528 SGLang\uff0c\u6bcf\u4e2a H100 \u8282\u70b9\u6bcf\u79d2\u53ef\u4ee5\u751f\u6210 25 \u4e2a\u7b54\u6848\uff08\u901f\u5ea6\u63d0\u5347\u4e86\u8fd1\u4e24\u500d\uff09<\/strong>\uff0c\u8fd9\u4f7f\u5f97 512 \u4e2a H100 \u8282\u70b9\u4e0a\u6bcf\u5929\u80fd\u751f\u6210 30 \u4e07\u4e2a\u95ee\u9898\u7684\u7b54\u6848\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u5728\u540e\u7eed\u7684\u8fc7\u6ee4\u548c\u4f18\u5316\u8fc7\u7a0b\u4e2d\u63d0\u4f9b\u66f4\u5927\u7684\u7075\u6d3b\u6027\uff0c\u56e2\u961f\u4e3a\u6bcf\u4e2a\u95ee\u9898\u751f\u6210\u4e86\u4e24\u4e2a\u7b54\u6848 \u2014\u2014 \u6709\u65f6\u751a\u81f3\u751f\u6210\u56db\u4e2a\u3002\u8fd9\u6837\u4e00\u6765\uff0c\u4e0d\u4ec5\u590d\u523b\u51fa\u4e86\u7c7b\u4f3c\u4e8e DeepSeek R1 \u5141\u8bb8\u8fdb\u884c<mark data-type=\"tech_methods\" data-id=\"c84b6d7e-447f-4bcb-b4a5-c807d7b8a5f7\">\u62d2\u7edd\u91c7\u6837<\/mark>\u7684\u65b9\u6cd5\uff0c\u8fd8\u80fd\u4f7f\u6570\u636e\u96c6\u80fd\u591f\u9002\u7528\u4e8e\u5982 DPO \u7b49\u504f\u597d\u4f18\u5316\u65b9\u6cd5\u3002<\/section>\n<section><\/section>\n<section>\u6570\u636e\u751f\u6210\u811a\u672c\uff1ahttps:\/\/github.com\/huggingface\/open-r1\/tree\/main\/slurm\u3002<\/section>\n<section><\/section>\n<section><strong>\u6570\u636e\u8fc7\u6ee4<\/strong><\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u786e\u4fdd\u6570\u636e\u96c6\u4e2d\u53ea\u5305\u542b\u9ad8\u8d28\u91cf\u4e14\u6b63\u786e\u7684\u63a8\u7406\u7ed3\u679c\uff0cOpen R1 \u56e2\u961f\u8bbe\u8ba1\u4e86\u4e00\u5957\u6570\u5b66\u9a8c\u8bc1\u7cfb\u7edf\uff0c\u7528\u4e8e\u81ea\u52a8\u6bd4\u5bf9 LLM \u751f\u6210\u7684\u590d\u6742\u6570\u5b66\u8868\u8fbe\u5f0f\u7b54\u6848\u4e0e\u6570\u636e\u96c6\u4e2d\u7684\u6807\u51c6\u7b54\u6848\u3002<\/section>\n<section><\/section>\n<section>\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\uff0cOpenR1 \u56e2\u961f\u53d1\u73b0\u5927\u7ea6 55% \u7684\u95ee\u9898\u81f3\u5c11\u6709\u4e00\u4e2a\u6b63\u786e\u7b54\u6848\u3002\u7136\u800c\uff0cNuminaMath 1.5 \u6570\u636e\u96c6\u4e2d\u6709\u5f88\u591a\u7b54\u6848\u662f\u7a7a\u7684\uff0c\u6216\u8005\u683c\u5f0f\u4e0d\u7b26\u5408\u9a8c\u8bc1\u6807\u51c6\uff0c\u8fd9\u90fd\u7ed9\u81ea\u52a8\u9a8c\u8bc1\u5e26\u6765\u4e86\u56f0\u96be\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0cOpen R1 \u56e2\u961f\u5148\u662f\u5bf9 Math-Verify \u5de5\u5177\u8fdb\u884c\u4e86\u6539\u8fdb\uff0c\u4f7f\u5176\u80fd\u591f\u5904\u7406\u66f4\u591a\u4e0d\u5e38\u89c1\u7684\u7b54\u6848\u683c\u5f0f\uff0c\u518d\u4f7f\u7528 Llama-3.3-70B-Instruct \u6a21\u578b\u8fdb\u884c\u4e8c\u6b21\u8bc4\u4f30\u3002<\/section>\n<section><\/section>\n<section>\u5177\u4f53\u6765\u8bf4\uff0c\u5bf9\u4e8e\u90a3\u4e9b\u88ab Math-Verify \u5224\u5b9a\u4e3a\u9519\u8bef\u7684\u7b54\u6848\uff0c\u4f7f\u7528 Llama-3.3-70B-Instruct \u6a21\u578b\u91cd\u65b0\u8bc4\u4f30\uff0c\u8bc6\u522b\u5b9e\u9645\u4e0a\u6b63\u786e\u4f46\u56e0\u683c\u5f0f\u95ee\u9898\u88ab\u9519\u5224\u7684\u7b54\u6848\u3002\u6700\u7ec8\uff0c\u4ed6\u4eec\u627e\u56de\u4e86 2.5 \u4e07\u6761\u88ab\u300c\u8bef\u5224\u300d\u7684\u6570\u636e\u3002<\/section>\n<section><\/section>\n<section>\u4f18\u5316 Math-Verify \u5de5\u5177\uff1a\u5bf9 Math-Verify \u5de5\u5177\u8fdb\u884c\u4e86\u6539\u8fdb\uff0c\u4f7f\u5176\u80fd\u591f\u5904\u7406\u66f4\u591a\u4e0d\u5e38\u89c1\u7684\u7b54\u6848\u683c\u5f0f\u3002<\/section>\n<section><\/section>\n<section>\u8ba9 Llama-3.3-70B-Instruct \u300c\u4f5c\u5224\u5b98\u300d\u7684\u63d0\u793a\u8bcd\u5982\u4e0b\uff1a<\/section>\n<section>\n<pre data-lang=\"sql\"><section><code>You are a mathematical answer validator. You will be provided with a mathematical problem and you need to compare the answer in the reference solution, and the final answer in a model's solution to determine if they are equivalent, even if formatted differently.<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>PROBLEM:<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>{problem}<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>REFERENCE SOLUTION:<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>{answer}<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>MODEL'S SOLUTION:<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>{generation}<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>Focus ONLY on comparing the final mathematical answer provided by the model while ignoring differences in:<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>- Formatting (e.g., \\boxed{{}} vs plain text)<\/code><\/section><section><code>- Multiple choice formatting (e.g., \"A\" vs full solution)<\/code><\/section><section><code>- Order of coordinate pairs or solutions<\/code><\/section><section><code>- Equivalent mathematical expressions or notation variations<\/code><\/section><section><code>- If the model's answer is nonsense, return \"Verdict: AMBIGUOUS\"<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>Start with a brief explanation of your comparison (2-3 sentences). Then output your final answer in one of the following formats:<\/code><code>\r\n<\/code><\/section><section>\r\n<\/section><section><code>- \"Verdict: EQUIVALENT\"<\/code><\/section><section><code>- \"Verdict: DIFFERENT\"<\/code><\/section><section><code>- \"Verdict: AMBIGUOUS\"<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section>\u5bf9\u4e8e\u90a3\u4e9b\u5305\u542b\u591a\u4e2a\u6b63\u786e\u7b54\u6848\u7684\u6570\u636e\u884c\uff0c\u56e2\u961f\u5c1d\u8bd5\u4f7f\u7528\u5956\u52b1\u6a21\u578b\uff08RM\uff09\u4f5c\u4e3a\u6700\u7ec8\u7b5b\u9009\u5668\u6765\u9009\u62e9\u6700\u4f73\u7b54\u6848\u3002\u5177\u4f53\u64cd\u4f5c\u5982\u4e0b\uff1a<\/section>\n<section><\/section>\n<section>\u9996\u5148\uff0c\u4ece\u6bcf\u4e2a\u5305\u542b\u591a\u4e2a\u6b63\u786e\u7b54\u6848\u7684\u6570\u636e\u884c\u4e2d\uff0c\u53bb\u6389\uff08\u2026\uff09\uff0c\u63d0\u53d6\u6700\u7ec8\u7b54\u6848\uff1b\u7b2c\u4e8c\uff0c\u5c06\u95ee\u9898\u548c\u63d0\u53d6\u7684\u7b54\u6848\u8f93\u5165\u5230\u914d\u7f6e\u4e86 vLLM \u7684 Qwen\/Qwen2.5-Math-RM-72B \u6a21\u578b\u4e2d\uff0c\u83b7\u53d6\u6bcf\u4e2a\u7b54\u6848\u7684\u8bc4\u5206\uff1b\u63a5\u7740\uff0c\u6839\u636e\u6a21\u578b\u8bc4\u5206\uff0c\u5bf9\u6bcf\u4e2a\u5305\u542b\u591a\u4e2a\u6b63\u786e\u7b54\u6848\u7684\u6570\u636e\u884c\u6392\u540d\uff0c\u9009\u62e9\u6392\u540d\u6700\u9ad8\u7684\u7b54\u6848\u7eb3\u5165\u8bad\u7ec3\u6570\u636e\u96c6\u3002<\/section>\n<section><\/section>\n<section>\u9057\u61be\u7684\u662f\uff0c\u6d88\u878d\u5b9e\u9a8c\u8868\u660e\uff0c<strong>\u8fd9\u79cd\u65b9\u6cd5\u5e76\u6ca1\u6709\u6bd4\u968f\u673a\u9009\u62e9\u4e00\u4e2a\u6b63\u786e\u7b54\u6848\u5e26\u6765\u66f4\u597d\u7684\u6a21\u578b\u6027\u80fd<\/strong>\u3002Open R1 \u56e2\u961f\u7684\u5224\u65ad\u662f\uff0c\u53ef\u80fd\u5728\u4f7f\u7528\u5956\u52b1\u6a21\u578b\u8bc4\u5206\u65f6\uff0c\u4e0d\u4ec5\u8981\u8003\u8651\u6700\u7ec8\u7b54\u6848\uff0c\u8fd8\u8981\u5305\u62ec\u63a8\u7406\u8fc7\u7a0b\u3002<\/section>\n<section><\/section>\n<section><strong>Open R1 \u8bad\u7ec3\u51fa\u6765\u7684\u65b0\u6a21\u578b VS DeepSeek-Distill-Qwen-7B<\/strong><\/section>\n<section><\/section>\n<section>Open R1 \u5728 OpenR1-Math-220k \u7684\u57fa\u7840\u4e0a\uff0c\u5bf9 Qwen2.5-Math-Instruct \u8fdb\u884c\u4e86 3 \u8f6e\u5fae\u8c03\uff0c<mark data-type=\"concepts\" data-id=\"444ad8d1-7262-46c9-94b7-fe0aa9227006\">\u5b66\u4e60\u7387<\/mark>\u4e3a 5e-5\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u5c06\u4e0a\u4e0b\u6587\u957f\u5ea6\u4ece 4k \u6269\u5c55\u5230 32k\uff0c\u4ed6\u4eec\u5c06 RoPE \u9891\u7387\u63d0\u9ad8\u5230 300k\u3002\u8bad\u7ec3\u9075\u5faa\u7ebf\u6027<mark data-type=\"concepts\" data-id=\"444ad8d1-7262-46c9-94b7-fe0aa9227006\">\u5b66\u4e60\u7387<\/mark><mark data-type=\"tech_methods\" data-id=\"1ef8dd91-7c80-46f2-811e-0722d3d333d4\">\u8c03\u5ea6<\/mark>\uff0c\u5176\u4e2d\u5305\u542b 10% \u7684\u9884\u70ed\u9636\u6bb5\u3002<\/section>\n<section><\/section>\n<section>\u4e0b\u8868\u5c55\u793a\u4e86\u5728 lighteval \u4e0a OpenR1-Qwen-7B\u3001DeepSeek-Distill-Qwen-7B \u548c OpenThinker-7B \u7684\u6027\u80fd\u5bf9\u6bd4\uff0c\u53ef\u4ee5\u770b\u51fa\u5728\u6570\u5b66\u6210\u7ee9\u4e0a\uff0cOpenR1-Qwen-7B \u548c DeepSeek-Distill-Qwen-7B \u5dee\u8ddd\u4e0d\u662f\u975e\u5e38\u660e\u663e\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-e3971cdae8c3d0388d1334cf3edc120d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-e3971cdae8c3d0388d1334cf3edc120d.png\"><\/a><\/section>\n<section><em><sup>\u672c\u5468 AIME 2025 \u5f00\u8d5b\uff0c\u6765\u81ea\u82cf\u9ece\u4e16\u8054\u90a6\u7406\u5de5\u5b66\u9662\u7684\u7814\u7a76\u4eba\u5458\u7528\u65b0\u9898\u6765\u6d4b\u8bc4\u4e86\u591a\u6b3e\u6a21\u578b\uff0c\u5374\u53d1\u73b0\u5404\u79cd\u6a21\u578b\u7684\u6570\u5b66\u80fd\u529b\u4e0b\u964d\u4e86 10-20 \u4e2a\u767e\u5206\u70b9\u4e4b\u95f4\u3002\u6709\u7814\u7a76\u8005\u53d1\u73b0\u8fd9\u4e9b\u300c\u5168\u65b0\u300d\u7684 AIME \u9898\u53ef\u80fd\u5df2\u7ecf\u5728\u4e92\u8054\u7f51\u8bba\u575b\u6cc4\u9732\uff0c\u8fd9\u6709\u53ef\u80fd\u9020\u6210\u610f\u5916\u7684<mark data-type=\"concepts\" data-id=\"af836eef-be90-4143-a022-46fae3904f0e\">\u8fc7\u62df\u5408<\/mark>\u95ee\u9898\uff0c\u8fd9\u4e5f\u51f8\u663e\u4e86\u65b0\u9c9c\u6d4b\u8bd5\u6570\u636e\u7684\u56f0\u5883\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u540c\u65f6\uff0c\u5f00\u6e90\u793e\u533a\u4e5f\u4ece\u591a\u4e2a\u89d2\u5ea6\u63a2\u7d22\u4e86 GRPO\uff0c\u6709\u591a\u4e2a\u7814\u7a76\u5b9e\u9a8c\u5ba4\u8868\u660e\uff0c<strong>\u5927\u7ea6 1000 \u4e2a\u9ad8\u8d28\u91cf\u7684\u8bad\u7ec3\u6837\u672c\u53ef\u80fd\u5c31\u8db3\u4ee5\u5728\u73b0\u6709\u7684\u5f00\u6e90\u6a21\u578b\u4e2d\u5f15\u53d1\u63a8\u7406\u80fd\u529b<\/strong>\u3002<\/section>\n<section><\/section>\n<section><strong>\u542f\u793a\u548c\u65b0\u7684\u95ee\u9898<\/strong><\/section>\n<section><\/section>\n<section><strong>LLM \u662f\u5426\u9700\u8981\u4f7f\u7528\u81ea\u7136\u8bed\u8a00\u8fdb\u884c\u63a8\u7406\uff1f<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-3ffc2bfb2b19016f8bcf09ba0d4f8921.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-3ffc2bfb2b19016f8bcf09ba0d4f8921.png\"><\/a><\/section>\n<section>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2502.05171<\/section>\n<section><\/section>\n<section>\u9a6c\u91cc\u5170\u5927\u5b66\u7684\u4e00\u7bc7\u8bba\u6587\u8868\u660e\uff0c\u901a\u8fc7\u4f7f\u7528\u5faa\u73af<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark>\uff0c\u53ef\u4ee5\u5728\u6f5c\u5728\u7a7a\u95f4\u4e2d\u9690\u5f0f\u63a8\u7406\uff0c\u4ece\u800c\u5728\u6d4b\u8bd5\u65f6\u6269\u5c55\u8ba1\u7b97\u80fd\u529b\uff0c\u8fd9\u7c7b\u4f3c\u4e8e Meta \u7684 Coconut\u3002\u8fd9\u4e9b\u65b9\u6cd5\u7684\u4f18\u52bf\u5728\u4e8e\u5b83\u4eec\u7684\u8ba1\u7b97\u6548\u7387\u66f4\u9ad8\uff1a\u901a\u8fc7\u63a2\u7d22\u6f5c\u5728\u7a7a\u95f4\uff0c\u65e0\u9700\u751f\u6210\u5927\u91cf\u300c\u601d\u8003\u300dtoken \u5373\u53ef\u83b7\u5f97\u9ad8\u6027\u80fd\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-162d8033a81a0dfce799fdb69a86905d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-162d8033a81a0dfce799fdb69a86905d.png\"><\/a><\/section>\n<section><strong>\u662f\u5426\u8981\u8f6c\u5411\u66f4\u5c0f\u3001\u9ad8\u8d28\u91cf\u7684\u63a8\u7406\u6570\u636e\uff1f<\/strong><\/section>\n<section><\/section>\n<section>\u5c3d\u7ba1 DeepSeek R1 \u4f7f\u7528\u4e86 600k \u63a8\u7406\u8f68\u8ff9\u8fdb\u884c\u84b8\u998f\uff0c\u4f46\u6700\u65b0\u7814\u7a76\u8868\u660e\uff0c<strong>\u590d\u6742\u7684\u63a8\u7406\u80fd\u529b\u5e76\u975e\u5355\u7eaf\u901a\u8fc7\u5927\u89c4\u6a21\u8bad\u7ec3\u5728<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark>\u4e2d\u5b9e\u73b0\uff0c\u800c\u662f\u53ef\u4ee5\u901a\u8fc7\u5c11\u91cf\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u6837\u672c\u8fbe\u6210<\/strong>\u3002<\/section>\n<section><\/section>\n<section>s1K \u6570\u636e\u96c6\u5c31\u662f\u4e00\u4e2a\u5f88\u597d\u7684\u4f8b\u5b50\u3002\u5b83\u5305\u542b 1000 \u4e2a\u7ecf\u8fc7\u7cbe\u5fc3\u6311\u9009\u7684\u6570\u5b66\u95ee\u9898\uff0c\u4ee5\u53ca\u4ece Gemini Flash \u84b8\u998f\u51fa\u7684\u63a8\u7406\u8f68\u8ff9\u3002\u5728\u9009\u62e9\u95ee\u9898\u65f6\uff0c\u7814\u7a76\u8005\u6ce8\u91cd\u96be\u5ea6\u3001\u591a\u6837\u6027\u548c\u8d28\u91cf\u3002\u901a\u8fc7\u5728 s1K \u6570\u636e\u96c6\u4e0a\u5bf9 Qwen2.5-32B-Instruct \u8fdb\u884c\u5fae\u8c03\uff0c\u7814\u7a76\u8005\u6210\u529f\u4f7f\u5176\u5728\u7ade\u8d5b\u6570\u5b66<mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u6d4b\u8bd5\u4e2d\u8d85\u8fc7\u4e86 OpenAI \u7684 o1-preview\uff0c\u6700\u9ad8\u63d0\u5347\u4e86 27%\u3002<\/section>\n<section><\/section>\n<section>\u53e6\u4e00\u4e2a\u6570\u636e\u96c6 LIMO \u4e5f\u8fdb\u4e00\u6b65\u9a8c\u8bc1\u4e86\u8fd9\u4e00\u7406\u5ff5\u3002\u5b83\u4ec5\u4f7f\u7528 817 \u4e2a\u8bad\u7ec3\u6837\u672c\uff0c\u5c31\u5728 AIME \u548c MATH <mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u6d4b\u8bd5\u4e2d\u53d6\u5f97\u4e86\u51fa\u8272\u7684\u8868\u73b0\u3002LIMO \u7684\u4f5c\u8005\u63a8\u6d4b\uff0c\u5f53\u6a21\u578b\u5728\u9884\u8bad\u7ec3\u9636\u6bb5\u5df2\u7ecf\u79ef\u7d2f\u4e86\u4e30\u5bcc\u7684\u77e5\u8bc6\u540e\uff0c\u53ef\u80fd\u53ea\u9700\u8981\u5c11\u91cf\u7ed3\u6784\u826f\u597d\u7684\u6837\u672c\uff0c\u5c31\u80fd\u89e3\u9501\u9ad8\u7ea7\u63a8\u7406\u80fd\u529b\u3002<\/section>\n<section><\/section>\n<section><strong>CoT \u957f\u5ea6\uff1a\u9884\u7b97\u5f3a\u5236\u4e0e\u5956\u52b1\u5851\u9020<\/strong><\/section>\n<section><\/section>\n<section>Qwen2.5-32B-Instruct \u6a21\u578b\u5728 s1K \u6570\u636e\u96c6\u4e0a\u5fae\u8c03\u540e\u8868\u73b0\u51fa\u8272\uff0c\u5176\u4e2d\u4e00\u4e2a\u5173\u952e\u56e0\u7d20\u662f\u91c7\u7528\u4e86\u300c\u9884\u7b97\u5f3a\u5236\u300d\u3002\u8fd9\u662f\u4e00\u79cd\u6d4b\u8bd5\u65f6\u7684\u8ba1\u7b97\u6280\u672f\uff0c\u901a\u8fc7\u5728\u6a21\u578b\u751f\u6210\u4e2d\u6dfb\u52a0\u300c\u7b49\u5f85\u300dtoken \u6765\u5ef6\u957f\u63a8\u7406\u65f6\u95f4\uff0c\u6216\u8005\u6dfb\u52a0\u300c\u7ed3\u675f\u601d\u8003\u300d\u7684 token \u6765\u622a\u65ad\u63a8\u7406\u3002<\/section>\n<section><\/section>\n<section>\u8fd9\u79cd\u65b9\u6cd5\u4f7f\u7814\u7a76\u8005\u80fd\u591f\u7075\u6d3b\u8c03\u6574\u6a21\u578b\u7684\u601d\u8003\u65f6\u95f4\uff0c\u5e76\u53d1\u73b0\u968f\u7740\u601d\u8003\u65f6\u95f4\u7684\u589e\u52a0\uff0c\u6a21\u578b\u5728\u4e0d\u540c\u6570\u5b66<mark data-type=\"concepts\" data-id=\"308c3a45-0fee-4ec6-858e-85b15f440fc0\">\u57fa\u51c6<\/mark>\u6d4b\u8bd5\u4e2d\u7684\u51c6\u786e\u6027\u4e5f\u968f\u4e4b\u63d0\u9ad8\u3002<\/section>\n<section><\/section>\n<section>\u540c\u6837\uff0cYeo \u7b49\u4eba\u5728\u7814\u7a76\u300aDemystifying Long Chain-of-Thought Reasoning in LLMs\u300b\u4e2d\u63a2\u8ba8\u4e86\u601d\u7ef4\u94fe\uff08CoT\uff09\u957f\u5ea6\u5bf9\u6a21\u578b\u6027\u80fd\u7684\u5f71\u54cd\u3002\u4ed6\u4eec\u5f15\u5165\u4e86\u4e00\u79cd\u540d\u4e3a\u300c\u4f59\u5f26\u5956\u52b1\u300d\u7684\u65b0\u5956\u52b1\u51fd\u6570\uff0c\u7528\u4e8e\u5728\u6b63\u786e\u751f\u6210\u65f6\u6fc0\u52b1\u8f83\u77ed\u7684 CoT\uff0c\u5728\u9519\u8bef\u751f\u6210\u65f6\u6fc0\u52b1\u8f83\u957f\u7684 CoT\u3002\u8fd9\u79cd\u5956\u52b1\u673a\u5236\u5728\u6a21\u578b\u7684\u6700\u5927\u4e0a\u4e0b\u6587\u5927\u5c0f\u6709\u9650\u4e14\u5e73\u5747\u54cd\u5e94\u957f\u5ea6\u53ef\u80fd\u5931\u63a7\u7684\u60c5\u51b5\u4e0b\uff0c\u80fd\u591f\u7a33\u5b9a<mark data-type=\"tech_methods\" data-id=\"ee1a8f69-3170-4ddf-b2b6-47d91c844425\">\u5f3a\u5316\u5b66\u4e60<\/mark>\u8bad\u7ec3\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-2ad7d28fc5a8d309ad7b35f8a1520c36.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-2ad7d28fc5a8d309ad7b35f8a1520c36.png\"><\/a><\/section>\n<section>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2502.03373<\/section>\n<section><\/section>\n<section>\u6b64\u5916\uff0c\u5f53\u6a21\u578b\u5728\u5904\u7406\u96be\u9898\u65f6\u51fa\u73b0\u5956\u52b1\u52ab\u6301\u7684\u8ff9\u8c61\uff08\u5373\u901a\u8fc7\u91cd\u590d\u800c\u975e\u771f\u6b63\u89e3\u51b3\u95ee\u9898\u6765\u589e\u52a0 CoT \u957f\u5ea6\uff09\uff0c\u7814\u7a76\u8005\u8fd8\u4f1a\u91c7\u7528\u91cd\u590d\u60e9\u7f5a\u673a\u5236\uff0c\u4ee5\u907f\u514d\u6a21\u578b\u9677\u5165\u65e0\u6548\u7684\u5faa\u73af\u63a8\u7406\u3002<\/section>\n<section><\/section>\n<section>Open R1 \u56e2\u961f\u8868\u793a\uff0c\u73b0\u5728 GRPO \u5df2\u7ecf\u5728 TRL \u4e2d\u987a\u5229\u8fd0\u884c\uff0c\u4ed6\u4eec\u6b63\u5728\u5f00\u5c55\u4e00\u7cfb\u5217\u5e7f\u6cdb\u7684\u5b9e\u9a8c\uff0c\u4ee5\u4e86\u89e3\u54ea\u4e9b\u8d85<mark data-type=\"concepts\" data-id=\"2e982b73-88e2-41e8-a430-f7ae5a9af4bf\">\u53c2\u6570<\/mark>\u548c\u5956\u52b1\u51fd\u6570\u5bf9\u8bad\u7ec3\u7684\u5f71\u54cd\u6700\u5927\u3002<\/section>\n<section><\/section>\n<section>\u90a3\u5c31\u8ba9\u6211\u4eec\u9759\u5f85 Open R1 \u7684\u597d\u6d88\u606f\u5427\u3002<\/section>\n<section><\/section>\n<section><em><sup>\u53c2\u8003\u94fe\u63a5\uff1a<\/sup><\/em><\/section>\n<section><sup><em>https:\/\/huggingface.co\/blog\/open-r1\/update-2<\/em><\/sup><\/section>\n<section><sup><em>https:\/\/x.com\/_lewtun\/status\/1889002019316506684<\/em><\/sup><\/section>\n<section><em><sup>https:\/\/huggingface.co\/open-r1<\/sup><\/em><\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-11-5\" target=\"_blank\">\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u5f00\u6e9022\u4e07\u6761DeepS [&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],"class_list":["post-38235","post","type-post","status-publish","format-standard","hentry","category-news","tag-68"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.9 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86 - \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=38235\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86\",\"datePublished\":\"2025-02-12T02:00:55+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235\"},\"wordCount\":292,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png\",\"keywords\":[\"\u5927\u6a21\u578b\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=38235#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=38235\",\"name\":\"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png\",\"datePublished\":\"2025-02-12T02:00:55+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=38235\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2025\\\/02\\\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38235#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86\"}]},{\"@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":"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86 - \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=38235","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=38235#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=38235"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86","datePublished":"2025-02-12T02:00:55+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=38235"},"wordCount":292,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=38235#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png","keywords":["\u5927\u6a21\u578b"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=38235#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=38235","url":"https:\/\/17aitech.com\/?p=38235","name":"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=38235#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=38235#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png","datePublished":"2025-02-12T02:00:55+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=38235#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=38235"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=38235#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-724cfbfe62f16519d24b3cc65ffc44bd.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=38235#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u5f00\u6e9022\u4e07\u6761DeepSeek R1\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff01\u4f60\u4e5f\u80fd\u590d\u73b0DeepSeek\u4e86"}]},{"@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\/38235","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=38235"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38235\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38235"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38235"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38235"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}