{"id":64573,"date":"2025-09-18T07:37:27","date_gmt":"2025-09-17T23:37:27","guid":{"rendered":"https:\/\/swarma.org\/?p=64573"},"modified":"2025-09-18T07:37:27","modified_gmt":"2025-09-17T23:37:27","slug":"nature%e5%b0%81%e9%9d%a2%e6%96%87%e7%ab%a0-deepseek-r1-%e9%80%9a%e8%bf%87%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e6%bf%80%e5%8a%b1%e7%9a%84llm%e6%8e%a8%e7%90%86","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=64573","title":{"rendered":"Nature\u5c01\u9762\u6587\u7ae0: DeepSeek-R1 \u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u6fc0\u52b1\u7684LLM\u63a8\u7406"},"content":{"rendered":"<div class='wxsyncmain'>\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 17px;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none\" data-pm-slice=\"0 0 []\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section data-mpa-powered-by=\"yiban.io\" data-pm-slice=\"0 0 []\" style=\"margin: 0px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;font-family: PingFangSC-light;font-size: 15px\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;width: 661px;vertical-align: top\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px 8px;padding: 0px;max-width: 100%;clear: both;min-height: 1em;text-align: left\"><img class=\"rich_pages wxw-img\" data-aiimageid=\"36121418193747968\" data-backh=\"325\" data-backw=\"578\" data-croporisrc=\"http:\/\/mmbiz.qpic.cn\/mmbiz_png\/wibWV1DB7tWLs8IKWSEh3VEs2oqhtxJ32yAk6T43wcibQLwUPniagHAJtiaabppPxn4HCoHdVXqW7FNBAlCewuQfRQ\/0?wx_fmt=png\" data-cropselx2=\"578\" data-cropsely2=\"325\" data-imgfileid=\"100236307\" data-ratio=\"0.5625\" data-s=\"300,640\" data-w=\"1024\" style=\"margin-top: 0px;margin-bottom: 0px;padding: 0px;max-width: 100%;vertical-align: bottom;background-color: transparent;width: 100%\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-60094c3cc023a64f3efe35aff1ddcf84.png\"><\/section>\n<\/section>\n<section style=\"margin: 10px 0px 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;width: 661px\">\n<section style=\"margin: 0px;padding: 0px 3px 0px 0px;max-width: 100%;float: left;line-height: 1\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;text-align: left\">\n<section style=\"margin: 0px;padding: 0px 0px 0px 10px;max-width: 100%;width: auto;vertical-align: top;min-width: 10%;height: auto;line-height: 0\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;width: 25px;height: 10px;vertical-align: top;overflow: hidden;line-height: 0;border-style: solid solid none;border-width: 3px 3px 2px;border-radius: 0px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/section>\n<\/section>\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;text-align: justify;font-size: 16px;line-height: 1.5\">\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: both;min-height: 1em\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">\u5bfc\u8bed<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;width: 24px;height: 10px;vertical-align: top;overflow: hidden;line-height: 0;border-style: solid solid none;border-width: 3px 3px 2px;border-radius: 0px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin: 0px;padding: 0px 4px;max-width: 100%;clear: right;min-height: 4.5em !important\">\n<section style=\"margin: 5px 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px 8px;max-width: 100%;font-size: 13px;line-height: 2;letter-spacing: 0.544px\">\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: none;min-height: 1em\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><\/strong><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span>\u5c31\u5728\u521a\u521a\uff0c\u4e00\u7bc7\u9898\u76ee\u4e3a\uff1a\u300aDeepSeek-R1 \u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u6fc0\u52b1\u5927\u578b\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u63a8\u7406\u300b\uff08DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning\uff09\u7684\u6587\u7ae0\u767b\u4e0a\u4e86Nature\uff0c\u800c\u4e14\u8fd8\u662f\u5c01\u9762\u6587\u7ae0\u3002<\/span><\/span><\/strong><\/p>\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: none;min-height: 1em\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/span><\/strong><\/p>\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: none;min-height: 1em\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;font-size: 13px;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;font-family: PingFangSC-light;font-size: 13px;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px;text-align: start\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">\u5173\u952e\u8bcd\uff1a<\/span><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;font-size: 13px;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;font-family: PingFangSC-light;font-size: 13px;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px;text-align: start\"><span data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;p&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;-webkit-tap-highlight-color: transparent;margin: 0px;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;clear: none;min-height: 1em;visibility: visible;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;strong&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;-webkit-tap-highlight-color: transparent;margin: 0px;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;visibility: visible;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;span&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;-webkit-tap-highlight-color: transparent;margin: 0px;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-size: 13px;letter-spacing: 0.544px;visibility: visible;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;strong&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;-webkit-tap-highlight-color: transparent;margin: 0px;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;color: rgb(63, 63, 63);font-family: PingFangSC-light;font-size: 13px;letter-spacing: 0.544px;white-space: normal;background-color: rgb(246, 246, 246);visibility: visible;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;strong&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;-webkit-tap-highlight-color: transparent;margin: 0px;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;color: rgb(0, 0, 0);font-family: &quot;Helvetica Neue&quot;;letter-spacing: 0.544px;text-align: start;visibility: visible;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\">LLM\u63a8\u7406\u3001\u5f3a\u5316\u5b66\u4e60\u3001\u89c4\u5219\u5316\u5956\u52b1<\/span><strong style=\"margin: 0px;padding: 0px;max-width: 100%;font-family: PingFangSC-light;font-size: 12px;font-style: normal;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none\" data-pm-slice=\"0 0 []\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;text-align: start\"><span><br \/><\/span><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin: 10px 0px 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;width: 661px\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;clear: both;line-height: 0\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin: 10px 0px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-indent: 0px;text-transform: none;font-family: PingFangSC-light;font-size: 15px;text-align: center\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;vertical-align: middle;line-height: 0;text-align: right\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><img alt=\"\u56fe\u7247\" class=\"rich_pages wxw-img\" data-ratio=\"0.07314814814814814\" data-type=\"png\" data-w=\"1080\" style=\"margin: 0px;padding: 0px;max-width: 100%;vertical-align: middle;height: auto !important;width: 677px !important\" src=\"\"><img class=\"rich_pages wxw-img\" \/><img class=\"rich_pages wxw-img\" \/><img class=\"rich_pages wxw-img\" \/><img class=\"rich_pages wxw-img\" \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section data-mpa-powered-by=\"yiban.io\" data-pm-slice=\"3 3 []\">\n<section style=\"text-align: right\"><span style=\"font-size: 13px;, Arial, sans-serif;letter-spacing: 0.544px\"><span style=\"font-weight: bold\">JK\u4e28\u4f5c\u8005<\/span><\/span><span style=\"font-size: 13px;, Arial, sans-serif;letter-spacing: 0.544px\" data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;powered-by&quot;:&quot;xiumi.us&quot;,&quot;data-mpa-powered-by&quot;:&quot;yiban.io&quot;,&quot;data-pm-slice&quot;:&quot;0 0 []&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;text-align: right;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\"><br \/><\/span><\/section>\n<section style=\"text-align: right\"><span style=\"text-align: right;font-size: 13px;, Arial, sans-serif;letter-spacing: 0.544px\" data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;powered-by&quot;:&quot;xiumi.us&quot;,&quot;data-mpa-powered-by&quot;:&quot;yiban.io&quot;,&quot;data-pm-slice&quot;:&quot;0 0 []&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;text-align: right;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\"><span style=\"font-weight: bold\">\u5f20\u5029<\/span><\/span><span style=\"font-size: 13px;, Arial, sans-serif;letter-spacing: 0.544px\" data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;powered-by&quot;:&quot;xiumi.us&quot;,&quot;data-mpa-powered-by&quot;:&quot;yiban.io&quot;,&quot;data-pm-slice&quot;:&quot;0 0 []&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;text-align: right;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\"><span style=\"font-weight: bold\">\u4e28\u7f16\u8f91<\/span><\/span><\/section>\n<section style=\"text-align: right\"><span style=\"font-size: 13px;, Arial, sans-serif;letter-spacing: 0.544px\" data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;powered-by&quot;:&quot;xiumi.us&quot;,&quot;data-mpa-powered-by&quot;:&quot;yiban.io&quot;,&quot;data-pm-slice&quot;:&quot;0 0 []&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;text-align: right;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\"><br \/><\/span><\/section>\n<\/section>\n<section style=\"margin-left: 8px;margin-right: 8px\"><img class=\"rich_pages wxw-img\" data-backh=\"285\" data-backw=\"562\" data-height=\"548\" data-imgfileid=\"100236272\" data-ratio=\"0.5074074074074074\" data-type=\"png\" data-w=\"1080\" data-width=\"1080\" style=\"width:100%\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-ff7e9a9ea7609445df4f432b9f401d1e.png\"><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u8fd9\u7bc7\u6587\u7ae0\u7684\u4f5c\u8005\u4eba\u6570\uff0c\u7adf\u7136\u8fbe\u5230\u4e86\u5c06\u8fd1200\u4eba\uff0c\u771f\u53ef\u8c13\u96c6\u4f53\u667a\u6167\u4e4b\u4f5c\u3002\u5728\u8fd9\u4e2a\u957f\u957f\u7684\u4f5c\u8005\u5217\u8868\u4e2d\uff0c\u4f60\u80fd\u627e\u5230\u6881\u6587\u950b\u7684\u540d\u5b57\u5417\uff1f<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px\"><img class=\"rich_pages wxw-img\" data-backh=\"236\" data-backw=\"562\" data-height=\"638\" data-imgfileid=\"100236269\" data-ratio=\"0.4203703703703704\" data-type=\"png\" data-w=\"1080\" data-width=\"1518\" style=\"width:100%\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-cc9e4297906ba17bce83dea9787c2785.png\"><\/section>\n<section style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><img class=\"rich_pages wxw-img\" data-ratio=\"1.0537037037037038\" data-type=\"png\" data-w=\"1080\" style=\"width: 100%\" data-width=\"1522\" data-height=\"1604\" data-backw=\"562\" data-backh=\"592\" data-imgfileid=\"100236273\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-d4cbc09cf61de6b608b6647631bbe658.png\" \/><\/section>\n<blockquote>\n<p style=\"text-align: left\"><span>\u8bba\u6587\u9898\u76ee\uff1aDeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning<\/span><\/p>\n<p style=\"text-align: left\"><span>\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/www.nature.com\/articles\/s41586-025-09422-z<\/span><\/p>\n<\/blockquote>\n<section style=\"margin-bottom: 0px;text-align: left\"><span style=\"font-size: 15px\"><span><br \/><\/span><\/span><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u4e86\u66f4\u597d\u5730\u4e86\u89e3\u672c\u6587\u5185\u5bb9\uff0c\u6211\u4eec\u7ffb\u8bd1\u4e86\u5168\u6587\u6587\u7ae0\uff0c\u5e76\u914d\u4ee5\u5173\u952e\u6982\u5ff5\u7684\u8bf4\u660e\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/p>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u6458\u8981<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u901a\u7528\u63a8\u7406\u4e00\u76f4\u662f\u4eba\u5de5\u667a\u80fd\uff08AI\uff09\u9886\u57df\u957f\u671f\u800c\u8270\u5de8\u7684\u6311\u6218\u3002\u8fd1\u5e74\u6765\uff0c\u4ee5\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u548c\u601d\u7ef4\u94fe\uff08CoT\uff09\u63d0\u793a\u4e3a\u4ee3\u8868\u7684\u7a81\u7834\uff0c\u5728\u57fa\u7840\u63a8\u7406\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u663e\u8457\u6210\u529f\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u6210\u529f\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u4f9d\u8d56\u4e8e\u5927\u91cf\u4eba\u5de5\u6807\u6ce8\u7684\u793a\u8303\u6570\u636e\uff0c\u800c\u4e14\u73b0\u6709\u6a21\u578b\u7684\u80fd\u529b\u5bf9\u4e8e\u66f4\u590d\u6742\u7684\u95ee\u9898\u4ecd\u663e\u4e0d\u8db3\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u672c\u6587\u8868\u660e\uff0c\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u7eaf\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u6765\u6fc0\u52b1LLM\u7684\u63a8\u7406\u80fd\u529b\uff0c\u4ece\u800c\u4ea7\u751f\u65e0\u9700\u4eba\u5de5\u6807\u6ce8\u7684\u63a8\u7406\u8f68\u8ff9\u3002\u672c\u6587\u6240\u63d0\u51fa\u7684RL\u6846\u67b6\u4fc3\u8fdb\u4e86\u9ad8\u7ea7\u63a8\u7406\u6a21\u5f0f\u7684\u6d8c\u73b0\u5f0f\u53d1\u5c55\uff0c\u4f8b\u5982\u81ea\u6211\u53cd\u601d\uff08Self reflection\uff09\u3001\u9a8c\u8bc1\u4ee5\u53ca\u52a8\u6001\u7b56\u7565\u9002\u5e94\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u56e0\u6b64\uff0c\u8bad\u7ec3\u540e\u7684\u6a21\u578b\u5728\u6570\u5b66\u3001\u7f16\u7a0b\u7ade\u8d5b\u548cSTEM\u9886\u57df\u7b49\u53ef\u9a8c\u8bc1\u7684\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u66f4\u4f18\u7684\u6027\u80fd\uff0c\u8d85\u8d8a\u4e86\u901a\u8fc7\u5e38\u89c4\u7684\u57fa\u4e8e\u4eba\u5de5\u793a\u8303\u7684\u76d1\u7763\u5b66\u4e60\u8bad\u7ec3\u5f97\u5230\u7684\u540c\u7c7b\u6a21\u578b\u3002\u6b64\u5916\uff0c\u8fd9\u4e9b\u5927\u578b\u6a21\u578b\u6240\u8868\u73b0\u51fa\u7684\u6d8c\u73b0\u63a8\u7406\u6a21\u5f0f\u8fd8\u53ef\u4ee5\u7cfb\u7edf\u6027\u5730\u7528\u4e8e\u6307\u5bfc\u5e76\u63d0\u5347\u5c0f\u578b\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u3002<\/span><\/span><\/p>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u5f15\u8a00<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u63a8\u7406\u80fd\u529b\u4f5c\u4e3a\u4eba\u7c7b\u667a\u80fd\u7684\u57fa\u77f3\uff0c\u4f7f\u6211\u4eec\u80fd\u591f\u5b8c\u6210\u4ece\u6570\u5b66\u95ee\u9898\u6c42\u89e3\u5230\u903b\u8f91\u63a8\u65ad\u4e0e\u7f16\u7a0b\u7b49\u590d\u6742\u7684\u8ba4\u77e5\u4efb\u52a1\u3002\u4eba\u5de5\u667a\u80fd\u7684\u6700\u65b0\u8fdb\u5c55\u8868\u660e\uff0c\u5f53\u89c4\u6a21\u8db3\u591f\u5927\u65f6\uff0cLLM \u4f1a\u5448\u73b0\u51fa\u6d8c\u73b0\u884c\u4e3a\uff0c\u5176\u4e2d\u5305\u62ec\u63a8\u7406\u80fd\u529b\u3002\u7136\u800c\uff0c\u5728\u9884\u8bad\u7ec3\u9636\u6bb5\u83b7\u5f97\u8fd9\u79cd\u80fd\u529b\u901a\u5e38\u9700\u8981\u8017\u8d39\u5927\u91cf\u7684\u8ba1\u7b97\u8d44\u6e90\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e0e\u6b64\u540c\u65f6\uff0c\u4e00\u6761\u4e92\u8865\u7684\u7814\u7a76\u65b9\u5411\u8868\u660e\uff0c\u53ef\u4ee5\u901a\u8fc7 CoT\uff08\u601d\u7ef4\u94fe\uff09\u63d0\u793a\u6709\u6548\u589e\u5f3a LLM\u3002\u8be5\u6280\u672f\u8981\u4e48\u63d0\u4f9b\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u5c11\u91cf\u793a\u4f8b\uff0c\u8981\u4e48\u4f7f\u7528\u8bf8\u5982\u201cLet\u2019s think step by step\uff08\u8ba9\u6211\u4eec\u4e00\u6b65\u4e00\u6b65\u60f3\uff09\u201d\u8fd9\u7c7b\u6781\u7b80\u63d0\u793a\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u751f\u6210\u4e2d\u95f4\u63a8\u7406\u6b65\u9aa4\uff0c\u4ece\u800c\u663e\u8457\u63d0\u5347\u5176\u5728\u590d\u6742\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u3002\u540c\u6837\u5730\uff0c\u5728\u540e\u8bad\u7ec3\u9636\u6bb5\u8ba9\u6a21\u578b\u5b66\u4e60\u9ad8\u8d28\u91cf\u7684\u591a\u6b65\u63a8\u7406\u8f68\u8ff9\u65f6\uff0c\u4e5f\u89c2\u5bdf\u5230\u4e86\u8fdb\u4e00\u6b65\u7684\u6027\u80fd\u63d0\u5347\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5c3d\u7ba1\u8fd9\u4e9b\u65b9\u6cd5\u884c\u4e4b\u6709\u6548\uff0c\u4f46\u4e5f\u5b58\u5728\u663e\u8457\u5c40\u9650\u3002\u5b83\u4eec\u5bf9\u4eba\u5de5\u6807\u6ce8\u63a8\u7406\u8f68\u8ff9\u7684\u4f9d\u8d56\u4f1a\u964d\u4f4e\u53ef\u6269\u5c55\u6027\u5e76\u5f15\u5165\u8ba4\u77e5\u504f\u5dee\u3002\u6b64\u5916\uff0c\u5c06\u6a21\u578b\u9650\u5b9a\u4e3a\u590d\u73b0\u4eba\u7c7b\u601d\u7ef4\u8fc7\u7a0b\uff0c\u5929\u7136\u4f1a\u7528\u4eba\u7c7b\u63d0\u4f9b\u7684\u793a\u4f8b\u4e3a\u5176\u80fd\u529b\u8bbe\u9650\uff0c\u4ece\u800c\u963b\u788d\u5bf9\u66f4\u4f18\u3001\u975e\u4eba\u7c7b\u5f0f\u63a8\u7406\u8def\u5f84\u7684\u63a2\u7d22\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u6211\u4eec\u5e0c\u671b\u5728\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u6846\u67b6\u4e2d\uff0c\u901a\u8fc7\u81ea\u6211\u8fdb\u5316\u6765\u6316\u6398 LLM \u5728\u63a8\u7406\u65b9\u9762\u7684\u6f5c\u529b\uff0c\u5e76\u5c3d\u91cf\u51cf\u5c11\u5bf9\u4eba\u5de5\u6807\u6ce8\u7684\u4f9d\u8d56\u3002\u5177\u4f53\u800c\u8a00\uff0c\u6211\u4eec\u57fa\u4e8e DeepSeek-V3-Base\uff0c\u91c7\u7528\u7ec4\u76f8\u5bf9\u7b56\u7565\u4f18\u5316\uff08Group Relative Policy Optimization\uff0c\u7b80\u79f0GRPO\uff09\u4f5c\u4e3a\u6211\u4eec\u7684 RL \u6846\u67b6\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5956\u52b1\u4fe1\u53f7\u4ec5\u4f9d\u636e\u6700\u7ec8\u9884\u6d4b\u4e0e\u771f\u5b9e\u7b54\u6848\u7684\u6b63\u786e\u6027\u6765\u8bbe\u5b9a\uff0c\u800c\u4e0d\u5bf9\u63a8\u7406\u8fc7\u7a0b\u672c\u8eab\u65bd\u52a0\u4efb\u4f55\u7ea6\u675f\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u6211\u4eec\u7ed5\u8fc7\u4e86\u5e38\u89c4\u7684\u5728 RL \u8bad\u7ec3\u524d\u8fdb\u884c\u7684\u76d1\u7763\u5fae\u8c03\uff08 Supervised Fine-Tuning\uff0c\u7b80\u79f0SFT\uff09\u9636\u6bb5\u3002\u4e4b\u6240\u4ee5\u505a\u51fa\u8fd9\u4e00\u8bbe\u8ba1\u9009\u62e9\uff0c\u662f\u56e0\u4e3a\u6211\u4eec\u5047\u8bbe\u7531\u4eba\u5b9a\u4e49\u7684\u63a8\u7406\u6a21\u5f0f\u53ef\u80fd\u4f1a\u9650\u5236\u6a21\u578b\u7684\u63a2\u7d22\uff1b\u76f8\u53cd\uff0c\u4e0d\u53d7\u9650\u5236\u7684 RL \u8bad\u7ec3\u66f4\u80fd\u6fc0\u52b1 LLM \u6d8c\u73b0\u65b0\u7684\u63a8\u7406\u80fd\u529b\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u901a\u8fc7\u8fd9\u4e00\u8fc7\u7a0b\uff08\u8be6\u7ec6\u5185\u5bb9\u89c1\u4e0b\u4e00\u8282\uff09\uff0c\u6211\u4eec\u7684\u6a21\u578b\uff08\u79f0\u4e3a DeepSeek-R1-Zero\uff09\u81ea\u7136\u5730\u53d1\u5c55\u51fa\u4e86\u591a\u6837\u800c\u590d\u6742\u7684\u63a8\u7406\u884c\u4e3a\u3002\u4e3a\u89e3\u51b3\u63a8\u7406\u95ee\u9898\uff0c\u6a21\u578b\u8868\u73b0\u51fa\u751f\u6210\u66f4\u957f\u56de\u7b54\u7684\u503e\u5411\uff0c\u5e76\u5728\u6bcf\u4e2a\u56de\u7b54\u4e2d\u878d\u5165\u9a8c\u8bc1\u3001\u81ea\u6211\u53cd\u601d\u4ee5\u53ca\u5bf9\u66ff\u4ee3\u65b9\u6848\u7684\u63a2\u7d22\u3002\u5c3d\u7ba1\u6211\u4eec\u5e76\u672a\u660e\u786e\u6559\u6388\u6a21\u578b\u5982\u4f55\u8fdb\u884c\u63a8\u7406\uff0c\u5b83\u4ecd\u7136\u901a\u8fc7 RL \u6210\u529f\u5b66\u4e60\u5230\u4e86\u66f4\u4f18\u7684\u63a8\u7406\u7b56\u7565\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5c3d\u7ba1 DeepSeek\u2011R1\u2011Zero \u5c55\u793a\u4e86\u51fa\u8272\u7684\u63a8\u7406\u80fd\u529b\uff0c\u4f46\u5b83\u4e5f\u9762\u4e34\u4e00\u4e9b\u6311\u6218\uff0c\u4f8b\u5982\u53ef\u8bfb\u6027\u8f83\u5dee\u4ee5\u53ca\u8bed\u8a00\u6df7\u7528\uff0c\u5076\u5c14\u4f1a\u5728\u4e00\u6b21 CoT \u56de\u7b54\u4e2d\u540c\u65f6\u51fa\u73b0\u82f1\u6587\u548c\u4e2d\u6587\u3002\u6b64\u5916\uff0cDeepSeek\u2011R1\u2011Zero \u7684\u57fa\u4e8e\u89c4\u5219\u7684\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u9636\u6bb5\u8fc7\u4e8e\u4e13\u6ce8\u4e8e\u63a8\u7406\u4efb\u52a1\uff0c\u5bfc\u81f4\u5728\u5199\u4f5c\u548c\u5f00\u653e\u9886\u57df\u95ee\u7b54\u7b49\u66f4\u5e7f\u6cdb\u9886\u57df\u7684\u8868\u73b0\u53d7\u9650\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 DeepSeek\u2011R1\uff0c\u8fd9\u662f\u4e00\u79cd\u901a\u8fc7\u591a\u9636\u6bb5\u5b66\u4e60\u6846\u67b6\u8bad\u7ec3\u7684\u6a21\u578b\uff0c\u8be5\u6846\u67b6\u7ed3\u5408\u4e86\u62d2\u7edd\u91c7\u6837\u3001\u5f3a\u5316\u5b66\u4e60\u548c\u76d1\u7763\u5fae\u8c03\uff0c\u5177\u4f53\u7ec6\u8282\u89c1\u201cDeepSeek\u2011R1\u201d\u4e00\u8282\u3002\u8be5\u8bad\u7ec3\u6d41\u7a0b\u4f7f DeepSeek\u2011R1 \u80fd\u591f\u7ee7\u627f\u5176\u524d\u8eab DeepSeek\u2011R1\u2011Zero \u7684\u63a8\u7406\u80fd\u529b\uff0c\u540c\u65f6\u901a\u8fc7\u8fdb\u4e00\u6b65\u7684\u975e\u63a8\u7406\u6570\u636e\uff0c\u4f7f\u6a21\u578b\u884c\u4e3a\u4e0e\u4eba\u7c7b\u504f\u597d\u4fdd\u6301\u4e00\u81f4\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u4ee5\u66f4\u4f4e\u7684\u80fd\u8017\u8ba9\u66f4\u591a\u4eba\u80fd\u591f\u4f7f\u7528\u5f3a\u5927\u7684 AI\uff0c\u6211\u4eec\u84b8\u998f\u4e86\u82e5\u5e72\u66f4\u5c0f\u7684\u6a21\u578b\u5e76\u516c\u5f00\u53d1\u5e03\u3002\u8fd9\u4e9b\u84b8\u998f\u6a21\u578b\u8868\u73b0\u51fa\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u8d85\u8fc7\u4e86\u5b83\u4eec\u6700\u521d\u4ec5\u505a\u6307\u4ee4\u5fae\u8c03\u7684\u5bf9\u5e94\u7248\u672c\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u76f8\u4fe1\uff0c\u8fd9\u4e9b\u6307\u4ee4\u5fae\u8c03\u540e\u7684\u84b8\u998f\u7248\u672c\u4e5f\u5c06\u6781\u5927\u5730\u60e0\u53ca\u79d1\u7814\u793e\u533a\uff0c\u4e3a\u7406\u89e3\u957f\u94fe\u5f0f\u601d\u7ef4\u63a8\u7406\u6a21\u578b\u7684\u673a\u5236\u3001\u4ee5\u53ca\u63a8\u52a8\u66f4\u5f3a\u5927\u63a8\u7406\u6a21\u578b\u7684\u53d1\u5c55\u63d0\u4f9b\u5b9d\u8d35\u8d44\u6e90\u3002\u6211\u4eec\u5c06\u6309\u7167\u201c\u4ee3\u7801\u53ef\u7528\u6027\u201d\u4e00\u8282\u4e2d\u7684\u8bf4\u660e\uff0c\u5411\u516c\u4f17\u53d1\u5e03 DeepSeek\u2011R1\u2011Zero\u3001DeepSeek\u2011R1\u3001\u6570\u636e\u6837\u672c\u4ee5\u53ca\u84b8\u998f\u6a21\u578b\u3002<\/span><\/span><\/p>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><span><span style=\"font-size: 15px\">DeepSeek\u2011R1\u2011Zero<\/span><\/span><\/span><\/strong><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u5728\u5927\u89c4\u6a21\u4e0a\u5b9e\u73b0\u5bf9 DeepSeek\u2011R1\u2011Zero \u7684\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\uff0c\u6211\u4eec\u91c7\u7528\u4e86\u9ad8\u6548\u7684 RL \u6d41\u6c34\u7ebf\u3002\u5177\u4f53\u800c\u8a00\uff0c\u6211\u4eec\u4f7f\u7528 GRPO \u4f5c\u4e3a\u6211\u4eec\u7684 RL \u7b97\u6cd5\uff0c\u8be6\u89c1\u201cMethods\u201d\u90e8\u5206\u7684\u201cGRPO\u201d\u3002\u6b64\u5916\uff0c\u6211\u4eec\u91c7\u7528\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\u7cfb\u7edf\u6765\u8ba1\u7b97\u201c\u51c6\u786e\u6027\u5956\u52b1\u201d\u548c\u201c\u683c\u5f0f\u5956\u52b1\u201d\uff0c\u5176\u8be6\u7ec6\u65b9\u6cd5\u5728\u201cMethods\u201d\u90e8\u5206\u7684\u201cReward design\u201d\u4e2d\u7ed9\u51fa\u3002\u5e76\u4e14\uff0c\u6211\u4eec\u7684\u9ad8\u6027\u80fd RL \u57fa\u7840\u8bbe\u65bd\u5728\u8865\u5145\u6750\u6599\u7684\u7b2c 2.1 \u8282\u4e2d\u8fdb\u884c\u4e86\u63cf\u8ff0\uff0c\u4ee5\u786e\u4fdd\u8bad\u7ec3\u5177\u6709\u53ef\u6269\u5c55\u6027\u4e0e\u9ad8\u6548\u7387\u3002<\/span><\/span><\/p>\n<section style=\"margin-bottom: 0px;letter-spacing: 0.578px\">\n<section>\n<section>\n<section data-tools=\"135\u7f16\u8f91\u5668\" data-id=\"88077\">\n<section style=\"margin-right: auto;margin-left: auto;padding-top: 15px\">\n<section style=\"margin-bottom: -14px;justify-content: flex-end\">\n<section style=\"margin-top: -14px;width: 5px;height: 5px;border-right: none;border-top: none;overflow: hidden\"><span><br \/><\/span><\/section>\n<\/section>\n<section data-width=\"100%\" style=\"margin-right: auto;margin-left: auto;width: 578px;border-width: 1px;border-style: solid\">\n<section style=\"padding: 3px\">\n<section style=\"padding: 10px;border-width: 1px;border-style: solid\">\n<section style=\"font-size: 15px;letter-spacing: 1.5px;line-height: 25px\">\n<p style=\"vertical-align: inherit\"><span style=\"letter-spacing: 0.578px\"><span>\u76f8\u5173\u8bcd\u6761<a class=\"normal_text_link\" target=\"_blank\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247707721&amp;idx=1&amp;sn=52c0370762a13598d3c76cf090ecfac3&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"font-size: 15px\">DeepSeek-R1-Zero<\/span><\/a><\/span><\/span><\/p>\n<p style=\"vertical-align: inherit;text-align: left\"><span style=\"letter-spacing: 0.578px\"><span><span style=\"font-size: 15px\">https:\/\/wiki.swarma.org\/index.php\/DeepSeek-R1-Zero<\/span><\/span><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin-top: -5px;width: 5px;height: 5px;border-left: none;border-bottom: none;overflow: hidden\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<p style=\"vertical-align: inherit\"><span><br \/><\/span><\/p>\n<section data-tools=\"135\u7f16\u8f91\u5668\" data-id=\"145047\" style=\"margin-bottom: 0px\">\n<section style=\"margin: 10px auto\">\n<section style=\"border-width: 1px;border-style: solid;padding: 10px 15px\">\n<section data-autoskip=\"1\" style=\"line-height: 1.75em;letter-spacing: 1.5px;font-size: 14px\">\n<p><span><span style=\"font-size: 15px;font-weight: bold\">\u4ec0\u4e48\u662fGRPO\uff1f<\/span><\/span><\/p>\n<p><span><br \/><\/span><\/p>\n<p style=\"margin-left: 0px;margin-right: 0px\"><span><span style=\"font-size: 15px\">DeepSeek-R1-Zero\u6a21\u578b\u4f7f\u7528\u4e86\u4e00\u79cd\u65b0\u578b\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\uff0c\u5373\u7ec4\u76f8\u5bf9\u7b56\u7565\u4f18\u5316\u7b97\u6cd5\uff08Group Relative Policy Optimization, \u7b80\u79f0GRPO\uff09\uff0c\u8be5\u7b97\u6cd5\u662f\u5bf9\u7ecf\u5178\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u8fd1\u7aef\u7b56\u7565\u4f18\u5316\uff08Proximal Policy Optimization, PPO\uff09\u7684\u6539\u8fdb\u3002\u4e8c\u8005\u6700\u5927\u7684\u533a\u522b\u5728\u4e8e\uff1aPPO\u7b97\u6cd5\u9700\u8981\u5b66\u4e60\u4f18\u5316\u4e00\u4e2a\u4ef7\u503c\u7f51\u7edc\uff08Value Network\uff09\u6765\u5145\u5f53\u8bc4\u5224\u8005\u7684\u51b3\u7b56\u6765\u5bf9\u751f\u6210\u7684\u7b56\u7565\u8fdb\u884c\u8bc4\u4f30\u3002\u4f46\u662f\uff0c\u8bad\u7ec3\u4e00\u4e2a\u4ef7\u503c\u7f51\u7edc\u9700\u8981\u8017\u8d39\u5927\u91cf\u5185\u5b58\uff0c\u56e0\u6b64GRPO\u7b97\u6cd5\u53bb\u6389\u4e86\u4ef7\u503c\u7f51\u7edc\uff0c\u5e76\u901a\u8fc7\u5f15\u5165\u7ec4\u76f8\u5bf9\u5956\u52b1\uff08Group Relative Reward\uff09\u6765\u4f18\u5316\u7b56\u7565\uff0c\u4f7f\u5f97\u7b56\u7565\u5728\u7ec4\u5185\u76f8\u5bf9\u8868\u73b0\u66f4\u597d\uff0c\u800c\u4e0d\u662f\u4ec5\u4ec5\u4f9d\u8d56\u4e8e\u7edd\u5bf9\u5956\u52b1\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 0px;margin-right: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 0px;margin-right: 0px\"><span><span style=\"font-size: 15px\">\u4e3a\u4e86\u66f4\u597d\u5730\u7406\u89e3GRPO\uff0c\u6211\u4eec\u7ed8\u5236\u4e86\u4f20\u7edf\u7684PPO\u7b97\u6cd5\u4e0eGRPO\u7b97\u6cd5\u7684\u5bf9\u6bd4\u56fe\uff1a<\/span><\/span><\/p>\n<section style=\"margin-left: 0px;margin-right: 0px\"><img class=\"rich_pages wxw-img\" data-backh=\"247\" data-backw=\"546\" data-height=\"620\" data-imgfileid=\"100236271\" data-ratio=\"0.4527777777777778\" data-type=\"png\" data-w=\"1080\" data-width=\"1368\" style=\"width:100%\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-0f2ba820fa3753d2274f2299a78a3351.png\"><\/section>\n<p style=\"margin-left: 0px;margin-right: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 0px;margin-right: 0px;margin-bottom: 0px\"><span><span style=\"font-size: 15px\">\u7531\u8be5\u56fe\u53ef\u4ee5\u6e05\u695a\u5730\u770b\u51fa\uff0cGRPO\u653e\u5f03\u4e86\u4f20\u7edf\u7684PPO\u4e2d\u7684\u4ef7\u503c\u6a21\u578b\uff0c\u8f6c\u800c\u4ece\u7ec4\u5206\u6570\u4e2d\u4f30\u8ba1\u57fa\u7ebf\uff0c\u663e\u8457\u51cf\u5c11\u4e86\u8bad\u7ec3\u8d44\u6e90\u3002<\/span><\/span><\/p>\n<\/section>\n<\/section>\n<section style=\"justify-content: flex-end\">\n<section style=\"padding-right: 5px;padding-left: 5px;margin-top: -8px;margin-right: 15px\">\n<section style=\"width: 4px;height: 13px;margin-right: 2px;margin-left: 2px\"><span><br \/><\/span><\/section>\n<section style=\"width: 4px;height: 13px;margin-right: 2px;margin-left: 2px\"><span><br \/><\/span><\/section>\n<section style=\"width: 4px;height: 13px;margin-right: 2px;margin-left: 2px\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<p style=\"margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u66f4\u5177\u4f53\u5730\uff0c\u6211\u4eec\u5728 DeepSeek\u2011V3 Base \u4e0a\u5e94\u7528 RL \u6280\u672f\u6765\u8bad\u7ec3 DeepSeek\u2011R1\u2011Zero\u3002\u8bad\u7ec3\u671f\u95f4\uff0c\u6211\u4eec\u8bbe\u8ba1\u4e86\u4e00\u4e2a\u76f4\u89c2\u7684\u6a21\u677f\uff0c\u8981\u6c42 DeepSeek\u2011R1\u2011Zero \u5148\u751f\u6210\u63a8\u7406\u8fc7\u7a0b\uff0c\u518d\u7ed9\u51fa\u6700\u7ec8\u7b54\u6848\u3002\u63d0\u793a\u6a21\u677f\u5982\u4e0b\u6240\u793a\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.16666666666666666\" data-type=\"png\" data-w=\"1080\" style=\"width:100%\" data-width=\"1224\" data-height=\"204\" data-backw=\"562\" data-backh=\"94\" data-imgfileid=\"100236270\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-c52a4181504e481dc6c44144f179ceb5.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u201c\u4e00\u6bb5\u7528\u6237\u4e0e\u52a9\u624b\u4e4b\u95f4\u7684\u5bf9\u8bdd\u3002\u7528\u6237\u63d0\u51fa\u4e00\u4e2a\u95ee\u9898\uff0c\u52a9\u624b\u6765\u89e3\u51b3\u3002\u52a9\u624b\u5148\u5728\u8111\u6d77\u4e2d\u601d\u8003\u63a8\u7406\u8fc7\u7a0b\uff0c\u7136\u540e\u5411\u7528\u6237\u7ed9\u51fa\u7b54\u6848\u3002\u63a8\u7406\u8fc7\u7a0b\u548c\u7b54\u6848\u5206\u522b\u7528&nbsp;<\/span><\/span><code><span><span style=\"font-size: 15px\">&lt;think&gt;...&lt;\/think&gt;<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u548c&nbsp;<\/span><\/span><code><span><span style=\"font-size: 15px\">&lt;answer&gt;...&lt;\/answer&gt;<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u6807\u7b7e\u5305\u88f9\uff0c\u5373\uff1a<\/span><\/span><code><span><span style=\"font-size: 15px\">&lt;think&gt; reasoning process here &lt;\/think&gt;&lt;answer&gt; answer here &lt;\/answer&gt;<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u3002User:&nbsp;<\/span><\/span><code><span><span style=\"font-size: 15px\">prompt<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u3002Assistant:\u201d<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5176\u4e2d&nbsp;<\/span><\/span><code><span><span style=\"font-size: 15px\">prompt<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u4f1a\u5728\u8bad\u7ec3\u65f6\u66ff\u6362\u4e3a\u5177\u4f53\u7684\u63a8\u7406\u95ee\u9898\u3002\u6211\u4eec\u6709\u610f\u5c06\u7ea6\u675f\u9650\u5236\u5728\u8fd9\u4e00\u7ed3\u6784\u5316\u683c\u5f0f\u4e0a\uff0c\u907f\u514d\u4efb\u4f55\u4e0e\u5185\u5bb9\u76f8\u5173\u7684\u504f\u7f6e\uff0c\u4ee5\u786e\u4fdd\u6211\u4eec\u80fd\u591f\u51c6\u786e\u89c2\u5bdf\u5230\u6a21\u578b\u5728 RL \u8fc7\u7a0b\u4e2d\u7684\u81ea\u7136\u6f14\u8fdb\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u56fe 1a \u5c55\u793a\u4e86 DeepSeek\u2011R1\u2011Zero \u5728\u7f8e\u56fd\u9080\u8bf7\u6570\u5b66\u8003\u8bd5\uff08AIME\uff092024 \u57fa\u51c6\u4e0a\u7684\u8bad\u7ec3\u5168\u8fc7\u7a0b\u8868\u73b0\u6548\u679c\uff0c\u5176\u4e2d AIME 2024 \u7684\u5e73\u5747 pass@1 \u5206\u6570\u663e\u8457\u63d0\u5347\uff0c\u4ece\u521d\u59cb\u7684 15.6% \u8dc3\u5347\u81f3 77.9%\u3002\u6b64\u5916\uff0c\u501f\u52a9\u81ea\u4e00\u81f4\u6027\u89e3\u7801\uff0c\u6a21\u578b\u6027\u80fd\u8fd8\u80fd\u8fdb\u4e00\u6b65\u63d0\u5347\uff0c\u51c6\u786e\u7387\u8fbe\u5230 86.7%\u3002\u8fd9\u4e00\u8868\u73b0\u5927\u5e45\u8d85\u8fc7\u4e86\u6240\u6709 AIME \u4eba\u7c7b\u53c2\u8d5b\u8005\u7684\u5e73\u5747\u6c34\u5e73\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><img class=\"rich_pages wxw-img\" data-height=\"782\" data-imgfileid=\"100236276\" data-ratio=\"0.43425925925925923\" data-type=\"png\" data-w=\"1080\" data-width=\"1799\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-aebb80ea6be40e0abb9ba4b8c49c42f4.png\"><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 12px\">\u56fe 1\uff1aDeepSeek\u2011R1\u2011Zero \u5728\u6574\u4e2a\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u51c6\u786e\u7387\u4e0e\u8f93\u51fa\u957f\u5ea6\u3002a\uff0cDeepSeek-R1\u2011Zero \u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684 AIME \u51c6\u786e\u7387\u3002AIME \u4ee5\u6570\u5b66\u95ee\u9898\u4e3a\u8f93\u5165\u3001\u4ee5\u6570\u503c\u4e3a\u8f93\u51fa\uff0c\u5176\u793a\u4f8b\u89c1\u6269\u5c55\u6570\u636e\u8868 1\u3002pass@1 \u548c cons@16 \u7684\u8bf4\u660e\u89c1\u8865\u5145\u4fe1\u606f\u7b2c 4.1 \u8282\u3002\u57fa\u7ebf\u4e3a AIME \u6bd4\u8d5b\u4e2d\u4eba\u7c7b\u53c2\u8d5b\u8005\u53d6\u5f97\u7684\u5e73\u5747\u5206\u3002b\uff0cDeepSeek-R1\u2011Zero \u5728\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u8fc7\u7a0b\u4e2d\u4e8e\u8bad\u7ec3\u96c6\u4e0a\u7684\u5e73\u5747\u56de\u590d\u957f\u5ea6\u3002DeepSeek-R1\u2011Zero \u4f1a\u81ea\u7136\u5730\u5728\u66f4\u957f\u601d\u8003\u65f6\u95f4\u4e0b\u5b66\u4e60\u89e3\u51b3\u63a8\u7406\u4efb\u52a1\u3002\u6ce8\u610f\uff0c\u8bad\u7ec3\u6b65\uff08training step\uff09\u6307\u4e00\u6b21\u7b56\u7565\u66f4\u65b0\u64cd\u4f5c\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u9664\u6570\u5b66\u7ade\u8d5b\u5916\uff0c\u5982\u8865\u5145\u6750\u6599\u56fe 8 \u6240\u793a\uff0cDeepSeek\u2011R1\u2011Zero \u5728\u7f16\u7a0b\u7ade\u8d5b\u4ee5\u53ca\u7814\u7a76\u751f\u6c34\u5e73\u7684\u751f\u7269\u3001\u7269\u7406\u548c\u5316\u5b66\u95ee\u9898\u4e0a\u4e5f\u53d6\u5f97\u4e86\u663e\u8457\u6210\u7ee9\u3002\u8fd9\u4e9b\u7ed3\u679c\u51f8\u663e\u4e86\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u5728\u63d0\u5347\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u63a8\u7406\u80fd\u529b\u65b9\u9762\u7684\u6709\u6548\u6027\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u9664\u4e86\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u63a8\u7406\u80fd\u529b\u9010\u6b65\u589e\u5f3a\u4e4b\u5916\uff0cDeepSeek\u2011R1\u2011Zero \u8fd8\u5728\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u8bad\u7ec3\u4e2d\u5c55\u73b0\u51fa\u81ea\u6211\u8fdb\u5316\u7684\u884c\u4e3a\u3002\u5982\u56fe 1b \u6240\u793a\uff0cDeepSeek\u2011R1\u2011Zero \u7684\u601d\u8003\u65f6\u95f4\u5728\u6574\u4e2a\u8bad\u7ec3\u4e2d\u7a33\u6b65\u589e\u957f\uff0c\u8fd9\u5b8c\u5168\u7531\u5176\u5185\u5728\u9002\u5e94\u6027\u9a71\u52a8\uff0c\u800c\u975e\u5916\u90e8\u6539\u52a8\u3002\u501f\u52a9\u8f83\u957f\u7684\u601d\u7ef4\u94fe\uff08CoT\uff09\uff0c\u6a21\u578b\u6301\u7eed\u7cbe\u70bc\u5176\u63a8\u7406\uff0c\u5728\u6bcf\u6b21\u56de\u7b54\u4e2d\u751f\u6210\u6570\u767e\u5230\u6570\u5343\u4e2a token\uff0c\u4ee5\u63a2\u7d22\u5e76\u6539\u8fdb\u5176\u89e3\u51b3\u95ee\u9898\u7684\u7b56\u7565\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u601d\u8003\u65f6\u95f4\u7684\u589e\u957f\u6709\u52a9\u4e8e\u590d\u6742\u884c\u4e3a\u7684\u81ea\u4e3b\u6f14\u5316\u3002\u5177\u4f53\u800c\u8a00\uff0cDeepSeek\u2011R1\u2011Zero \u8d8a\u6765\u8d8a\u591a\u5730\u8868\u73b0\u51fa\u9ad8\u7ea7\u63a8\u7406\u7b56\u7565\uff0c\u5982\u53cd\u601d\u5f0f\u63a8\u7406\u548c\u5bf9\u66ff\u4ee3\u65b9\u6848\u7684\u7cfb\u7edf\u6027\u63a2\u7d22\uff08\u89c1\u6269\u5c55\u6570\u636e\u56fe 1a\uff09\uff0c\u8fd9\u663e\u8457\u63d0\u5347\u4e86\u5176\u5728\u6570\u5b66\u4e0e\u7f16\u7a0b\u7b49\u53ef\u9a8c\u8bc1\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0cDeepSeek\u2011R1\u2011Zero \u51fa\u73b0\u4e86\u4e00\u4e2a\u201c\u7075\u5149\u4e00\u73b0\u201d\u7684\u65f6\u523b\uff08\u89c1\u8868 1\uff09\uff1a\u5728\u53cd\u601d\u9636\u6bb5\u5bf9\u201cwait\uff08\u7b49\u5f85\uff09\u201d\u4e00\u8bcd\u7684\u4f7f\u7528\u7a81\u7136\u589e\u52a0\uff08\u89c1\u6269\u5c55\u6570\u636e\u56fe 1b\uff09\u3002\u8fd9\u4e00\u65f6\u523b\u6807\u5fd7\u7740\u63a8\u7406\u6a21\u5f0f\u7684\u663e\u8457\u53d8\u5316\uff0c\u5e76\u6e05\u6670\u5c55\u793a\u4e86 DeepSeek\u2011R1\u2011Zero \u7684\u81ea\u6211\u8fdb\u5316\u8fc7\u7a0b\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px\"><img class=\"rich_pages wxw-img\" data-height=\"808\" data-imgfileid=\"100236275\" data-ratio=\"0.8840262582056893\" data-type=\"png\" data-w=\"914\" data-width=\"914\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-8558f3e30e56878c29a940cf8e06d7e1.png\"><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">DeepSeek\u2011R1\u2011Zero \u7684\u81ea\u6211\u8fdb\u5316\u51f8\u663e\u4e86\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u7684\u529b\u91cf\u4e0e\u9b45\u529b\uff1a\u4e0e\u5176\u660e\u786e\u6559\u6a21\u578b\u5982\u4f55\u89e3\u51b3\u95ee\u9898\uff0c\u6211\u4eec\u53ea\u9700\u63d0\u4f9b\u5408\u9002\u7684\u6fc0\u52b1\uff0c\u6a21\u578b\u4fbf\u80fd\u81ea\u4e3b\u53d1\u5c55\u51fa\u5148\u8fdb\u7684\u95ee\u9898\u6c42\u89e3\u7b56\u7565\u3002\u8fd9\u63d0\u9192\u6211\u4eec\uff0cRL \u5177\u6709\u91ca\u653e\u66f4\u9ad8\u5c42\u6b21 LLM \u80fd\u529b\u7684\u6f5c\u529b\uff0c\u4e3a\u672a\u6765\u66f4\u81ea\u4e3b\u3001\u66f4\u5177\u9002\u5e94\u6027\u7684\u6a21\u578b\u94fa\u5e73\u9053\u8def\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5c3d\u7ba1 DeepSeek\u2011R1\u2011Zero \u5c55\u73b0\u51fa\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u4f46\u5b83\u4e5f\u9762\u4e34\u82e5\u5e72\u95ee\u9898\u3002\u7531\u4e8e DeepSeek\u2011V3 Base \u5728\u591a\u79cd\u8bed\u8a00\uff08\u5c24\u5176\u662f\u82f1\u6587\u548c\u4e2d\u6587\uff09\u4e0a\u8bad\u7ec3\uff0cDeepSeek\u2011R1\u2011Zero \u5728\u53ef\u8bfb\u6027\u548c\u8bed\u8a00\u6df7\u7528\u65b9\u9762\u8868\u73b0\u6b20\u4f73\u3002\u4e3a\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u6211\u4eec\u5f00\u53d1\u4e86 DeepSeek\u2011R1\uff0c\u5176\u8bad\u7ec3\u6d41\u6c34\u7ebf\u5982\u56fe 2 \u6240\u793a\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.4740740740740741\" data-type=\"png\" data-w=\"1080\" style=\"width:100%\" data-width=\"1787\" data-height=\"847\" data-backw=\"562\" data-backh=\"266\" data-imgfileid=\"100236278\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-e076275df76623724672f09a8bd54427.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\" data-mpa-action-id=\"mfokwvy81iem\" data-pm-slice=\"0 0 []\"><span>\u56fe 2\uff1aDeepSeek\u2011R1 \u7684\u591a\u9636\u6bb5\u6d41\u6c34\u7ebf\u3002\u5173\u4e8e DeepSeek\u2011V3 Base \u548c DeepSeek\u2011V3 \u7684\u8be6\u7ec6\u80cc\u666f\u89c1\u8865\u5145\u4fe1\u606f\u7b2c 1.1 \u8282\u3002\u6a21\u578b DeepSeek\u2011R1 Dev1\u3001Dev2 \u548c Dev3 \u8868\u793a\u8be5\u6d41\u6c34\u7ebf\u4e2d\u7684\u4e2d\u95f4\u68c0\u67e5\u70b9\u3002<\/span><\/span><\/p>\n<ul style=\"margin-left: 8px;margin-right: 8px\" class=\"list-paddingleft-1\">\n<li style=\"text-align: left\">\n<p style=\"margin-bottom: 8px\"><span><span style=\"font-size: 15px\">\u5728\u521d\u59cb\u9636\u6bb5\uff0c\u6211\u4eec\u6536\u96c6\u4e86\u5927\u91cf\u5177\u5907\u5bf9\u8bdd\u5f0f\u3001\u7b26\u5408\u4eba\u7c7b\u601d\u7ef4\u8fc7\u7a0b\u7684\u51b7\u542f\u52a8\u6570\u636e\uff0c\u8be6\u7ec6\u5185\u5bb9\u89c1\u8865\u5145\u6750\u6599\u7b2c 2.3.2 \u8282\u3002<\/span><\/span><\/p>\n<\/li>\n<li style=\"text-align: left\">\n<p style=\"margin-bottom: 8px\"><span><span style=\"font-size: 15px\">\u968f\u540e\u8fdb\u884c RL \u8bad\u7ec3\uff0c\u5e76\u91c7\u7528\u201c\u7b2c\u4e00\u9636\u6bb5 RL \u8bad\u7ec3\u7ec6\u8282\u201d\u7684\u8d85\u53c2\u6570\u8bbe\u7f6e\uff1b\u6570\u636e\u7ec6\u8282\u89c1\u8865\u5145\u6750\u6599\u7b2c 2.3.1 \u8282\uff0c\u7528\u4e8e\u63d0\u5347\u6a21\u578b\u5728\u5bf9\u8bdd\u5f0f\u601d\u7ef4\u8fc7\u7a0b\u4e0e\u8bed\u8a00\u4e00\u81f4\u6027\u65b9\u9762\u7684\u8868\u73b0\u3002<\/span><\/span><\/p>\n<\/li>\n<li style=\"text-align: left\">\n<p style=\"margin-bottom: 8px\"><span><span style=\"font-size: 15px\">\u63a5\u7740\uff0c\u6211\u4eec\u518d\u6b21\u5e94\u7528\u62d2\u7edd\u91c7\u6837\u5e76\u8fdb\u884c\u4e00\u6b21 SFT\uff08\u76d1\u7763\u5fae\u8c03\uff09\u3002\u6b64\u9636\u6bb5\u5c06\u63a8\u7406\u4e0e\u975e\u63a8\u7406\u6570\u636e\u96c6\u4e00\u5e76\u7eb3\u5165 SFT \u8fc7\u7a0b\uff0c\u8be6\u89c1\u8865\u5145\u6750\u6599\u7b2c 2.3.3 \u8282\uff0c\u4f7f\u6a21\u578b\u4e0d\u4ec5\u80fd\u5728\u63a8\u7406\u4efb\u52a1\u4e0a\u8868\u73b0\u4f18\u5f02\uff0c\u4e5f\u80fd\u5c55\u73b0\u51fa\u9ad8\u7ea7\u5199\u4f5c\u80fd\u529b\u3002<\/span><\/span><\/p>\n<\/li>\n<li style=\"text-align: left\">\n<p><span><span style=\"font-size: 15px\">\u4e3a\u4e86\u8fdb\u4e00\u6b65\u4f7f\u6a21\u578b\u4e0e\u4eba\u7c7b\u504f\u597d\u5bf9\u9f50\uff0c\u6211\u4eec\u8bbe\u8ba1\u4e86\u7b2c\u4e8c\u9636\u6bb5\u7684 RL\uff0c\u7528\u4e8e\u589e\u5f3a\u6a21\u578b\u7684\u6709\u7528\u6027\u4e0e\u65e0\u5bb3\u6027\uff0c\u540c\u65f6\u6301\u7eed\u6253\u78e8\u5176\u63a8\u7406\u80fd\u529b\u3002\u5956\u52b1\u7ec6\u8282\u89c1 Methods \u4e2d\u7684\u201c\u5956\u52b1\u8bbe\u8ba1\u201d\u90e8\u5206\uff0c\u7b2c\u4e8c\u9636\u6bb5 RL \u7684\u8d85\u53c2\u6570\u89c1 Methods \u7684\u201c\u7b2c\u4e8c\u9636\u6bb5 RL \u8bad\u7ec3\u7ec6\u8282\u201d\u3002<\/span><\/span><\/p>\n<\/li>\n<\/ul>\n<p style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u5728\u4ee5\u4e0b\u57fa\u51c6\u4e0a\u8bc4\u4f30\u6a21\u578b\uff1aMMLU\u3001MMLU\u2011Redux\u3001MMLU\u2011Pro\u3001DROP\u3001C\u2011Eval\u3001IFEval\u3001FRAMES\u3001GPQA Diamond\u3001SimpleQA\u3001C\u2011SimpleQA\u3001CLUEWSC\u3001AlpacaEval 2.0\uff08\u53c2\u8003\u6587\u732e 22\uff09\u3001Arena\u2011Hard\u3001SWE\u2011bench Verified\u3001Aider\u2011Polyglot\u3001LiveCodeBench\uff082024\u201108\u20142025\u201101\uff09\u3001Codeforces\u3001\u4e2d\u56fd\u9ad8\u4e2d\u6570\u5b66\u7ade\u8d5b\uff08CNMO 2024\uff09\u4ee5\u53ca AIME 2024\uff08\u53c2\u8003\u6587\u732e 29\uff09\u3002\u8fd9\u4e9b\u57fa\u51c6\u7684\u8be6\u7ec6\u4fe1\u606f\u89c1\u8865\u5145\u8868 15\u201429\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.5851851851851851\" data-type=\"png\" data-w=\"1080\" style=\"width: 100%\" data-width=\"1834\" data-height=\"1074\" data-backw=\"562\" data-backh=\"329\" data-imgfileid=\"100236277\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-f74865e168675545c2b6a51fc6634b64.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\" data-mpa-action-id=\"mfokx9rp17na\" data-pm-slice=\"0 0 []\"><span>\u8868 2 \u603b\u7ed3\u4e86 DeepSeek\u2011R1 \u5728\u56fe 2 \u6240\u793a\u7684\u591a\u4e2a\u5f00\u53d1\u9636\u6bb5\u4e2d\u7684\u8868\u73b0\u3002\u5c06 DeepSeek\u2011R1\u2011Zero \u4e0e DeepSeek\u2011R1 Dev1 \u6bd4\u8f83\uff0c\u5728\u6307\u4ee4\u8ddf\u968f\u65b9\u9762\u6709\u663e\u8457\u63d0\u5347\uff0c\u8fd9\u5728 IF\u2011Eval \u548c Arena\u2011Hard \u57fa\u51c6\u4e0a\u7684\u66f4\u9ad8\u5f97\u5206\u4e2d\u5f97\u5230\u4f53\u73b0\u3002\u7136\u800c\uff0c\u7531\u4e8e\u51b7\u542f\u52a8\u6570\u636e\u96c6\u89c4\u6a21\u6709\u9650\uff0cDev1 \u7684\u63a8\u7406\u6027\u80fd\u76f8\u6bd4 DeepSeek\u2011R1\u2011Zero \u51fa\u73b0\u4e00\u5b9a\u4e0b\u964d\uff0c\u6700\u660e\u663e\u7684\u662f\u5728 AIME \u57fa\u51c6\u4e0a\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cDeepSeek\u2011R1 Dev2 \u5728\u9700\u8981\u9ad8\u7ea7\u63a8\u7406\u6280\u80fd\u7684\u57fa\u51c6\u4e0a\u8868\u73b0\u51fa\u663e\u8457\u63d0\u5347\uff0c\u5305\u62ec\u805a\u7126\u4e8e\u4ee3\u7801\u751f\u6210\u3001\u6570\u5b66\u95ee\u9898\u6c42\u89e3\u548c STEM \u76f8\u5173\u4efb\u52a1\u7684\u57fa\u51c6\u3002\u9488\u5bf9\u901a\u7528\u4efb\u52a1\u7684\u57fa\u51c6\uff08\u5982 AlpacaEval 2.0\uff09\u4ec5\u663e\u793a\u51fa\u8f83\u5c0f\u7684\u6539\u8fdb\u3002\u8fd9\u4e9b\u7ed3\u679c\u8868\u660e\uff0c\u4ee5\u63a8\u7406\u4e3a\u5bfc\u5411\u7684 RL \u5728\u5f71\u54cd\u7528\u6237\u504f\u597d\u5bfc\u5411\u57fa\u51c6\u6709\u9650\u7684\u60c5\u51b5\u4e0b\uff0c\u4ecd\u80fd\u663e\u8457\u589e\u5f3a\u63a8\u7406\u80fd\u529b\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">DeepSeek\u2011R1 Dev3 \u5c06\u63a8\u7406\u4e0e\u975e\u63a8\u7406\u6570\u636e\u96c6\u5171\u540c\u6574\u5408\u8fdb SFT \u6d41\u6c34\u7ebf\uff0c\u4ece\u800c\u63d0\u5347\u6a21\u578b\u5728\u63a8\u7406\u4e0e\u901a\u7528\u8bed\u8a00\u751f\u6210\u4efb\u52a1\u4e0a\u7684\u719f\u7ec3\u5ea6\u3002\u4e0e Dev2 \u76f8\u6bd4\uff0cDeepSeek\u2011R1 Dev3 \u5728 AlpacaEval 2.0 \u548c Aider\u2011Polyglot \u4e0a\u53d6\u5f97\u4e86\u663e\u8457\u6027\u80fd\u63d0\u5347\uff0c\u8fd9\u5f52\u56e0\u4e8e\u6211\u4eec\u5f15\u5165\u4e86\u5927\u89c4\u6a21\u975e\u63a8\u7406\u8bed\u6599\u548c\u4ee3\u7801\u5de5\u7a0b\u7c7b\u6570\u636e\u96c6\u3002\u6700\u540e\uff0c\u5728 DeepSeek\u2011R1 Dev3 \u4e0a\u4f7f\u7528\u6df7\u5408\u7684\u63a8\u7406\u805a\u7126\u6570\u636e\u4e0e\u901a\u7528\u6570\u636e\u8fdb\u884c\u5168\u9762\u7684 RL \u8bad\u7ec3\uff0c\u6211\u4eec\u5f97\u5230\u4e86\u6700\u7ec8\u7684 DeepSeek\u2011R1\u3002\u7531\u4e8e\u5728\u6b64\u524d\u9636\u6bb5\u5df2\u8fdb\u884c\u4e86\u5927\u91cf\u4ee5\u63a8\u7406\u4e3a\u5bfc\u5411\u7684 RL\uff0c\u6a21\u578b\u5728\u4ee3\u7801\u4e0e\u6570\u5b66\u57fa\u51c6\u4e0a\u4ec5\u83b7\u5f97\u4e86\u5c0f\u5e45\u6539\u8fdb\u3002\u6700\u7ec8\u7248 DeepSeek\u2011R1 \u7684\u4e3b\u8981\u63d0\u5347\u4f53\u73b0\u5728\u901a\u7528\u7684\u6307\u4ee4\u8ddf\u968f\u4e0e\u7528\u6237\u504f\u597d\u7c7b\u57fa\u51c6\u4e0a\uff0c\u5176\u4e2d AlpacaEval 2.0 \u63d0\u5347\u4e86 25%\uff0cArena\u2011Hard \u63d0\u5347\u4e86 17%\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u8fd8\u5728\u8865\u5145\u4fe1\u606f\u7b2c 4.2 \u8282\u5c06 DeepSeek\u2011R1 \u4e0e\u5176\u4ed6\u6a21\u578b\u8fdb\u884c\u5bf9\u6bd4\u3002\u6a21\u578b\u5b89\u5168\u6027\u8bc4\u4f30\u89c1\u8865\u5145\u4fe1\u606f\u7b2c 4.3 \u8282\u3002\u5168\u9762\u7684\u8bc4\u4f30\u5206\u6790\u89c1\u8865\u5145\u4fe1\u606f\u7b2c 5 \u8282\uff0c\u5176\u4e2d\u5305\u62ec\u4e0e DeepSeek\u2011V3 \u7684\u6bd4\u8f83\u3001\u5728\u65b0\u6d4b\u8bd5\u96c6\u4e0a\u7684\u6027\u80fd\u8bc4\u4f30\u3001\u6309\u7c7b\u522b\u7ec6\u5206\u7684\u6570\u5b66\u80fd\u529b\uff0c\u4ee5\u53ca\u6d4b\u8bd5\u65f6\u95f4\u7f29\u653e\u884c\u4e3a\u7684\u7814\u7a76\u3002\u8865\u5145\u4fe1\u606f\u7b2c 6 \u8282\u663e\u793a\uff0c\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\u53ef\u4ee5\u8fc1\u79fb\u5230\u66f4\u5c0f\u7684\u6a21\u578b\u4e0a\u3002<\/span><\/span><\/p>\n<section style=\"margin-bottom: 0px;letter-spacing: 0.578px\">\n<section>\n<section>\n<section data-tools=\"135\u7f16\u8f91\u5668\" data-id=\"88077\">\n<section style=\"margin-right: auto;margin-left: auto;padding-top: 15px\">\n<section style=\"margin-bottom: -14px;justify-content: flex-end\">\n<section style=\"margin-top: -14px;width: 5px;height: 5px;border-right: none;border-top: none;overflow: hidden\"><span><br \/><\/span><\/section>\n<\/section>\n<section data-width=\"100%\" style=\"margin-right: auto;margin-left: auto;width: 578px;border-width: 1px;border-style: solid\">\n<section style=\"padding: 3px\">\n<section style=\"padding: 10px;border-width: 1px;border-style: solid\">\n<section style=\"font-size: 15px;letter-spacing: 1.5px;line-height: 25px\">\n<p style=\"vertical-align: inherit\"><span style=\"letter-spacing: 0.578px\"><span>\u76f8\u5173\u9605\u8bfb\uff1a<a class=\"normal_text_link\" target=\"_blank\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247707721&amp;idx=1&amp;sn=52c0370762a13598d3c76cf090ecfac3&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"font-size: 15px\">DeepSeek-R1<\/span><\/a><\/span><\/span><\/p>\n<p style=\"vertical-align: inherit;text-align: left\"><span style=\"letter-spacing: 0.578px\"><span><span style=\"font-size: 15px\">https:\/\/wiki.swarma.org\/index.php\/DeepSeek-R1<\/span><\/span><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin-top: -5px;width: 5px;height: 5px;border-left: none;border-bottom: none;overflow: hidden\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<p style=\"vertical-align: inherit\"><span><br \/><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin-bottom: 0px;letter-spacing: 0.578px\">\n<section>\n<section>\n<p style=\"margin-right: 8px;margin-left: 8px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u4f26\u7406\u4e0e\u5b89\u5168\u58f0\u660e<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u968f\u7740 DeepSeek\u2011R1 \u63a8\u7406\u80fd\u529b\u7684\u63d0\u5347\uff0c\u6211\u4eec\u6df1\u523b\u610f\u8bc6\u5230\u6f5c\u5728\u7684\u4f26\u7406\u98ce\u9669\u3002\u6bd4\u5982\uff0cR1 \u53ef\u80fd\u906d\u53d7\u8d8a\u72f1\u653b\u51fb\uff0c\u4ece\u800c\u751f\u6210\u5371\u9669\u5185\u5bb9\uff08\u5982\u7206\u70b8\u7269\u5236\u9020\u65b9\u6848\uff09\uff1b\u800c\u589e\u5f3a\u7684\u63a8\u7406\u80fd\u529b\u4e5f\u53ef\u80fd\u8ba9\u6a21\u578b\u7ed9\u51fa\u5728\u53ef\u64cd\u4f5c\u6027\u4e0e\u53ef\u6267\u884c\u6027\u4e0a\u66f4\u5f3a\u7684\u89c4\u5212\u65b9\u6848\u3002\u6b64\u5916\uff0c\u516c\u5f00\u53d1\u5e03\u7684\u6a21\u578b\u8fd8\u5bb9\u6613\u88ab\u8fdb\u4e00\u6b65\u5fae\u8c03\uff0c\u4ece\u800c\u524a\u5f31\u5176\u56fa\u6709\u7684\u5b89\u5168\u9632\u62a4\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5728\u8865\u5145\u4fe1\u606f\u7b2c 4.3 \u8282\u4e2d\uff0c\u6211\u4eec\u4ece\u591a\u4e2a\u89d2\u5ea6\u7ed9\u51fa\u4e86\u4e00\u4efd\u5168\u9762\u7684\u5b89\u5168\u62a5\u544a\uff0c\u5305\u62ec\uff1a\u5728\u5f00\u6e90\u4e0e\u81ea\u5efa\u5b89\u5168\u8bc4\u6d4b\u57fa\u51c6\u4e0a\u7684\u8868\u73b0\u3001\u8de8\u591a\u79cd\u8bed\u8a00\u4e0e\u8d8a\u72f1\u653b\u51fb\u4e0b\u7684\u5b89\u5168\u6c34\u5e73\u7b49\u3002\u8fd9\u4e9b\u5168\u9762\u7684\u5b89\u5168\u5206\u6790\u5f97\u51fa\u7ed3\u8bba\uff1a\u4e0e\u5176\u4ed6\u5148\u8fdb\u6a21\u578b\u76f8\u6bd4\uff0cDeepSeek\u2011R1 \u6a21\u578b\u7684\u56fa\u6709\u5b89\u5168\u6c34\u5e73\u603b\u4f53\u5904\u4e8e\u4e2d\u7b49\u6c34\u5e73\uff08\u53ef\u4e0e GPT\u20114o\uff082024\u201105\u201113\uff09\u76f8\u5f53\uff09\u3002\u6b64\u5916\uff0c\u7ed3\u5408\u98ce\u9669\u63a7\u5236\u4f53\u7cfb\u540e\uff0c\u6a21\u578b\u7684\u5b89\u5168\u7b49\u7ea7\u53ef\u63d0\u5347\u5230\u66f4\u9ad8\u6807\u51c6\u3002<\/span><\/span><\/p>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u7ed3\u8bba\u3001\u5c40\u9650\u6027\u4e0e\u672a\u6765\u5de5\u4f5c<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u63d0\u51fa\u4e86 DeepSeek\u2011R1\u2011Zero \u548c DeepSeek\u2011R1\uff0c\u8fd9\u4e24\u8005\u4f9d\u8d56\u4e8e\u5927\u89c4\u6a21\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u6765\u6fc0\u53d1\u6a21\u578b\u7684\u63a8\u7406\u884c\u4e3a\u3002\u6211\u4eec\u7684\u7ed3\u679c\u8868\u660e\uff0c\u9884\u8bad\u7ec3\u68c0\u67e5\u70b9\u672c\u8eab\u5c31\u5bf9\u590d\u6742\u63a8\u7406\u4efb\u52a1\u5177\u6709\u76f8\u5f53\u5927\u7684\u52a9\u529b\u3002\u6211\u4eec\u8ba4\u4e3a\uff0c\u91ca\u653e\u8fd9\u79cd\u6f5c\u529b\u7684\u5173\u952e\u4e0d\u5728\u4e8e\u5927\u89c4\u6a21\u4eba\u5de5\u6807\u6ce8\uff0c\u800c\u5728\u4e8e\u4e3a RL \u63d0\u4f9b\u56f0\u96be\u7684\u63a8\u7406\u95ee\u9898\u3001\u53ef\u9760\u7684\u9a8c\u8bc1\u5668\u4ee5\u53ca\u5145\u8db3\u7684\u8ba1\u7b97\u8d44\u6e90\u3002\u5728 RL \u8fc7\u7a0b\u4e2d\uff0c\u8bf8\u5982\u81ea\u6211\u6821\u9a8c\u4e0e\u53cd\u601d\u7b49\u590d\u6742\u7684\u63a8\u7406\u884c\u4e3a\u4f3c\u4e4e\u90fd\u80fd\u591f\u81ea\u7136\u6d8c\u73b0\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5373\u4f7f DeepSeek\u2011R1 \u5728\u63a8\u7406\u57fa\u51c6\u4e0a\u53d6\u5f97\u4e86\u524d\u6cbf\u7ed3\u679c\uff0c\u5b83\u4ecd\u7136\u9762\u4e34\u82e5\u5e72\u80fd\u529b\u4e0a\u7684\u9650\u5236\uff0c\u5177\u4f53\u5982\u4e0b\u6240\u8ff0\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e\u5de5\u5177\u4f7f\u7528<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u76ee\u524d\uff0cDeepSeek\u2011R1 \u7684\u7ed3\u6784\u5316\u8f93\u51fa\u80fd\u529b\u76f8\u8f83\u4e8e\u73b0\u6709\u6a21\u578b\u4ecd\u4e0d\u7406\u60f3\u3002\u6b64\u5916\uff0cDeepSeek\u2011R1 \u4e0d\u80fd\u4f7f\u7528\u8bf8\u5982\u641c\u7d22\u5f15\u64ce\u4e0e\u8ba1\u7b97\u5668\u7b49\u5de5\u5177\u6765\u63d0\u5347\u8f93\u51fa\u6548\u679c\u3002\u4e0d\u8fc7\uff0c\u7531\u4e8e\u4e3a\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e\u5de5\u5177\u4f7f\u7528\u642d\u5efa\u4e00\u4e2a RL \u73af\u5883\u5e76\u4e0d\u56f0\u96be\uff0c\u6211\u4eec\u76f8\u4fe1\u8fd9\u4e00\u95ee\u9898\u5c06\u5728\u4e0b\u4e2a\u7248\u672c\u4e2d\u5f97\u5230\u89e3\u51b3\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">Token \u6548\u7387<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e0d\u540c\u4e8e\u4f20\u7edf\u7684\u6d4b\u8bd5\u65f6\u8ba1\u7b97\u6269\u5c55\u65b9\u6cd5\uff08\u5982\u591a\u6570\u6295\u7968\u6216\u8499\u7279\u5361\u7f57\u6811\u641c\u7d22 MCTS\uff09\uff0cDeepSeek\u2011R1 \u4f1a\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u6839\u636e\u95ee\u9898\u7684\u590d\u6742\u5ea6\u52a8\u6001\u5206\u914d\u8ba1\u7b97\u8d44\u6e90\u3002\u5177\u4f53\u800c\u8a00\uff0c\u5b83\u5728\u89e3\u51b3\u7b80\u5355\u4efb\u52a1\u65f6\u4f7f\u7528\u66f4\u5c11\u7684 token\uff0c\u800c\u5728\u5904\u7406\u590d\u6742\u4efb\u52a1\u65f6\u4f1a\u751f\u6210\u66f4\u591a\u7684 token\u3002\u5c3d\u7ba1\u5982\u6b64\uff0c\u5728 token\u5229\u7528\u6548\u7387\u65b9\u9762\u4ecd\u6709\u4f18\u5316\u7a7a\u95f4\uff0c\u56e0\u4e3a\u5728\u56de\u7b54\u8f83\u4e3a\u7b80\u5355\u7684\u95ee\u9898\u65f6\uff0c\u6211\u4eec\u4ecd\u53ef\u89c2\u5bdf\u5230\u8fc7\u5ea6\u63a8\u7406\uff08\u5373\u201coverthinking\u201d\uff09\u7684\u60c5\u51b5\u53d1\u751f\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u8bed\u8a00\u6df7\u7528<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">DeepSeek\u2011R1 \u76ee\u524d\u9488\u5bf9\u4e2d\u6587\u548c\u82f1\u6587\u8fdb\u884c\u4e86\u4f18\u5316\uff0c\u8fd9\u5728\u5904\u7406\u5176\u4ed6\u8bed\u8a00\u7684\u67e5\u8be2\u65f6\u53ef\u80fd\u4f1a\u5bfc\u81f4\u8bed\u8a00\u6df7\u7528\u7684\u95ee\u9898\u3002\u6bd4\u5982\uff0c\u5f53\u67e5\u8be2\u4f7f\u7528\u82f1\u8bed\u6216\u4e2d\u6587\u4ee5\u5916\u7684\u8bed\u8a00\u65f6\uff0cDeepSeek\u2011R1 \u53ef\u80fd\u4f1a\u7528\u82f1\u8bed\u8fdb\u884c\u63a8\u7406\u5e76\u7ed9\u51fa\u56de\u7b54\u3002\u6211\u4eec\u8ba1\u5212\u5728\u540e\u7eed\u66f4\u65b0\u4e2d\u89e3\u51b3\u8fd9\u4e00\u9650\u5236\u3002\u8be5\u9650\u5236\u53ef\u80fd\u4e0e\u57fa\u7840\u68c0\u67e5\u70b9\u6709\u5173\uff1aDeepSeek\u2011V3 Base \u4e3b\u8981\u4f7f\u7528\u4e2d\u6587\u548c\u82f1\u6587\uff0c\u56e0\u6b64\u5728\u63a8\u7406\u4e2d\u8fd9\u4e24\u79cd\u8bed\u8a00\u7684\u8868\u73b0\u66f4\u597d\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u63d0\u793a\u5de5\u7a0b<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5728\u8bc4\u6d4b DeepSeek\u2011R1 \u65f6\uff0c\u6211\u4eec\u53d1\u73b0\u5b83\u5bf9\u63d0\u793a\u8f83\u4e3a\u654f\u611f\u3002\u5c11\u6837\u672c\u63d0\u793a\u5f80\u5f80\u4f1a\u964d\u4f4e\u5176\u6027\u80fd\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u5efa\u8bae\u7528\u6237\u5728\u96f6\u6837\u672c\u8bbe\u7f6e\u4e0b\uff0c\u76f4\u63a5\u3001\u6e05\u6670\u5730\u63cf\u8ff0\u95ee\u9898\u5e76\u6307\u5b9a\u8f93\u51fa\u683c\u5f0f\uff0c\u4ee5\u83b7\u5f97\u6700\u4f73\u7ed3\u679c\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u8f6f\u4ef6\u5de5\u7a0b\u4efb\u52a1<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u7531\u4e8e\u8bc4\u6d4b\u65f6\u95f4\u8f83\u957f\u4f1a\u5f71\u54cd RL \u6d41\u7a0b\u7684\u6548\u7387\uff0c\u5927\u89c4\u6a21 RL \u76ee\u524d\u5c1a\u672a\u5728\u8f6f\u4ef6\u5de5\u7a0b\u4efb\u52a1\u4e0a\u5f97\u5230\u5e7f\u6cdb\u5e94\u7528\u3002\u56e0\u6b64\uff0cDeepSeek\u2011R1 \u5728\u8f6f\u4ef6\u5de5\u7a0b\u57fa\u51c6\u4e0a\u5c1a\u672a\u663e\u793a\u51fa\u76f8\u5bf9\u4e8e DeepSeek\u2011V3 \u7684\u5de8\u5927\u63d0\u5347\u7a7a\u95f4\u3002\u672a\u6765\u7684\u7248\u672c\u5c06\u901a\u8fc7\u5728\u8f6f\u4ef6\u5de5\u7a0b\u6570\u636e\u4e0a\u5b9e\u65bd\u62d2\u7edd\u91c7\u6837\uff0c\u6216\u5728 RL \u8fc7\u7a0b\u4e2d\u5f15\u5165\u5f02\u6b65\u8bc4\u6d4b\u7b49\u65b9\u5f0f\u6765\u63d0\u5347\u6548\u7387\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u8d85\u8d8a\u5177\u4f53\u80fd\u529b\u9650\u5236\uff1a\u7eaf RL \u65b9\u6cd5\u672c\u8eab\u7684\u5185\u5728\u6311\u6218\uff1a<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span><span style=\"font-size: 15px\">\u5956\u52b1\u9ed1\u5ba2\uff08Reward hacking\uff09<\/span><\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u7eaf\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u7684\u6210\u529f\u4f9d\u8d56\u4e8e\u53ef\u9760\u7684\u5956\u52b1\u4fe1\u53f7\u3002\u672c\u7814\u7a76\u901a\u8fc7\u9762\u5411\u63a8\u7406\u9886\u57df\u7684\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\u6a21\u578b\u6765\u786e\u4fdd\u5956\u52b1\u7684\u53ef\u9760\u6027\u3002\u7136\u800c\uff0c\u5bf9\u4e8e\u67d0\u4e9b\u4efb\u52a1\uff08\u5982\u5199\u4f5c\uff09\uff0c\u8fd9\u79cd\u53ef\u9760\u7684\u5956\u52b1\u6a21\u578b\u5f88\u96be\u6784\u5efa\u3002\u5982\u679c\u5956\u52b1\u4fe1\u53f7\u7531\u6a21\u578b\u800c\u975e\u9884\u5b9a\u4e49\u89c4\u5219\u7ed9\u51fa\uff0c\u968f\u7740\u8bad\u7ec3\u63a8\u8fdb\uff0c\u5b83\u66f4\u5bb9\u6613\u88ab\u201c\u94bb\u7a7a\u5b50\u201d\uff0c\u8fd9\u610f\u5473\u7740\u7b56\u7565\u6a21\u578b\u53ef\u80fd\u4f1a\u627e\u5230\u6377\u5f84\u6765\u201c\u9ed1\u201d\u6389\u5956\u52b1\u6a21\u578b\u3002\u56e0\u6b64\uff0c\u5bf9\u4e8e\u90a3\u4e9b\u65e0\u6cd5\u901a\u8fc7\u53ef\u9760\u5956\u52b1\u6a21\u578b\u8fdb\u884c\u6709\u6548\u8bc4\u4f30\u7684\u590d\u6742\u4efb\u52a1\uff0c\u5355\u7eaf\u6269\u5927\u7eaf RL \u65b9\u6cd5\u7684\u89c4\u6a21\u4ecd\u662f\u4e00\u9879\u5f00\u653e\u6311\u6218\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5728\u672c\u5de5\u4f5c\u4e2d\uff0c\u5bf9\u4e8e\u65e0\u6cd5\u83b7\u5f97\u53ef\u9760\u5956\u52b1\u4fe1\u53f7\u7684\u4efb\u52a1\uff0cDeepSeek\u2011R1 \u91c7\u7528\u4eba\u5de5\u6807\u6ce8\u6765\u6784\u5efa\u76d1\u7763\u6570\u636e\uff0c\u5e76\u4ec5\u8fdb\u884c\u6570\u767e\u6b65\u7684 RL\u3002\u6211\u4eec\u5e0c\u671b\u672a\u6765\u80fd\u591f\u83b7\u5f97\u66f4\u9c81\u68d2\u7684\u5956\u52b1\u6a21\u578b\uff0c\u4ee5\u89e3\u51b3\u6b64\u7c7b\u95ee\u9898\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u968f\u7740\u50cf DeepSeek\u2011R1 \u8fd9\u6837\u7684\u7eaf RL \u65b9\u6cd5\u7684\u51fa\u73b0\uff0c\u672a\u6765\u5728\u89e3\u51b3\u4efb\u4f55\u80fd\u88ab\u9a8c\u8bc1\u5668\u6709\u6548\u8bc4\u4f30\u7684\u95ee\u9898\u65b9\u9762\u90fd\u8574\u542b\u5de8\u5927\u6f5c\u529b\uff0c\u800c\u4e0d\u8bba\u5176\u5bf9\u4eba\u7c7b\u800c\u8a00\u6709\u591a\u590d\u6742\u3002\u914d\u5907\u6b64\u7c7b\u5148\u8fdb RL \u6280\u672f\u7684\u673a\u5668\uff0c\u6709\u671b\u5728\u8fd9\u4e9b\u9886\u57df\u8d85\u8d8a\u4eba\u7c7b\u80fd\u529b\uff0c\u5176\u52a8\u529b\u6e90\u4e8e\u5b83\u4eec\u901a\u8fc7\u8bd5\u9519\u8fed\u4ee3\u6765\u4f18\u5316\u6027\u80fd\u7684\u80fd\u529b\u3002\u7136\u800c\uff0c\u5bf9\u4e8e\u90a3\u4e9b\u96be\u4ee5\u6784\u5efa\u53ef\u9760\u5956\u52b1\u6a21\u578b\u7684\u4efb\u52a1\uff0c\u4ecd\u7136\u5b58\u5728\u6311\u6218\u3002\u5728\u8fd9\u7c7b\u60c5\u5f62\u4e0b\uff0c\u7f3a\u4e4f\u7a33\u5065\u7684\u53cd\u9988\u673a\u5236\u53ef\u80fd\u4f1a\u51cf\u7f13\u8fdb\u5c55\uff0c\u8fd9\u8868\u660e\u672a\u6765\u7814\u7a76\u5e94\u81f4\u529b\u4e8e\u5f00\u53d1\u521b\u65b0\u65b9\u6cd5\uff0c\u4ee5\u4fbf\u4e3a\u8fd9\u4e9b\u590d\u6742\u3001\u96be\u4ee5\u9a8c\u8bc1\u7684\u95ee\u9898\u5b9a\u4e49\u5e76\u5b8c\u5584\u5956\u52b1\u7ed3\u6784\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6b64\u5916\uff0c\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u4f7f\u7528\u5de5\u5177\u4e5f\u5c55\u73b0\u51fa\u663e\u8457\u524d\u666f\u3002\u65e0\u8bba\u662f\u4f7f\u7528\u7f16\u8bd1\u5668\u6216\u641c\u7d22\u5f15\u64ce\u6765\u68c0\u7d22\u6216\u8ba1\u7b97\u5fc5\u8981\u4fe1\u606f\uff0c\u8fd8\u662f\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u4f7f\u7528\u751f\u7269\u6216\u5316\u5b66\u8bd5\u5242\u6765\u9a8c\u8bc1\u6700\u7ec8\u7ed3\u679c\uff0c\u8fd9\u79cd\u5de5\u5177\u589e\u5f3a\u578b\u63a8\u7406\u7684\u878d\u5408\u90fd\u53ef\u80fd\u6781\u5927\u62d3\u5c55\u673a\u5668\u9a71\u52a8\u89e3\u51b3\u65b9\u6848\u7684\u9002\u7528\u8303\u56f4\u4e0e\u51c6\u786e\u6027\u3002<\/span><\/span><\/p>\n<section style=\"margin-bottom: 0px;letter-spacing: 0.578px\">\n<section>\n<section>\n<section data-tools=\"135\u7f16\u8f91\u5668\" data-id=\"88077\">\n<section style=\"margin-right: auto;margin-left: auto;padding-top: 15px\">\n<section style=\"margin-bottom: -14px;justify-content: flex-end\">\n<section style=\"margin-top: -14px;width: 5px;height: 5px;border-right: none;border-top: none;overflow: hidden\"><span><br \/><\/span><\/section>\n<\/section>\n<section data-width=\"100%\" style=\"margin-right: auto;margin-left: auto;width: 578px;border-width: 1px;border-style: solid\">\n<section style=\"padding: 3px\">\n<section style=\"padding: 10px;border-width: 1px;border-style: solid\">\n<section style=\"font-size: 15px;letter-spacing: 1.5px;line-height: 25px\">\n<p style=\"vertical-align: inherit\"><span style=\"letter-spacing: 0.578px\"><span>\u76f8\u5173\u9605\u8bfb\uff1a<\/span><span data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;p&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;margin-left: 8px;margin-right: 8px;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\"><a class=\"normal_text_link\" target=\"_blank\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247707721&amp;idx=1&amp;sn=52c0370762a13598d3c76cf090ecfac3&amp;scene=21&amp;token=585105048&amp;lang=zh_CN#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"font-size: 15px\">DeepSeek R1<\/span><\/a><\/span><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin-top: -5px;width: 5px;height: 5px;border-left: none;border-bottom: none;overflow: hidden\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<p style=\"vertical-align: inherit\"><span><br \/><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u65b9\u6cd5<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">GRPO<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">GRPO \u662f\u6211\u4eec\u7528\u4e8e\u8bad\u7ec3 DeepSeek\u2011R1\u2011Zero \u548c DeepSeek\u2011R1 \u7684\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u7b97\u6cd5\u3002\u5b83\u6700\u521d\u88ab\u63d0\u51fa\u662f\u4e3a\u4e86\u7b80\u5316\u8bad\u7ec3\u6d41\u7a0b\u5e76\u964d\u4f4e\u8fd1\u7aef\u7b56\u7565\u4f18\u5316\uff08PPO\uff09\u7684\u8d44\u6e90\u6d88\u8017\uff0c\u8be5\u65b9\u6cd5\u5728\u5927\u8bed\u8a00\u6a21\u578b\u7684 RL \u9636\u6bb5\u88ab\u5e7f\u6cdb\u91c7\u7528\u3002GRPO \u7684\u6d41\u7a0b\u5c55\u793a\u4e8e\u6269\u5c55\u6570\u636e\u56fe 2 \u4e2d\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5bf9\u4e8e\u6bcf\u4e2a\u95ee\u9898 (q)\uff0cGRPO \u4f1a\u4ece\u65e7\u7b56\u7565<\/span><\/span><sub data-pm-slice=\"0 0 []\"><span><img alt=\"image.png\" class=\"rich_pages wxw-img\" data-ratio=\"0.5113636363636364\" data-type=\"png\" data-w=\"176\" style=\"vertical-align: baseline;width: 35px\" width=\"35\" data-width=\"35px\" data-imgfileid=\"100236310\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-76748e609295d114195ff047224cf4a7.png\" \/><img class=\"rich_pages wxw-img\" \/><\/span><\/sub><span data-mpa-action-id=\"mfoltxxb12an\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">\u4e2d\u91c7\u6837\u4e00\u7ec4\u8f93\u51fa o<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolu5id1swk\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">1<\/span><\/sub><span data-mpa-action-id=\"mfolu0im1p4m\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">, o<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolu2ut13tf\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">2<\/span><\/sub><span data-mpa-action-id=\"mfolui7hd2v\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">, \u2026, o<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolujbx17o5\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">G<\/span><\/sub><span data-mpa-action-id=\"mfolvqgh2v6\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">\uff0c\u968f\u540e\u901a\u8fc7\u6700\u5927\u5316\u4ee5\u4e0b\u76ee\u6807\u6765\u4f18\u5316\u7b56\u7565\u6a21\u578b (\u03c0<\/span><\/span><sub style=\"font-size: 15px\"><span>\u03b8<\/span><\/sub><span><span style=\"font-size: 15px\">)\u3002<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;text-align: center\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.2953703703703704\" data-type=\"png\" data-w=\"1080\" style=\"width: 100%\" data-width=\"1900\" data-height=\"562\" data-backw=\"562\" data-backh=\"166\" data-imgfileid=\"100236274\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-85b89694fe14251fca706f12e041fa54.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span data-mpa-action-id=\"mfolw7aax9a\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">\u5176\u4e2d\uff0c(\u03c0<\/span><\/span><sub style=\"font-size: 15px\"><span>ref<\/span><\/sub><span data-mpa-action-id=\"mfolxnog19pw\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">) \u662f\u53c2\u8003\u7b56\u7565\uff0c\u03f5\u548c \u03b2 \u4e3a\u8d85\u53c2\u6570\uff0cA<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolyug222mw\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">i<\/span><\/sub><span data-mpa-action-id=\"mfoly3kt1m6e\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">&nbsp;\u662f\u4f18\u52bf\u9879\uff08advantage\uff09\uff0c\u5176\u901a\u8fc7\u6bcf\u7ec4\u5bf9\u5e94\u4e8e\u5404\u4e2a\u8f93\u51fa\u7684\u5956\u52b1 r<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolypeumr2\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">1<\/span><\/sub><span data-mpa-action-id=\"mfoly6tmwgd\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">, r<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolym6wuuq\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">2<\/span><\/sub><span data-mpa-action-id=\"mfolyc6bd4u\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">, \u2026, r<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfolydkpbyd\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">G<\/span><\/sub><span><span style=\"font-size: 15px\">) \u8ba1\u7b97\u5f97\u5230\uff1a<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.10277777777777777\" data-type=\"png\" data-w=\"1080\" style=\"width:100%\" data-width=\"1886\" data-height=\"194\" data-backw=\"562\" data-backh=\"58\" data-imgfileid=\"100236284\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-086306739a2e6bba304bd0b1ecb2437c.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u5728\u8865\u5145\u4fe1\u606f\u7b2c 1.3 \u8282\u4e2d\u7ed9\u51fa\u4e86 GRPO \u4e0e PPO \u7684\u5bf9\u6bd4\u3002<\/span><\/span><\/p>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u5956\u52b1\u8bbe\u8ba1<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5956\u52b1\u662f\u8bad\u7ec3\u4fe1\u53f7\u7684\u6765\u6e90\uff0c\u51b3\u5b9a\u4e86\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u4f18\u5316\u7684\u65b9\u5411\u3002\u5bf9\u4e8e DeepSeek\u2011R1\u2011Zero\uff0c\u6211\u4eec\u4f7f\u7528\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\uff0c\u4e3a\u6570\u5b66\u3001\u7f16\u7a0b\u4e0e\u903b\u8f91\u63a8\u7406\u7b49\u9886\u57df\u7684\u6570\u636e\u63d0\u4f9b\u7cbe\u786e\u53cd\u9988\u3002\u5bf9\u4e8e DeepSeek\u2011R1\uff0c\u6211\u4eec\u5728\u6b64\u57fa\u7840\u4e0a\u8fdb\u4e00\u6b65\u6269\u5c55\uff1a\u5bf9\u63a8\u7406\u7c7b\u6570\u636e\u91c7\u7528\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\uff0c\u5bf9\u901a\u7528\u6570\u636e\u91c7\u7528\u57fa\u4e8e\u6a21\u578b\u7684\u5956\u52b1\uff0c\u4ece\u800c\u63d0\u5347\u5b66\u4e60\u8fc7\u7a0b\u5728\u4e0d\u540c\u9886\u57df\u95f4\u7684\u9002\u5e94\u6027\u3002<\/span><\/span><\/p>\n<p style=\"text-indent: 0px;margin-left: 8px;margin-right: 8px\"><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><span style=\"font-size: 15px\" data-mpa-action-id=\"mfol7nog1vxh\">\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1<\/span><\/span><\/strong><\/strong><\/span><\/strong><\/strong><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u7684\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\u7cfb\u7edf\u4e3b\u8981\u5305\u542b\u4e24\u7c7b\u5956\u52b1\uff1a\u51c6\u786e\u6027\u5956\u52b1\u4e0e\u683c\u5f0f\u5956\u52b1\u3002<\/span><\/span><\/p>\n<ul style=\"margin-left: 8px;margin-right: 8px\" class=\"list-paddingleft-1\">\n<li style=\"text-align: left\">\n<p><span><span style=\"font-size: 15px\">\u51c6\u786e\u6027\u5956\u52b1\uff1a\u8bc4\u4f30\u56de\u7b54\u662f\u5426\u6b63\u786e\u3002\u6bd4\u5982\uff0c\u5bf9\u4e8e\u5177\u6709\u786e\u5b9a\u6027\u7ed3\u679c\u7684\u6570\u5b66\u9898\uff0c\u8981\u6c42\u6a21\u578b\u6309\u6307\u5b9a\u683c\u5f0f\uff08\u4f8b\u5982\u5728\u65b9\u6846\u5185\uff09\u7ed9\u51fa\u6700\u7ec8\u7b54\u6848\uff0c\u4ee5\u4fbf\u8fdb\u884c\u53ef\u9760\u7684\u57fa\u4e8e\u89c4\u5219\u7684\u6b63\u786e\u6027\u6821\u9a8c\u3002\u7c7b\u4f3c\u5730\uff0c\u5bf9\u4e8e\u7f16\u7a0b\u7ade\u8d5b\u7c7b\u63d0\u793a\u8bcd\uff0c\u53ef\u4f7f\u7528\u7f16\u8bd1\u5668\u4f9d\u636e\u4e00\u5957\u9884\u5b9a\u4e49\u6d4b\u8bd5\u7528\u4f8b\u6765\u8bc4\u4f30\u6a21\u578b\u7684\u8f93\u51fa\uff0c\u4ece\u800c\u5bf9\u6b63\u786e\u6027\u751f\u6210\u5ba2\u89c2\u53cd\u9988\u3002<\/span><\/span><\/p>\n<p><span><br \/><\/span><\/p>\n<\/li>\n<li style=\"text-align: left\">\n<p><span><span style=\"font-size: 15px\">\u683c\u5f0f\u5956\u52b1\uff1a\u7528\u7279\u5b9a\u7684\u683c\u5f0f\u8981\u6c42\u6765\u8865\u5145\u51c6\u786e\u6027\u5956\u52b1\u673a\u5236\u3002\u5177\u4f53\u800c\u8a00\uff0c\u6a21\u578b\u88ab\u6fc0\u52b1\u5c06\u5176\u63a8\u7406\u8fc7\u7a0b\u5c01\u88c5\u5728\u6307\u5b9a\u6807\u7b7e\u5185\uff0c\u7279\u522b\u662f\u4f7f\u7528&nbsp;<\/span><\/span><code><span><span style=\"font-size: 15px\">&lt;think&gt;<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u548c&nbsp;<\/span><\/span><code><span><span style=\"font-size: 15px\">&lt;\/think&gt;<\/span><\/span><\/code><span><span style=\"font-size: 15px\">\u3002\u8fd9\u6837\u53ef\u4ee5\u4f7f\u6a21\u578b\u7684\u601d\u8003\u8fc7\u7a0b\u88ab\u6e05\u6670\u5730\u5448\u73b0\uff0c\u589e\u5f3a\u53ef\u89e3\u91ca\u6027\uff0c\u5e76\u4fbf\u4e8e\u540e\u7eed\u5206\u6790\u3002<\/span><\/span><\/p>\n<\/li>\n<\/ul>\n<section style=\"margin-left: 8px;margin-right: 8px;text-align: center\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.07222222222222222\" data-type=\"png\" data-w=\"1080\" data-width=\"1904\" data-height=\"138\" data-imgfileid=\"100236279\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-98ff9a8310fb13588424329470b8aed0.png\" \/><\/section>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u57fa\u4e8e\u6a21\u578b\u7684\u5956\u52b1\u4e0e\u6743\u91cd\u8bbe\u7f6e<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u5c06\u51c6\u786e\u6027\u5956\u52b1\u4e0e\u683c\u5f0f\u5956\u52b1\u4ee5\u76f8\u540c\u6743\u91cd\u8fdb\u884c\u7ec4\u5408\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u6211\u4eec\u907f\u514d\u5728\u63a8\u7406\u4efb\u52a1\u4e2d\u4f7f\u7528\u795e\u7ecf\u5956\u52b1\u6a21\u578b\u2014\u2014\u65e0\u8bba\u662f\u57fa\u4e8e\u7ed3\u679c\u7684\u8fd8\u662f\u57fa\u4e8e\u8fc7\u7a0b\u7684\u3002\u505a\u51fa\u8fd9\u4e00\u51b3\u5b9a\u7684\u4f9d\u636e\u662f\u6211\u4eec\u89c2\u5bdf\u5230\uff1a\u5728\u5927\u89c4\u6a21 RL \u4e2d\uff0c\u795e\u7ecf\u5956\u52b1\u6a21\u578b\u5bb9\u6613\u53d7\u5230\u201c\u5956\u52b1\u9ed1\u5ba2\u201d\u7684\u5f71\u54cd\u3002\u6b64\u5916\uff0c\u91cd\u65b0\u8bad\u7ec3\u8fd9\u7c7b\u6a21\u578b\u9700\u8981\u5927\u91cf\u8ba1\u7b97\u8d44\u6e90\uff0c\u5e76\u4f1a\u7ed9\u8bad\u7ec3\u6d41\u6c34\u7ebf\u5f15\u5165\u66f4\u591a\u590d\u6742\u6027\uff0c\u4ece\u800c\u4f7f\u6574\u4f53\u4f18\u5316\u8fc7\u7a0b\u66f4\u52a0\u590d\u6742\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><strong style=\"text-align: left;letter-spacing: 0.544px\"><strong style=\"letter-spacing: 0.578px\"><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><span><span style=\"font-size: 15px\">\u57fa\u4e8e\u6a21\u578b\u7684\u5956\u52b1<\/span><\/span><\/span><\/strong><\/strong><\/strong><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5bf9\u4e8e\u901a\u7528\u6570\u636e\uff0c\u6211\u4eec\u91c7\u7528\u5956\u52b1\u6a21\u578b\u6765\u6355\u6349\u4eba\u7c7b\u5728\u590d\u6742\u4e14\u7ec6\u5fae\u573a\u666f\u4e2d\u7684\u504f\u597d\u3002\u6211\u4eec\u5728 DeepSeek\u2011V3 \u7684\u6d41\u6c34\u7ebf\u4e0a\u6784\u5efa\uff0c\u5e76\u4f7f\u7528\u7c7b\u4f3c\u7684\u504f\u597d\u5bf9\u4e0e\u8bad\u7ec3\u63d0\u793a\u7684\u5206\u5e03\u3002<\/span><\/span><\/p>\n<ul style=\"margin-left: 8px;margin-right: 8px\" class=\"list-paddingleft-1\">\n<li style=\"text-align: left\">\n<p style=\"margin-bottom: 8px\"><span><span style=\"font-size: 15px\">\u9488\u5bf9\u201c\u6709\u7528\u6027\uff08helpfulness\uff09\u201d\uff0c\u6211\u4eec\u53ea\u5173\u6ce8\u6700\u7ec8\u603b\u7ed3\uff0c\u786e\u4fdd\u8bc4\u4f30\u5f3a\u8c03\u56de\u7b54\u5bf9\u7528\u6237\u7684\u7528\u9014\u4e0e\u76f8\u5173\u6027\uff0c\u540c\u65f6\u5c3d\u91cf\u51cf\u5c11\u5bf9\u5e95\u5c42\u63a8\u7406\u8fc7\u7a0b\u7684\u5e72\u6270\u3002<\/span><\/span><\/p>\n<\/li>\n<li style=\"text-align: left\">\n<p><span><span style=\"font-size: 15px\">\u9488\u5bf9\u201c\u65e0\u5bb3\u6027\uff08harmlessness\uff09\u201d\uff0c\u6211\u4eec\u4f1a\u8bc4\u4f30\u6a21\u578b\u7684\u5b8c\u6574\u54cd\u5e94\uff0c\u5305\u62ec\u63a8\u7406\u8fc7\u7a0b\u4e0e\u603b\u7ed3\uff0c\u4ee5\u8bc6\u522b\u5e76\u7f13\u89e3\u5728\u751f\u6210\u8fc7\u7a0b\u4e2d\u53ef\u80fd\u51fa\u73b0\u7684\u4efb\u4f55\u6f5c\u5728\u98ce\u9669\u3001\u504f\u89c1\u6216\u6709\u5bb3\u5185\u5bb9\u3002<\/span><\/span><\/p>\n<\/li>\n<\/ul>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u6709\u7528\u6027\u5956\u52b1\u6a21\u578b\uff08Helpful reward model\uff09<\/span><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u4e86\u8bad\u7ec3\u6709\u7528\u6027\u5956\u52b1\u6a21\u578b\uff0c\u6211\u4eec\u9996\u5148\u4f7f\u7528 DeepSeek\u2011V3 \u4f9d\u636e Arena\u2011Hard \u63d0\u793a\u683c\u5f0f\uff08\u89c1\u8865\u5145\u4fe1\u606f\u7b2c 2.2 \u8282\uff09\u751f\u6210\u504f\u597d\u5bf9\u3002\u6bcf\u4e2a\u504f\u597d\u5bf9\u7531\u4e00\u4e2a\u7528\u6237\u95ee\u9898\u548c\u4e24\u4e2a\u5019\u9009\u7b54\u6848\u7ec4\u6210\u3002\u5bf9\u4e8e\u6bcf\u4e2a\u504f\u597d\u5bf9\uff0c\u6211\u4eec\u5411 DeepSeek\u2011V3 \u8be2\u95ee\u56db\u6b21\uff0c\u5e76\u5c06\u56de\u7b54\u968f\u673a\u6307\u5b9a\u4e3a\u201c\u56de\u7b54 A\u201d\u6216\u201c\u56de\u7b54 B\u201d\uff0c\u4ee5\u51cf\u5c11\u4f4d\u7f6e\u504f\u5dee\u3002\u6700\u7ec8\u7684\u504f\u597d\u5206\u6570\u7531\u8fd9\u56db\u6b21\u72ec\u7acb\u5224\u65ad\u7684\u5e73\u5747\u503c\u786e\u5b9a\uff0c\u4ec5\u4fdd\u7559\u5206\u5dee \u0394 \u5927\u4e8e 1 \u7684\u914d\u5bf9\uff0c\u4ee5\u786e\u4fdd\u5dee\u5f02\u5177\u6709\u610f\u4e49\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6b64\u5916\uff0c\u4e3a\u4e86\u6700\u5c0f\u5316\u4e0e\u957f\u5ea6\u76f8\u5173\u7684\u504f\u5dee\uff0c\u6211\u4eec\u786e\u4fdd\u6574\u4e2a\u6570\u636e\u96c6\u4e2d\u88ab\u9009\u4e2d\u548c\u88ab\u62d2\u7edd\u7684\u56de\u7b54\u5728\u957f\u5ea6\u4e0a\u5927\u4f53\u76f8\u5f53\u3002\u603b\u4f53\u800c\u8a00\uff0c\u6211\u4eec\u7cbe\u9009\u4e86 66,000 \u5bf9\u6570\u636e\u7528\u4e8e\u8bad\u7ec3\u5956\u52b1\u6a21\u578b\u3002\u8be5\u6570\u636e\u96c6\u4e2d\u7684\u63d0\u793a\u5747\u4e3a\u975e\u63a8\u7406\u95ee\u9898\uff0c\u6765\u6e90\u4e8e\u516c\u5f00\u7684\u5f00\u6e90\u6570\u636e\u96c6\uff0c\u6216\u6765\u81ea\u660e\u786e\u540c\u610f\u5171\u4eab\u5176\u6570\u636e\u4ee5\u7528\u4e8e\u6a21\u578b\u6539\u8fdb\u7684\u7528\u6237\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u7684\u5956\u52b1\u6a21\u578b\u67b6\u6784\u4e0e DeepSeek\u2011R1 \u4e00\u81f4\uff0c\u5e76\u5728\u6b64\u57fa\u7840\u4e0a\u589e\u52a0\u4e86\u4e00\u4e2a\u7528\u4e8e\u9884\u6d4b\u6807\u91cf\u504f\u597d\u5206\u6570\u7684\u5956\u52b1\u5934\uff08reward head\uff09\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;text-align: center\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.0824074074074074\" data-type=\"png\" data-w=\"1080\" style=\"width: 562px;height: 46px\" data-width=\"1888\" data-height=\"156\" data-imgfileid=\"100236282\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-788115a57e7d24ec6b4fcf76903e0406.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span data-mpa-action-id=\"mfoma9ti19uw\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">\u6709\u7528\u6027\u5956\u52b1\u6a21\u578b\u5728\u6574\u4e2a\u8bad\u7ec3\u6570\u636e\u96c6\u4e0a\u4ee5\u5355\u8f6e\uff08epoch\uff09\u65b9\u5f0f\u8bad\u7ec3\uff0c\u6279\u5927\u5c0f\u4e3a 256\uff0c\u5b66\u4e60\u7387\u4e3a 6<\/span>\u00d7<span style=\"font-size: 15px\">10<\/span><\/span><sup style=\"font-size: 15px\" data-mpa-action-id=\"mfomabkf1uzp\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">-6<\/span><\/sup><span><span style=\"font-size: 15px\">\u3002\u8bad\u7ec3\u671f\u95f4\u7684\u6700\u5927\u5e8f\u5217\u957f\u5ea6\u8bbe\u4e3a 8192 \u4e2a token\uff0c\u800c\u5728\u5956\u52b1\u6a21\u578b\u63a8\u7406\u9636\u6bb5\u4e0d\u65bd\u52a0\u660e\u786e\u7684\u957f\u5ea6\u9650\u5236\u3002<\/span><\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u5b89\u5168\u6027\u5956\u52b1\u6a21\u578b<\/span><\/span><\/h3>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u4e3a\u8bc4\u4f30\u5e76\u63d0\u5347\u6a21\u578b\u5b89\u5168\u6027\uff0c\u6211\u4eec\u6574\u7406\u4e86\u4e00\u4e2a\u5305\u542b 106,000 \u4e2a\u63d0\u793a\u7684\u6570\u636e\u96c6\uff0c\u8fd9\u4e9b\u63d0\u793a\u914d\u6709\u6a21\u578b\u751f\u6210\u7684\u56de\u7b54\uff0c\u5e76\u4f9d\u636e\u9884\u5b9a\u4e49\u7684\u5b89\u5168\u6307\u5357\u88ab\u6807\u6ce8\u4e3a\u201c\u5b89\u5168\u201d\u6216\u201c\u4e0d\u5b89\u5168\u201d\u3002\u4e0e\u6709\u7528\u6027\u5956\u52b1\u6a21\u578b\u4e2d\u4f7f\u7528\u7684\u6210\u5bf9\u635f\u5931\uff08pairwise loss\uff09\u4e0d\u540c\uff0c\u5b89\u5168\u6027\u5956\u52b1\u6a21\u578b\u4f7f\u7528\u70b9\u5f0f\u65b9\u6cd5\uff08pointwise methodology\uff09\u6765\u533a\u5206\u5b89\u5168\u4e0e\u4e0d\u5b89\u5168\u7684\u56de\u7b54\u3002\u5176\u8bad\u7ec3\u8d85\u53c2\u6570\u4e0e\u6709\u7528\u6027\u5956\u52b1\u6a21\u578b\u76f8\u540c\u3002<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;text-align: center\"><img class=\"rich_pages wxw-img\" data-ratio=\"0.08796296296296297\" data-type=\"png\" data-w=\"1080\" data-width=\"1890\" data-height=\"166\" data-imgfileid=\"100236280\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-dd913befce75a0e23f88b30f1c2d80b1.png\" \/><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span data-mpa-action-id=\"mfom1mec14m\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">\u5bf9\u4e8e\u901a\u7528\u67e5\u8be2\uff0c\u6bcf\u4e2a\u6837\u672c\u90fd\u4f1a\u88ab\u5f52\u5165\u5b89\u5168\u6570\u636e\u96c6\u6216\u6709\u7528\u6027\u6570\u636e\u96c6\u4e4b\u4e00\u3002\u5206\u914d\u7ed9\u6bcf\u4e2a\u67e5\u8be2\u7684\u901a\u7528\u5956\u52b1 Reward<\/span><\/span><sub style=\"font-size: 15px\" data-mpa-action-id=\"mfom1tfm1hej\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">general<\/span><\/sub><span><span style=\"font-size: 15px\">&nbsp;\u4e0e\u5176\u6240\u5c5e\u6570\u636e\u96c6\u4e2d\u5b9a\u4e49\u7684\u76f8\u5e94\u5956\u52b1\u4e00\u81f4\u3002<\/span><\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h3>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u8bad\u7ec3\u7ec6\u8282<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">DeepSeek\u2011R1\u2011Zero \u7684\u8bad\u7ec3\u7ec6\u8282<\/span><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span data-mpa-action-id=\"mfom0ug7q8n\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">\u4e3a\u4e86\u8bad\u7ec3 DeepSeek\u2011R1\u2011Zero\uff0c\u6211\u4eec\u5c06\u5b66\u4e60\u7387\u8bbe\u4e3a (3<\/span>\u00d7<span style=\"font-size: 15px\">10<\/span><\/span><sup style=\"font-size: 15px\" data-mpa-action-id=\"mfom0x9wdwc\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">-6<\/span><\/sup><span><span style=\"font-size: 15px\">)\uff0c\u5c06 KL\uff08Kullback\u2013Leibler\uff09\u7cfb\u6570\u8bbe\u4e3a 0.001\uff0c\u91c7\u6837\u6e29\u5ea6\u5728 rollout \u65f6\u8bbe\u4e3a 1\u3002\u5bf9\u4e8e\u6bcf\u4e2a\u95ee\u9898\uff0c\u6211\u4eec\u91c7\u6837 16 \u4e2a\u8f93\u51fa\uff1a\u5728 8.2k \u6b65\u4e4b\u524d\u7684\u6700\u5927\u957f\u5ea6\u4e3a 32,768 \u4e2a token\uff0c\u5728 8.2k \u6b65\u4e4b\u540e\u6700\u5927\u957f\u5ea6\u4e3a 65,536 \u4e2a token\u3002<\/span><\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u7531\u6b64\uff0cDeepSeek\u2011R1\u2011Zero \u7684\u6027\u80fd\u4e0e\u54cd\u5e94\u957f\u5ea6\u5728\u7b2c 8.2k \u6b65\u51fa\u73b0\u663e\u8457\u8dc3\u5347\uff1b\u8bad\u7ec3\u603b\u8ba1\u8fdb\u884c\u5230 10,400 \u6b65\uff0c\u5bf9\u5e94\u7ea6 1.6 \u4e2a\u8bad\u7ec3 epoch\u3002\u6bcf\u4e2a\u8bad\u7ec3\u6b65\u5305\u542b 32 \u4e2a\u4e0d\u540c\u7684\u95ee\u9898\uff0c\u56e0\u6b64\u6bcf\u6b65\u7684\u8bad\u7ec3\u6279\u5927\u5c0f\u4e3a 512\u3002\u6bcf 400 \u6b65\uff0c\u6211\u4eec\u7528\u6700\u65b0\u7684\u7b56\u7565\u6a21\u578b\u66ff\u6362\u53c2\u8003\u6a21\u578b\u3002\u4e3a\u52a0\u901f\u8bad\u7ec3\uff0c\u6bcf\u6b21 rollout \u4ea7\u751f 8,192 \u4e2a\u8f93\u51fa\uff0c\u968f\u673a\u5212\u5206\u4e3a 16 \u4e2a\u5c0f\u6279\u6b21\uff0c\u4ec5\u8bad\u7ec3\u4e00\u4e2a\u5185\u90e8 epoch\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u7b2c\u4e00\u9636\u6bb5 RL \u7684\u8bad\u7ec3\u7ec6\u8282<\/span><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5728 RL \u7684\u7b2c\u4e00\u9636\u6bb5\uff0c\u6211\u4eec\u5c06\u5b66\u4e60\u7387\u8bbe\u4e3a 3<\/span><\/span><span data-mpa-action-id=\"mfom0ug7q8n\" data-pm-slice=\"0 0 []\"><span data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;p&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;margin-left: 8px;margin-right: 8px;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;span&quot;,&quot;attributes&quot;:{&quot;data-mpa-action-id&quot;:&quot;mfom0ug7q8n&quot;,&quot;data-pm-slice&quot;:&quot;0 0 []&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\">\u00d7<\/span><\/span><span data-mpa-action-id=\"mfom2jp61d6c\" data-pm-slice=\"0 0 []\"><span><span style=\"font-size: 15px\">10<\/span><\/span><sup style=\"font-size: 15px\" data-mpa-action-id=\"mfom2l9q14pf\" data-pm-slice=\"0 0 []\"><span style=\"font-size: 12px\">-6<\/span><\/sup><span><span style=\"font-size: 15px\">\uff0cKL \u7cfb\u6570\u4e3a 0.001\uff0cGRPO \u7684\u88c1\u526a\u6bd4\u7387 (\u03f5) \u8bbe\u4e3a 10\uff0crollout \u7684\u91c7\u6837\u6e29\u5ea6\u4e3a 1\u3002\u5bf9\u4e8e\u6bcf\u4e2a\u95ee\u9898\uff0c\u6211\u4eec\u91c7\u6837 16 \u4e2a\u8f93\u51fa\uff0c\u6700\u5927\u957f\u5ea6\u4e3a 32,768 \u4e2a token\u3002\u6bcf\u4e2a\u8bad\u7ec3\u6b65\u5305\u542b 32 \u4e2a\u4e0d\u540c\u7684\u95ee\u9898\uff0c\u56e0\u6b64\u6bcf\u6b65\u7684\u8bad\u7ec3\u6279\u5927\u5c0f\u4e3a 512\u3002\u6bcf 400 \u6b65\uff0c\u6211\u4eec\u7528\u6700\u65b0\u7684\u7b56\u7565\u6a21\u578b\u66ff\u6362\u53c2\u8003\u6a21\u578b\u3002\u4e3a\u52a0\u901f\u8bad\u7ec3\uff0c\u6bcf\u6b21 rollout \u4ea7\u751f 8,192 \u4e2a\u8f93\u51fa\uff0c\u968f\u673a\u5212\u5206\u4e3a 16 \u4e2a\u5c0f\u6279\u6b21\uff0c\u4ec5\u8bad\u7ec3\u4e00\u4e2a\u5185\u90e8 epoch\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u4e3a\u4e86\u51cf\u8f7b\u5728 RL \u8bad\u7ec3\u671f\u95f4\u8bed\u8a00\u6df7\u7528\u7684\u95ee\u9898\uff0c\u6211\u4eec\u5f15\u5165\u4e86\u8bed\u8a00\u4e00\u81f4\u6027\u5956\u52b1\uff0c\u5176\u8ba1\u7b97\u65b9\u5f0f\u662f\u6839\u636e\u5728 CoT \u4e2d\u76ee\u6807\u8bed\u8a00\u8bcd\u8bed\u6240\u5360\u7684\u6bd4\u4f8b\u3002<\/span><\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;text-align: center\"><img class=\"rich_pages wxw-img\" data-height=\"188\" data-imgfileid=\"100236281\" data-ratio=\"0.09907407407407408\" data-type=\"png\" data-w=\"1080\" data-width=\"1902\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-f370e70c5580a218bf8fbd3953ca9529.png\"><\/section>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u7b2c\u4e8c\u9636\u6bb5 RL \u8bad\u7ec3\u7ec6\u8282\u4e0e\u8bed\u8a00\u4e00\u81f4\u6027\u5956\u52b1<\/span><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u867d\u7136\u8865\u5145\u6750\u6599\u7b2c 2.6 \u8282\u4e2d\u7684\u6d88\u878d\u5b9e\u9a8c\u663e\u793a\uff0c\u8fd9\u79cd\u5bf9\u9f50\u4f1a\u4f7f\u6a21\u578b\u6027\u80fd\u7565\u6709\u4e0b\u964d\uff0c\u4f46\u8be5\u5956\u52b1\u7b26\u5408\u4eba\u7c7b\u504f\u597d\uff0c\u4f7f\u6587\u672c\u66f4\u6613\u8bfb\u3002\u6211\u4eec\u5c06\u8bed\u8a00\u4e00\u81f4\u6027\u5956\u52b1\u76f4\u63a5\u52a0\u5230\u6700\u7ec8\u5956\u52b1\u4e2d\uff0c\u540c\u65f6\u5e94\u7528\u4e8e\u63a8\u7406\u7c7b\u4e0e\u975e\u63a8\u7406\u7c7b\u6570\u636e\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u8bf7\u6ce8\u610f\uff0c\u88c1\u526a\u6bd4\u7387\uff08clip ratio\uff09\u5728\u8bad\u7ec3\u4e2d\u8d77\u7740\u5173\u952e\u4f5c\u7528\u3002\u6570\u503c\u8f83\u4f4e\u4f1a\u5bfc\u81f4\u5927\u91cf token \u7684\u68af\u5ea6\u88ab\u622a\u65ad\uff0c\u4ece\u800c\u964d\u4f4e\u6a21\u578b\u6027\u80fd\uff1b\u800c\u6570\u503c\u8f83\u9ad8\u5219\u53ef\u80fd\u5728\u8bad\u7ec3\u4e2d\u5f15\u53d1\u4e0d\u7a33\u5b9a\u6027\u3002\u672c\u9636\u6bb5\u6240\u7528 RL \u6570\u636e\u7684\u7ec6\u8282\u89c1\u8865\u5145\u6750\u6599\u7b2c 2.3 \u8282\u3002<\/span><\/span><\/p>\n<h3 style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px;font-weight: bold\">\u7b2c\u4e8c\u9636\u6bb5 RL \u7684\u8bad\u7ec3\u7ec6\u8282<\/span><\/span><\/h3>\n<p style=\"margin-left: 8px;margin-right: 8px;margin-bottom: 0px\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u5177\u4f53\u800c\u8a00\uff0c\u6211\u4eec\u4f7f\u7528\u5956\u52b1\u4fe1\u53f7\u4e0e\u591a\u6837\u5316\u63d0\u793a\u5206\u5e03\u7684\u7ec4\u5408\u6765\u8bad\u7ec3\u6a21\u578b\u3002\u5bf9\u4e8e\u63a8\u7406\u6570\u636e\uff0c\u6211\u4eec\u9075\u5faa DeepSeek\u2011R1\u2011Zero \u4e2d\u63cf\u8ff0\u7684\u65b9\u6cd5\uff0c\u4f7f\u7528\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\u6765\u5f15\u5bfc\u6570\u5b66\u3001\u7f16\u7a0b\u4e0e\u903b\u8f91\u63a8\u7406\u7b49\u9886\u57df\u7684\u5b66\u4e60\u3002\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u6211\u4eec\u89c2\u5bdf\u5230\uff0c\u5f53 RL \u63d0\u793a\u6d89\u53ca\u591a\u79cd\u8bed\u8a00\u65f6\uff0cCoT \u5e38\u4f1a\u51fa\u73b0\u8bed\u8a00\u6df7\u7528\u3002\u5bf9\u4e8e\u901a\u7528\u6570\u636e\uff0c\u6211\u4eec\u4f7f\u7528\u5956\u52b1\u6a21\u578b\u6765\u5f15\u5bfc\u8bad\u7ec3\u3002\u6700\u7ec8\uff0c\u878d\u5408\u591a\u79cd\u5956\u52b1\u4fe1\u53f7\u4e0e\u591a\u6837\u7684\u6570\u636e\u5206\u5e03\uff0c\u4f7f\u6211\u4eec\u80fd\u591f\u5f97\u5230\u4e00\u4e2a\u4e0d\u4ec5\u5728\u63a8\u7406\u4e0a\u8868\u73b0\u51fa\u8272\u3001\u540c\u65f6\u4e5f\u4f18\u5148\u8003\u8651\u6709\u7528\u6027\u4e0e\u65e0\u5bb3\u6027\u7684\u6a21\u578b\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u7ed9\u5b9a\u4e00\u6279\u6570\u636e\uff0c\u5956\u52b1\u53ef\u8868\u8ff0\u4e3a\uff1a<\/span><\/span><\/p>\n<section style=\"margin-left: 8px;margin-right: 8px;text-align: center\"><img class=\"rich_pages wxw-img\" data-height=\"648\" data-imgfileid=\"100236285\" data-ratio=\"0.3425925925925926\" data-type=\"png\" data-w=\"1080\" data-width=\"1894\" src=\"\/wp-content\/uploads\/2025\/09\/wxsync-2025-09-0fbe41d59e9e9f47f99901a43c1343ae.png\"><\/section>\n<p style=\"margin-left: 8px;margin-right: 8px\"><span><span style=\"font-size: 15px\">\u7b2c\u4e8c\u9636\u6bb5\u7684 RL \u57fa\u672c\u6cbf\u7528\u4e86\u7b2c\u4e00\u9636\u6bb5\u7684\u5927\u591a\u6570\u53c2\u6570\uff0c\u4e3b\u8981\u533a\u522b\u662f\u5c06\u6e29\u5ea6\u964d\u4f4e\u5230 0.7\uff0c\u56e0\u4e3a\u6211\u4eec\u53d1\u73b0\u5728\u8be5\u9636\u6bb5\u4f7f\u7528\u66f4\u9ad8\u6e29\u5ea6\u4f1a\u5bfc\u81f4\u751f\u6210\u7ed3\u679c\u4e0d\u8fde\u8d2f\u3002\u8be5\u9636\u6bb5\u5171\u5305\u542b 1,700 \u4e2a\u8bad\u7ec3\u6b65\uff0c\u5176\u4e2d\u901a\u7528\u6307\u4ee4\u6570\u636e\u4e0e\u57fa\u4e8e\u504f\u597d\u7684\u5956\u52b1\u4ec5\u5728\u6700\u540e 400 \u4e2a\u8bad\u7ec3\u6b65\u4e2d\u5f15\u5165\u3002\u6211\u4eec\u53d1\u73b0\uff0c\u4f7f\u7528\u57fa\u4e8e\u6a21\u578b\u7684\u504f\u597d\u5956\u52b1\u4fe1\u53f7\u8fdb\u884c\u66f4\u591a\u8bad\u7ec3\u6b65\u53ef\u80fd\u4f1a\u5bfc\u81f4\u201c\u5956\u52b1\u9ed1\u5ba2\u201d\u73b0\u8c61\uff0c\u76f8\u5173\u5185\u5bb9\u8bb0\u5f55\u4e8e\u8865\u5145\u6750\u6599\u7b2c 2.5 \u8282\u3002<\/span><\/span><\/p>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u6570\u636e\u53ef\u7528\u6027\uff08Data availability\uff09<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px;text-align: left\"><span><span style=\"font-size: 15px\">\u6211\u4eec\u5728\u4ee5\u4e0b\u94fe\u63a5\u63d0\u4f9b\u4e86\u7528\u4e8e\u62d2\u7edd\u91c7\u6837\u4e0e RL \u63d0\u793a\u7684\u6570\u636e\u6837\u672c\uff1ahttps:\/\/github.com\/deepseek-ai\/DeepSeek-R1\uff08https:\/\/doi.org\/10.5281\/zenodo.15753193\uff09\u3002 &nbsp;\u5173\u4e8e\u6211\u4eec\u5b8c\u6574\u6570\u636e\u751f\u6210\u65b9\u6cd5\u7684\u7efc\u5408\u7edf\u8ba1\u4e0e\u7ec6\u8282\uff0c\u89c1\u8865\u5145\u6750\u6599\u7b2c 2.3 \u8282\u3002<\/span><\/span><\/p>\n<h2 style=\"margin-left: 8px;margin-right: 8px\"><span><br \/><\/span><\/h2>\n<h3 style=\", Arial, sans-serif;letter-spacing: 0.544px\">\n<section style=\"letter-spacing: 0.544px;text-align: right;font-size: 13px\">\n<section style=\"margin-top: 10px;margin-bottom: 10px;letter-spacing: 0.544px;text-align: center\">\n<section style=\"vertical-align: middle\">\n<section style=\"margin-bottom: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"clear: both;line-height: 0\">\n<section style=\"line-height: 0;width: 0px\"><\/section>\n<\/section>\n<\/section>\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 1.4\">\n<p><strong><strong style=\"text-align: left;letter-spacing: 0.544px\"><span style=\", Arial, sans-serif\"><strong style=\"letter-spacing: 0.578px\"><span>\u4ee3\u7801\u53ef\u7528\u6027\uff08Code availability\uff09<\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n<section style=\"margin-top: -2px;, Arial, sans-serif\">\n<section style=\"float: left;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<section style=\"float: right;width: 8px;height: 3px;line-height: 0\"><span><br \/><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px;text-align: left\"><span><span style=\"font-size: 15px\">\u5728 MIT \u8bb8\u53ef\u8bc1\u4e0b\uff0cDeepSeek\u2011R1\u2011Zero \u4e0e DeepSeek\u2011R1 \u7684\u8bad\u7ec3\u6743\u91cd\u53ef\u5728\u6b64\u83b7\u53d6\uff1ahttps:\/\/github.com\/deepseek-ai\/DeepSeek-R1\uff08https:\/\/doi.org\/10.5281\/zenodo.15753193\uff09\u3002 &nbsp;\u63a8\u7406\u811a\u672c\u53d1\u5e03\u4e8e\uff1ahttps:\/\/github.com\/deepseek-ai\/DeepSeek-V3\uff08https:\/\/doi.org\/10.5281\/zenodo.15753347\uff09\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px;text-align: left\"><span><span style=\"font-size: 15px\">\u795e\u7ecf\u7f51\u7edc\u4f7f\u7528 PyTorch \u5f00\u53d1\uff0c\u5206\u5e03\u5f0f\u6846\u67b6\u57fa\u4e8e\u6211\u4eec\u5185\u90e8\u7684 HAI\u2011LLM \u6846\u67b6\uff08https:\/\/www.high-flyer.cn\/en\/blog\/hai-llm\uff09\u3002\u63a8\u7406\u6846\u67b6\u57fa\u4e8e vLLM\u3002<\/span><\/span><\/p>\n<p style=\"margin-left: 8px;margin-right: 8px;text-align: left\"><span><span style=\"font-size: 15px\">\u6570\u636e\u5206\u6790\u4f7f\u7528 Python v3.8\uff08https:\/\/www.python.org\/\uff09\u3001NumPy v.1.23.1\uff08https:\/\/github.com\/numpy\/numpy\uff09\u3001Matplotlib v.3.5.2\uff08https:\/\/github.com\/matplotlib\/matplotlib\uff09\u4ee5\u53ca TensorBoard v.2.9.1\uff08https:\/\/github.com\/tensorflow\/tensorboard\uff09\u3002<\/span><\/span><\/p>\n<section><span data-lark-record-format=\"docx\/record\" style=\"font-style: normal;font-weight: normal;letter-spacing: normal;text-align: start;text-indent: 0px;text-transform: none;text-decoration: none\"><\/span><\/section>\n<section style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.578px;line-height: 1.75em\" data-pm-slice=\"5 5 []\"><span><strong><span style=\"font-size: 15px\"><span><br \/><\/span><\/span><\/strong><\/span><\/section>\n<p style=\"text-align: left;line-height: 1em;margin-bottom: 8px;margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\"><span><br \/><\/span><\/span><\/p>\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;font-family: PingFangSC-light;font-size: 15px;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none\" data-pm-slice=\"0 0 []\">\n<section style=\"margin: 0px;padding: 0px 6px;max-width: 100%;width: 661px;vertical-align: top\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;line-height: 1.4\">\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: both;min-height: 1em\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span>\u53c2\u8003\u6587\u732e<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;font-family: PingFangSC-light;font-size: 15px;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-indent: 0px;text-transform: none;text-align: center;width: 661px\">\n<section style=\"margin: 0px;padding: 0px 0px 0px 4px;max-width: 100%;width: 661px;height: 360px;vertical-align: top\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;text-align: justify;font-size: 13px\">\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: both;min-height: 1em\"><span><br \/><\/span><\/p>\n<p style=\"text-align: left;line-height: 1em;margin-bottom: 8px;margin-left: 8px;margin-right: 8px\" data-pm-slice=\"3 2 []\"><span style=\"font-size: 13px\"><span>1. Brown, T. B. et al. Language models are few-shot learners. In Advances in Neural Information Processing Systems 33(eds Larochelle, H. et al.) (ACM,2020).<\/span><\/span><\/p>\n<p style=\"text-align: left;line-height: 1em;margin-bottom: 8px;margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\"><span><span>2. OpenAl etal. GPT4 technical report. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2303.08774(2024).<\/span><\/span><\/span><\/p>\n<p style=\"text-align: left;line-height: 1em;margin-bottom: 8px;margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\"><span>3. Wei, J.et al. Chain-of-thought prompting elicits reasoning in large language models. In Advancesin Neural Information Processing Systems 35 (eds Koyejo, S. et al.) 24824-24837(ACM,2022).<\/span><\/span><\/p>\n<p style=\"text-align: left;line-height: 1em;margin-bottom: 8px;margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\"><span>5. Kaplan, J. et al. Scaling laws for neural language models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2001.08361 (2020).<\/span><\/span><\/p>\n<p style=\"text-align: left;line-height: 1em;margin-bottom: 8px;margin-left: 8px;margin-right: 8px\"><span style=\"font-size: 13px\"><span>6. Kojima, T., Gu, S. S., Reid, M., Matsuo, Y. &amp; Iwasawa, Y. Large language models are zero-shot reasoners. In Advances in Neural Information Processing Systems 35 (eds Oh, A. H. et al.) &nbsp;22199\u201322213 (ACM, 2022).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>7. Chung, H. W. et al. Scaling instruction-finetuned language models. J. Mach. Learn. Res.&nbsp;<span>25, 1\u201353 (2024).<\/span><\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>8. DeepSeek-AI et al. DeepSeek-V3 technical report. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2412.19437 (2025).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>9. Shao, Z. et al. DeepSeekMath: pushing the limits of mathematical reasoning in open language models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2402.03300 (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>10. Wang, X. et al. Self-consistency improves chain of thought reasoning in language models. In 11th International Conference on Learning Representations (ICLR, 2023).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>11. Hendrycks, D. et al. Measuring massive multitask language understanding.In 9th International Conference on Learning Representations (ICLR, 2021).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>12. Gema, A. P. et al. Are we done with MMLU? In Proc. 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (eds Chiruzzo, L. et al.) Vol. 1 (Long Papers), 5069\u20135096 (ACL, 2025).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>13. Wang, Y. et al. MMLU-Pro: a more robust and challenging multi-task language understanding benchmark. In Advances in Neural Information Processing Systems 37(eds Globersons, A. et al.) 95266\u201395290 (ACM, 2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>14. Dua, D. et al. DROP: a reading comprehension benchmark requiring discrete reasoning over paragraphs. In Proc. 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies Vol. 1 (Long and Short Papers) (eds Burstein, J. et al.) 2368\u20132378 (ACL, 2019).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>15. Huang, Y. et al. C-EVAL: a multi-level multi-discipline Chinese evaluation suite for foundation models. In Advances in Neural Information Processing Systems 36(eds Oh, A. et al.) 62991\u201363010 (ACM, 2023).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>16. Zhou, J. et al. Instruction-following evaluation for large language models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2311.07911 (2023).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>17. Krishna, S. et al. Fact, fetch, and reason: a unified evaluation of retrieval-augmented generation. In Proc. 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies Vol. 1 (Long Papers) 4745\u20134759 (ACL, 2025).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>18. Rein, D. et al. GPQA: a graduate-level Google-proof Q&amp;A benchmark. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2311.12022 (2023).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>19. OpenAI. Introducing SimpleQA; https:\/\/openai.com\/index\/introducing-simpleqa\/ (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>20. He, Y. et al. Chinese SimpleQA: a Chinese factuality evaluation for large language models. In Proc. 63rd Annual Meeting of the Association for Computational LinguisticsVol. 1 (Long Papers), 19182\u201319208 (ACL, 2025).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>21. Xu, L. et al. CLUE: a Chinese Language Understanding Evaluation benchmark. In Proc. 28th International Conference on Computational Linguistics (eds Scott, D. et al.) 4762\u20134772&nbsp;<\/span><\/span><span style=\"font-size: 13px;, Arial, sans-serif;letter-spacing: 0.034em;background-color: transparent\"><span>(International Committee on Computational Linguistics, 2020).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>22. Dubois, Y., Galambosi, B., Liang, P. &amp; Hashimoto, T. B. Length-controlled AlpacaEval: a simple way to debias automatic evaluators. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2404.04475 (2025).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>23. Li, T. et al. From crowdsourced data to high-quality benchmarks: Arena-Hard and BenchBuilder pipeline. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2406.11939 (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>24. OpenAI. Introducing SWE-bench verified; https:\/\/openai.com\/index\/introducing-swebench-verified\/ (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>25. Aider. Aider LLM leaderboards; https:\/\/aider.chat\/docs\/leaderboards\/ (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>26. Jain, N. et al. LiveCodeBench: holistic and contamination free evaluation of large&nbsp;<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>language models for code. In 13th International Conference on Learning Representations(ICLR, 2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>27. Mirzayanov, M. Codeforces; https:\/\/codeforces.com\/ (2025).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>28. Chinese Mathematical Society (CMS). Chinese National High School Mathematics Olympiad; https:\/\/www.cms.org.cn\/Home\/comp\/comp\/cid\/12.html (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>29. Mathematical Association of America. American Invitational Mathematics Examination;&nbsp;<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>https:\/\/maa.org\/maa-invitational-competitions (2024).<\/span><\/span><\/p>\n<p style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>30. OpenAI. Hello GPT-4o; https:\/\/openai.com\/index\/hello-gpt-4o\/ (2024).<\/span><\/span><\/p>\n<section style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>31. Schulman, J., Wolski, F., Dhariwal, P., Radford, A. &amp; Klimov, O. Proximal policy optimization &nbsp;<span>algorithms. Preprint at https:\/\/doi.org\/10.48550\/arXiv.1707.06347 (2017).<\/span><\/span><\/span><\/section>\n<section style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>32. Ouyang, L. et al. Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems 35 (eds Koyejo, S. et al.) 27730\u201327744 (ACM, 2022).<\/span><\/span><\/section>\n<section style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>33. Nano et al. deepseek-ai\/DeepSeek-R1: v1.0.0. Zenodo https:\/\/doi.org\/10.5281\/zenodo.15753192 (2025).<\/span><\/span><\/section>\n<section style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>34. Yu, X. et al. deepseek-ai\/DeepSeek-V3: v1.0.0. Zenodo https:\/\/doi.org\/10.5281\/zenodo.15753346 (2025).<\/span><\/span><\/section>\n<section style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>35. Paszke, A. et al. PyTorch: an imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems 32 (eds Wallach, H. M. et al.) 8026\u20138037 (ACM, 2019).<\/span><\/span><\/section>\n<section style=\"text-align: left;margin-left: 8px;margin-right: 8px;line-height: 1em;margin-bottom: 8px\"><span style=\"font-size: 13px\"><span>36. Kwon, W. et al. Efficient memory management for large language model serving with PagedAttention. In Proc. ACM SIGOPS 29th Symposium on Operating Systems Principles611\u2013626 (ACM, 2023).<\/span><\/span><\/section>\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: both;min-height: 1em\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><span><br \/><\/span><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin: 3px 0px 0px;padding: 0px;max-width: 100%;font-family: PingFangSC-light;font-size: 15px;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none\">\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;font-size: 14px;letter-spacing: 1px\">\n<p style=\"margin: 0px;padding: 0px;max-width: 100%;clear: both;min-height: 1em;text-align: center\"><span>\u53c2\u8003\u6587\u732e\u53ef<\/span><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span>\u4e0a\u4e0b\u6ed1\u52a8<\/span><\/strong><span>\u67e5\u770b<\/span><\/p>\n<\/section>\n<\/section>\n<section style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.578px;line-height: 1.75em\" data-pm-slice=\"5 5 []\"><span><strong><span style=\"font-size: 15px\"><span><br \/><\/span><\/span><\/strong><\/span><\/section>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px\"><span><br \/><\/span><\/p>\n<section style=\"margin-top: 10px;margin-bottom: 0px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px;text-align: center;justify-content: center\">\n<section style=\"width: auto;vertical-align: top;min-width: 10%;height: auto\">\n<section style=\"margin-bottom: -15px\">\n<section style=\"padding-right: 10px;padding-left: 10px;font-size: 16px;line-height: 2;letter-spacing: 3px\">\n<p><strong><span>\u5927\u6a21\u578b2.0\u8bfb\u4e66\u4f1a<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<section style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px;clear: both;min-height: 1em;line-height: 2em\"><span><br \/><\/span><\/section>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px;line-height: 2em\"><span style=\"letter-spacing: 0.544px;font-size: 15px\"><span>o1\u6a21\u578b\u4ee3\u8868\u5927\u8bed\u8a00\u6a21\u578b\u878d\u5408\u5b66\u4e60\u4e0e\u63a8\u7406\u7684\u65b0\u8303\u5f0f\u3002\u96c6\u667a\u4ff1\u4e50\u90e8\u8054\u5408\u5317\u4eac\u5e08\u8303\u5927\u5b66\u7cfb\u7edf\u79d1\u5b66\u5b66\u9662\u6559\u6388\u5f20\u6c5f\u3001Google DeepMind\u7814\u7a76\u79d1\u5b66\u5bb6\u51af\u7199\u680b\u3001\u963f\u91cc\u5df4\u5df4\u5f3a\u5316\u5b66\u4e60\u7814\u7a76\u5458\u738b\u7ef4\u57d9\u548c\u4e2d\u79d1\u9662\u4fe1\u5de5\u6240\u5f20\u6770\u5171\u540c\u53d1\u8d77<a class=\"normal_text_link\" target=\"_blank\" href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247701119&amp;idx=1&amp;sn=52535e032aff6eb5a0cbd5e7244d14a7&amp;chksm=e89888f2dfef01e40f71af107c5b46866fb08094fbfa95638206bbc00f0ef802455a33b4da92&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\">\u300c\u5927\u6a21\u578bII\uff1a\u878d\u5408\u5b66\u4e60\u4e0e\u63a8\u7406\u7684\u5927\u6a21\u578b\u65b0\u8303\u5f0f \u300d\u8bfb\u4e66\u4f1a<\/a>\uff0c\u672c\u6b21\u8bfb\u4e66\u4f1a\u5c06\u5173\u6ce8\u5927\u6a21\u578b\u63a8\u7406\u8303\u5f0f\u7684\u6f14\u8fdb\u3001\u57fa\u4e8e\u641c\u7d22\u4e0e\u8499\u7279\u5361\u6d1b\u6811\u7684\u63a8\u7406\u4f18\u5316\u3001\u57fa\u4e8e\u5f3a\u5316\u5b66\u4e60\u7684\u5927\u6a21\u578b\u4f18\u5316\u3001\u601d\u7ef4\u94fe\u65b9\u6cd5\u4e0e\u5185\u5316\u673a\u5236\u3001\u81ea\u6211\u6539\u8fdb\u4e0e\u63a8\u7406\u9a8c\u8bc1\u3002\u5e0c\u671b\u901a\u8fc7\u8bfb\u4e66\u4f1a\u63a2\u7d22o1\u5177\u4f53\u5b9e\u73b0\u7684\u6280\u672f\u8def\u5f84\uff0c\u5e2e\u52a9\u6211\u4eec\u66f4\u597d\u7684\u7406\u89e3\u673a\u5668\u63a8\u7406\u548c\u4eba\u5de5\u667a\u80fd\u7684\u672c\u8d28\u3002<\/span><span style=\", Arial, sans-serif;font-size: 15px;letter-spacing: 0.034em\"><span style=\"font-size: 15px;letter-spacing: 0.578px\"><span style=\"letter-spacing: 0.578px\"><span data-pm-slice=\"1 1 [&quot;para&quot;,{&quot;tagName&quot;:&quot;section&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;-webkit-tap-highlight-color: transparent;margin-right: 8px;margin-left: 8px;outline: 0px;font-family: &quot;PingFang SC&quot;, system-ui, -apple-system, &quot;system-ui&quot;, &quot;Helvetica Neue&quot;, &quot;Hiragino Sans GB&quot;, &quot;Microsoft YaHei UI&quot;, &quot;Microsoft YaHei&quot;, Arial, sans-serif;letter-spacing: 0.544px;background-color: rgb(255, 255, 255);line-height: 2em;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;span&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;font-family: mp-quote, &quot;PingFang SC&quot;, system-ui, -apple-system, BlinkMacSystemFont, &quot;Helvetica Neue&quot;, &quot;Hiragino Sans GB&quot;, &quot;Microsoft YaHei UI&quot;, &quot;Microsoft YaHei&quot;, Arial, sans-serif;font-size: 15px;letter-spacing: 0.034em;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;span&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;font-size: 15px;letter-spacing: 0.578px;&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;},&quot;node&quot;,{&quot;tagName&quot;:&quot;span&quot;,&quot;attributes&quot;:{&quot;style&quot;:&quot;color: rgb(136, 136, 136);letter-spacing: 0.578px;text-decoration: none solid rgb(136, 136, 136);&quot;},&quot;namespaceURI&quot;:&quot;http:\/\/www.w3.org\/1999\/xhtml&quot;}]\">\u8bfb\u4e66\u4f1a\u5df2\u5b8c\u7ed3\uff0c\u73b0\u5728\u62a5\u540d\u53ef\u52a0\u5165\u793e\u7fa4\u5e76\u89e3\u9501\u56de\u653e\u89c6\u9891\u6743\u9650\u3002<\/span><\/span><\/span><\/span><\/span><\/p>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px;line-height: 2em\"><span style=\"letter-spacing: 0.544px\"><span><br \/><\/span><\/span><\/p>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px;line-height: 2em\"><span style=\"font-size: 15px;letter-spacing: 0.578px\"><span class=\"js_jump_icon h5_image_link\"><span><img class=\"rich_pages wxw-img\" data-backh=\"321\" data-backw=\"562\" data-imgfileid=\"100217811\" data-ratio=\"0.5712962962962963\" data-w=\"1080\" style=\"width: 100%;height: auto\" src=\"\"><img class=\"rich_pages wxw-img\" \/><img class=\"rich_pages wxw-img\" \/><img class=\"rich_pages wxw-img\" \/><img class=\"rich_pages wxw-img\" \/><\/span><\/span><\/span><\/p>\n<p style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px;line-height: 2em\"><span style=\"font-size: 15px;letter-spacing: 0.578px\"><span><br \/><\/span><\/span><\/p>\n<p style=\"margin-right: 8px;margin-bottom: 5px;margin-left: 8px;font-size: 14px;letter-spacing: 0.544px;, Arial, sans-serif\"><span style=\", Arial, sans-serif;font-size: 15px;letter-spacing: 0.544px\"><span>\u8be6\u60c5\u8bf7\u89c1\uff1a<a class=\"normal_text_link\" target=\"_blank\" href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247701119&amp;idx=1&amp;sn=52535e032aff6eb5a0cbd5e7244d14a7&amp;chksm=e89888f2dfef01e40f71af107c5b46866fb08094fbfa95638206bbc00f0ef802455a33b4da92&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\">\u5927\u6a21\u578b2.0\u8bfb\u4e66\u4f1a\uff1a\u878d\u5408\u5b66\u4e60\u4e0e\u63a8\u7406\u7684\u5927\u6a21\u578b\u65b0\u8303\u5f0f\uff01<\/a><\/span><\/span><\/p>\n<section style=\"margin-right: 8px;margin-bottom: 0px;margin-left: 8px;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.578px;line-height: 1.75em\"><span><br \/><\/span><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\" data-pm-slice=\"0 0 []\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px;font-size: 16px;text-align: center;, Arial, sans-serif\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;text-align: left;font-family: PingFangSC-light;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;font-size: 15px;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/span><\/strong><\/strong><\/strong><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px;font-size: 16px;text-align: center;, Arial, sans-serif\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;text-align: left;font-family: PingFangSC-light;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;font-size: 15px;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">\u63a8\u8350\u9605\u8bfb<\/span><\/span><\/strong><\/strong><\/strong><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><\/strong><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">1.&nbsp;<a class=\"normal_text_link\" target=\"_blank\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247719435&amp;idx=1&amp;sn=8e8aceda649659bbadeea38fefa68939&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"text-decoration: underline\">\u5927\u6a21\u578b\u81ea\u6307\uff1a\u4ee3\u7801\u751f\u6210\u7684\u9012\u5f52\u5f0f\u81ea\u6211\u63d0\u5347<\/span><\/a><\/span><\/strong><span style=\"margin: 0px;padding: 0px;max-width: 100%;text-decoration: underline\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/strong><\/span><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">2.&nbsp;<a class=\"normal_text_link\" target=\"_blank\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247707721&amp;idx=1&amp;sn=52c0370762a13598d3c76cf090ecfac3&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"text-decoration: underline\">DeepSeek-R1\uff5c\u96c6\u667a\u767e\u79d1<\/span><\/a><\/span><\/strong><span style=\"margin: 0px;padding: 0px;max-width: 100%;text-decoration: underline\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/strong><\/span><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">3.&nbsp;<a class=\"normal_text_link\" target=\"_blank\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247708181&amp;idx=1&amp;sn=14c67704bd57b72cd5fbc417e428236c&amp;scene=21#wechat_redirect\" data-itemshowtype=\"11\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"text-decoration: underline\">\u706b\u904d\u5168\u7403\u7684DeepSeek\uff1a\u79d1\u5b66\u5bb6\u662f\u600e\u4e48\u7528\u5b83\u7684\uff1f<\/span><\/a><\/span><\/strong><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><\/strong><\/section>\n<section data-pm-slice=\"4 5 []\" style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"text-decoration: none\">4.&nbsp;<\/span><a class=\"normal_text_link\" target=\"_blank\" style=\"margin: 0px;padding: 0px;text-decoration: none;cursor: default;max-width: 100%\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247710871&amp;idx=1&amp;sn=f572fc0ff9abbc6b32185b835ed39122&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"text-decoration: underline\">\u6d8c\u73b0\u52a8\u529b\u5b66\u5982\u4f55\u7528\u6765\u5206\u6790\u590d\u6742\u7cfb\u7edf\uff1f | \u65b0\u8bfe\u4e0a\u7ebf<\/span><\/a><\/span><\/strong><\/span><span style=\"margin: 0px;padding: 0px;max-width: 100%;text-decoration: underline\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/strong><\/strong><\/span><\/section>\n<section style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%;letter-spacing: 0.544px\"><\/strong><strong style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">5.&nbsp;<\/span><\/strong><strong style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><a class=\"normal_text_link\" target=\"_blank\" style=\"margin: 0px;padding: 0px;text-decoration: none;cursor: default;max-width: 100%\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247715376&amp;idx=2&amp;sn=e9b6f441a1a3615be72bb0b60594c015&amp;scene=21#wechat_redirect\" data-itemshowtype=\"11\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"text-decoration: underline\">\u96c6\u667a\u5b66\u56ed\u7cbe\u54c1\u8bfe\u7a0b\u514d\u8d39\u5f00\u653e\uff0c\u89e3\u9501\u7cfb\u7edf\u79d1\u5b66\u4e0e AI \u65b0\u4e16\u754c<\/span><\/a><\/span><\/strong><\/section>\n<p style=\"margin: 0px 8px 8px;padding: 0px;max-width: 100%;clear: both;min-height: 1em;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\">6.&nbsp;<a class=\"normal_text_link\" target=\"_blank\" style=\"margin: 0px;padding: 0px;text-decoration: none;cursor: default;max-width: 100%\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247716932&amp;idx=1&amp;sn=1fcb8a78a7f0157ad35a15d99f7ee9c1&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\"><span style=\"text-decoration: underline\">\u9ad8\u8003\u5206\u6570\u53ea\u662f\u5f20\u5165\u573a\u5238\uff0c\u4f60\u7684\u79d1\u7814\u5192\u9669\u5728\u8fd9\u91cc\u542f\u822a\uff01<\/span><\/a><\/span><\/strong><\/p>\n<p style=\"margin: 0px 8px;padding: 0px;max-width: 100%;clear: both;min-height: 1em;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><strong data-pm-slice=\"0 0 []\" style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 14px;font-style: normal;text-align: justify;text-indent: 0px;text-transform: none;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><span style=\"text-decoration: none\">7.&nbsp;<\/span><\/span><\/strong><\/strong><\/span><span style=\"margin: 0px;padding: 0px;max-width: 100%;text-decoration: underline\"><strong style=\"margin: 0px;padding: 0px;max-width: 100%\"><strong data-pm-slice=\"0 0 []\" style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 14px;font-style: normal;text-align: justify;text-indent: 0px;text-transform: none;letter-spacing: 0.544px\"><strong data-pm-slice=\"0 0 []\" style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 14px;font-style: normal;letter-spacing: 0.544px;text-align: justify;text-indent: 0px;text-transform: none\"><strong data-pm-slice=\"0 0 []\" style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 14px;font-style: normal;text-align: justify;text-indent: 0px;text-transform: none;letter-spacing: 0.544px\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><a class=\"normal_text_link\" target=\"_blank\" style=\"margin: 0px;padding: 0px;text-decoration: none;cursor: default;max-width: 100%\" href=\"https:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247716805&amp;idx=1&amp;sn=176d16c8f5fd5045f380e90ff9c16416&amp;scene=21#wechat_redirect\" data-itemshowtype=\"0\" data-linktype=\"2\" rel=\"noopener noreferrer\">\u91cd\u6574\u5316\u7fa4\u4e0e\u975e\u7ebf\u6027\u7269\u7406\uff0c\u5bfb\u627e\u590d\u6742\u7cfb\u7edf\u8de8\u5c3a\u5ea6\u7684\u5206\u6790\u65b9\u6cd5\u4e28\u65b0\u8bfe\u53d1\u5e03<\/a><\/span><\/strong><\/strong><\/strong><\/strong><\/span><\/p>\n<p style=\"margin: 0px 8px;padding: 0px;max-width: 100%;clear: both;min-height: 1em;font-style: normal;font-weight: 400;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.578px;line-height: 2em\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/p>\n<section style=\"margin: 0px 0px 24px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 17px;font-style: normal;font-weight: 400;letter-spacing: 0.544px;text-indent: 0px;text-transform: none;text-align: center\"><img alt=\"\u56fe\u7247\" class=\"rich_pages wxw-img\" data-backh=\"325\" data-backw=\"578\" data-imgfileid=\"100236176\" data-ratio=\"0.562962962962963\" data-s=\"300,640\" data-type=\"jpeg\" data-w=\"1080\" style=\"margin: 0px;padding: 0px;max-width: 100%;vertical-align: bottom;height: auto !important;width: 677px !important\" src=\"\"><\/section>\n<p style=\"margin: 0px 8px;padding: 0px;max-width: 100%;clear: both;min-height: 1em;font-style: normal;font-weight: 400;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.578px;line-height: 2em\"><span style=\"margin: 0px;padding: 0px;max-width: 100%;, Arial, sans-serif;font-size: 15px;letter-spacing: 0.578px;text-indent: 0em\"><span style=\"margin: 0px;padding: 0px;max-width: 100%\"><br \/><\/span><\/span><\/p>\n<section style=\"margin: 0px;padding: 0px;max-width: 100%;font-style: normal;font-weight: 400;text-align: justify;text-indent: 0px;text-transform: none;, Arial, sans-serif;font-size: 14px;letter-spacing: 0.578px;line-height: 2em\"><span style=\"margin-top: 0px;margin-right: 0px;margin-left: 0px;padding: 0px;max-width: 100%;text-indent: 0em;font-family: PingFangSC-light;font-size: 15px;font-weight: 700;letter-spacing: 0.544px;text-align: left\"><span style=\"margin-top: 0px;margin-right: 0px;margin-left: 0px;padding: 0px;max-width: 100%\">\u70b9\u51fb\u201c\u9605\u8bfb\u539f\u6587\u201d\uff0c\u62a5\u540d\u8bfb\u4e66\u4f1a<\/span><\/span><\/section>\n<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u5bfc\u8bed \u5c31\u5728\u521a\u521a\uff0c\u4e00\u7bc7\u9898\u76ee\u4e3a\uff1a\u300aDeepSeek-R1 \u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u6fc0\u52b1\u5927\u578b\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u63a8\u7406\u300b\uff08DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning\uff09\u7684\u6587\u7ae0\u767b\u4e0a\u4e86Nature\uff0c\u800c\u4e14\u8fd8\u662f\u5c01\u9762\u6587\u7ae0\u3002 \u5173\u952e\u8bcd\uff1aLL&#8230;<\/p>\n","protected":false},"author":0,"featured_media":64550,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/64573"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=64573"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/64573\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/64550"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=64573"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=64573"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=64573"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=64573"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}