{"id":25509,"date":"2025-04-21T09:34:06","date_gmt":"2025-04-21T01:34:06","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=25509"},"modified":"2025-04-21T09:34:06","modified_gmt":"2025-04-21T01:34:06","slug":"%e6%8e%a8%e7%90%86%e6%a8%a1%e5%9e%8b%e5%85%b6%e5%ae%9e%e6%97%a0%e9%9c%80%e3%80%8c%e6%80%9d%e8%80%83%e3%80%8d%ef%bc%9f%e4%bc%af%e5%85%8b%e5%88%a9%e5%8f%91%e7%8e%b0%e6%9c%89%e6%97%b6%e8%b7%b3%e8%bf%87","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2025\/04\/21\/25509\/","title":{"rendered":"\u63a8\u7406\u6a21\u578b\u5176\u5b9e\u65e0\u9700\u300c\u601d\u8003\u300d\uff1f\u4f2f\u514b\u5229\u53d1\u73b0\u6709\u65f6\u8df3\u8fc7\u601d\u8003\u8fc7\u7a0b\u4f1a\u66f4\u5feb\u3001\u66f4\u51c6\u786e"},"content":{"rendered":"<div><\/div>\n<div><img data-dominant-color=\"b5a2ad\" data-has-transparency=\"false\" style=\"--dominant-color: #b5a2ad;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-25511\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2025\/04\/84651f26719f1db410059306d68a5fa245cda3-300x167-1.jpg\" width=\"300\" height=\"167\" alt=\"\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2025\/04\/84651f26719f1db410059306d68a5fa245cda3-300x167-1.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2025\/04\/84651f26719f1db410059306d68a5fa245cda3-300x167-1-150x84.jpg 150w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/div>\n<div class=\"article-desc\">\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u751f\u6210\u89e3\u7b54\u4e4b\u524d\u4f1a\u4ea7\u751f\u5197\u957f\u7684\u601d\u8003\u8fc7\u7a0b\uff0c\u8fd9\u79cd\u65b9\u5f0f\u5728\u63a8\u7406\u4efb\u52a1\u4e0a\u5df2\u7ecf\u53d6\u5f97\u4e86\u5f88\u597d\u7684\u6210\u679c\u3002\u8be5\u7814\u7a76\u5bf9\u8fd9\u4e00\u8fc7\u7a0b\u7684\u5fc5\u8981\u6027\u63d0\u51fa\u4e86\u8d28\u7591\uff0c\u4e3a\u6b64\u5f15\u5165\u4e86 NoThinking \u65b9\u6cd5\u3002<\/div>\n<div id=\"postspictures\" class=\"article-content\">\n<div id=\"container\" class=\"container am-engine\" data-v-1d7a5742=\"\" data-element=\"root\">\n<p>\u5f53 DeepSeek-R1\u3001OpenAI o1 \u8fd9\u6837\u7684\u5927\u578b\u63a8\u7406\u6a21\u578b\u8fd8\u5728\u901a\u8fc7\u589e\u52a0\u63a8\u7406\u65f6\u7684\u8ba1\u7b97\u91cf\u63d0\u5347\u6027\u80fd\u65f6\uff0c\u52a0\u5dde\u5927\u5b66\u4f2f\u514b\u5229\u5206\u6821\u4e0e\u827e\u4f26\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u6240\u7a81\u7136\u6254\u51fa\u4e86\u4e00\u9897\u6df1\u6c34\u70b8\u5f39\uff1a\u522b\u518d\u5377 token \u4e86\uff0c\u65e0\u9700\u663e\u5f0f\u601d\u7ef4\u94fe\uff0c\u63a8\u7406\u6a21\u578b\u4e5f\u80fd\u5b9e\u73b0\u9ad8\u6548\u4e14\u51c6\u786e\u7684\u63a8\u7406\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202504\/20\/e54fcf871ed1eaac0b7499ef6bed53e31e9e8c.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u8fd9\u9879\u7814\u7a76\u8ba4\u4e3a\u663e\u5f0f\u601d\u8003\u8fc7\u7a0b\u4f1a\u663e\u8457\u589e\u52a0 token \u4f7f\u7528\u91cf\u548c\u5ef6\u8fdf\uff0c\u5bfc\u81f4\u63a8\u7406\u6548\u7387\u4f4e\u4e0b\u3002<\/p>\n<p>\u5c31\u6bd4\u5982\u5728\u63a7\u5236\u5ef6\u8fdf\u6761\u4ef6\u65f6\uff0cNoThinking \u7684\u6548\u679c\u5c31\u663e\u8457\u597d\u4e8e Thinking\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202504\/20\/934a1ec676651c1117042817a197b22853371b.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u8fd9\u9879\u7814\u7a76\u7ed9\u51fa\u7684\u51fa\u4eba\u610f\u6599\u7684\u7ed3\u8bba\u5438\u5f15\u4e86\u4e0d\u5c11\u773c\u7403\u3002\u6bd4\u5982\u4e9a\u9a6c\u900a\u7814\u7a76\u591a\u6a21\u6001 LLM \u7684\u535a\u58eb\u540e Gabriele Berton \u603b\u7ed3\u8bf4\uff1aNoThinking \u65b9\u6cd5\u672c\u8d28\u4e0a\u5c31\u662f\u5f3a\u5236\u6a21\u578b\u8f93\u51fa\uff1a\u300c\u601d\u8003\uff1a\u597d\u5427\uff0c\u6211\u60f3\u6211\u5df2\u7ecf\u601d\u8003\u5b8c\u4e86\u3002\u300d<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202504\/20\/8466b9f8843ed4f9ace6955d55d61a3501f384.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5177\u4f53\u600e\u4e48\u56de\u4e8b\uff1f\u6765\u770b\u4e00\u4e0b\u539f\u6587\uff1a<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s4.51cto.com\/oss\/202504\/20\/396cb004233040fc647070fd8d63967f32fbd1.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<ul data-id=\"ua73dd4b-TkgMVS6D\">\n<li data-id=\"ld70c578-26wdk596\">\u8bba\u6587\u6807\u9898\uff1aReasoning Models Can Be Effective Without Thinking<\/li>\n<li data-id=\"ld70c578-3H2DoVb2\">\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2504.09858<\/li>\n<\/ul>\n<h4>\u4e0e Thinking \u76f8\u6bd4\uff0cNoThinking \u80fd\u66f4\u597d\u5730\u6743\u8861\u7cbe\u5ea6\u4e0e\u9884\u7b97\u7684\u5173\u7cfb<\/h4>\n<p><strong>\u5b9a\u4e49 Thinking \u548c NoThinking\u00a0<\/strong><\/p>\n<p>\u5927\u591a\u6570\u73b0\u4ee3\u63a8\u7406\u6a21\u578b\uff0c\u5982 R1 \u548c R1-Distill-Qwen\uff0c\u5728\u5176\u751f\u6210\u8fc7\u7a0b\u4e2d\u90fd\u9075\u5faa\u7c7b\u4f3c\u7684\u7ed3\u6784\uff1a\u5728\u601d\u8003\u6846\u5185\u7684\u63a8\u7406\u8fc7\u7a0b\uff0c\u4ee5 &lt;|beginning_of_thinking|&gt; \u548c &lt;|end_of_thinking|&gt; \u4e3a\u6807\u5fd7\uff0c\u7136\u540e\u662f\u6700\u7ec8\u7b54\u6848\u3002\u57fa\u4e8e\u8fd9\u79cd\u7ed3\u6784\uff0c\u5c06\u4e24\u79cd\u65b9\u6cd5\uff08Thinking \u548c NoThinking\uff09\u5b9a\u4e49\u5982\u4e0b\u3002<\/p>\n<ul data-id=\"ua73dd4b-OQ7rEmME\">\n<li data-id=\"ld70c578-wwGIWc5i\">Thinking\u00a0\u6307\u7684\u662f\u67e5\u8be2\u63a8\u7406\u6a21\u578b\u4ee5\u751f\u6210\u4ee5\u4e0b\u8f93\u51fa\u7684\u9ed8\u8ba4\u65b9\u6cd5\uff1a\u601d\u8003\u6846\u5185\u7684\u63a8\u7406\u8fc7\u7a0b\u3001\u6700\u7ec8\u89e3\u51b3\u65b9\u6848\u548c\u6700\u7ec8\u7b54\u6848\uff08\u56fe 1\uff08\u84dd\u8272\uff09\uff09\u3002<\/li>\n<li data-id=\"ld70c578-lUy43Uxq\">NoThinking\u00a0\u6307\u7684\u662f\u901a\u8fc7\u63d0\u793a\u7ed5\u8fc7\u663e\u5f0f\u63a8\u7406\u8fc7\u7a0b\uff0c\u76f4\u63a5\u751f\u6210\u6700\u7ec8\u89e3\u51b3\u65b9\u6848\u548c\u7b54\u6848\u7684\u65b9\u6cd5\u3002\u8fd9\u53ef\u901a\u8fc7\u5728\u89e3\u7801\u8fc7\u7a0b\u4e2d\u5f3a\u5236\u601d\u7ef4\u6846\u4e3a\u7a7a\u6765\u5b9e\u73b0\uff08\u56fe 1\uff08\u6a59\u8272\uff09\uff09\uff0c\u5982\u4e0b\u6240\u793a\u3002<\/li>\n<\/ul>\n<div>\n<div class=\"hljs-cto\">\n<div class=\"hljs-cto\"><button class=\"copy_btn disable\" data-clipboard-target=\"#code_id_0\">\u590d\u5236<\/button><\/p>\n<div class=\"code-toolbar\">\n<pre class=\"has-pre-numbering language-javascript\" tabindex=\"0\"><code class=\"language-javascript\"><span class=\"token operator\">&lt;<\/span><span class=\"token operator\">|<\/span>beginning_of_thinking<span class=\"token operator\">|<\/span><span class=\"token operator\">&gt;<\/span> \r\nOkay<span class=\"token punctuation\">,<\/span> <span class=\"token constant\">I<\/span> think <span class=\"token constant\">I<\/span> have finished thinking<span class=\"token punctuation\">.<\/span> \r\n<span class=\"token operator\">&lt;<\/span><span class=\"token operator\">|<\/span>end_of_thinking<span class=\"token operator\">|<\/span><span class=\"token operator\">&gt;<\/span><\/code><\/pre>\n<ul id=\"code_id_0\" class=\"pre-numbering\">\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<\/ul>\n<div class=\"toolbar\"><\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<p>\u4e3a\u4e86\u63a7\u5236\u4e24\u79cd\u65b9\u6cd5\u4e2d\u7684 token \u4f7f\u7528\u91cf\uff0c\u5f53\u6a21\u578b\u8fbe\u5230 token \u9884\u7b97\u65f6\uff0c\u5b83\u5c06\u88ab\u8feb\u751f\u6210\u300c\u6700\u7ec8\u7b54\u6848\u300d\uff0c\u4ee5\u786e\u4fdd\u7acb\u5373\u5f97\u5230\u6700\u7ec8\u7b54\u6848\u3002\u5982\u679c\u6a21\u578b\u5728\u8fbe\u5230 token \u9650\u5236\u65f6\u4ecd\u5728\u601d\u8003\u6846\u5185\uff0c&lt;|end_of_thinking|&gt; \u5c06\u9644\u52a0\u5728\u6700\u7ec8\u7b54\u6848\u6807\u7b7e\u4e4b\u524d\u3002<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s9.51cto.com\/oss\/202504\/20\/51257ba36f29b843ed9954a6be159030978092.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p><strong>\u5b9e\u9a8c\u8bbe\u7f6e<\/strong><\/p>\n<p>\u5b9e\u9a8c\u4f7f\u7528 DeepSeek-R1-Distill-Qwen-32B \u4f5c\u4e3a\u4e3b\u8981\u6a21\u578b\u3002\u5b83\u662f DeepSeek-R1 \u7684\u4e00\u79cd\u84b8\u998f\u7248\uff0c\u901a\u8fc7\u7528 Qwen- 32B \u521d\u59cb\u5316\u6a21\u578b\u5e76\u5728 DeepSeek-R1 \u751f\u6210\u7684\u6570\u636e\u4e0a\u8fdb\u884c\u8bad\u7ec3\u800c\u521b\u5efa\u3002\u6839\u636e\u62a5\u544a\uff0c\u5b83\u662f\u4f7f\u7528\u987a\u5e8f\u6d4b\u8bd5\u65f6\u95f4\u6269\u5c55\u7684\u6700\u5148\u8fdb\u63a8\u7406\u6a21\u578b\u4e4b\u4e00\uff0c\u4e0e\u89c4\u6a21\u66f4\u5927\u7684 DeepSeek R1-Distill-Llama-70B \u4e0d\u76f8\u4e0a\u4e0b\u3002<\/p>\n<p>\u8fd8\u63d0\u4f9b\u4e86\u4e00\u7cfb\u5217\u5177\u6709\u6311\u6218\u6027\u7684\u63a8\u7406\u57fa\u51c6\uff0c\u5305\u62ec\u6570\u5b66\u7ade\u8d5b\u3001\u7f16\u7801\u3001\u5965\u6797\u5339\u514b\u7ade\u8d5b\u95ee\u9898\u548c\u5b9a\u7406\u8bc1\u660e\u7b49\u4efb\u52a1\u3002\u53c8\u4ee5\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\u4e3a\u6307\u6807\uff0c\u5b83\u8861\u91cf\u7684\u662f\u5728\u6bcf\u4e2a\u95ee\u9898\u6240\u751f\u6210\u7684 n \u4e2a\u5b8c\u6574\u56de\u590d\u4e2d\uff0c\u968f\u673a\u9009\u53d6 k \u4e2a\u6837\u672c\uff0c\u5176\u4e2d\u81f3\u5c11\u6709\u4e00\u4e2a\u6b63\u786e\u8f93\u51fa\u7684\u6982\u7387\u3002<\/p>\n<p>\u5176\u5f62\u5f0f\u5316\u5b9a\u4e49\u4e3a<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s2.51cto.com\/oss\/202504\/20\/a1c48d494e60312cd4e7560798462ece99451e.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u5176\u4e2d n \u662f\u6bcf\u4e2a\u95ee\u9898\u7684\u91c7\u6837\u8f93\u51fa\u6570\u91cf\uff0cc \u662f\u6b63\u786e\u8f93\u51fa\u7684\u6570\u91cf\u3002<\/p>\n<p>\u5bf9\u4e8e\u5b9a\u7406\u8bc1\u660e\u6570\u636e\u96c6\uff08MiniF2F \u548c ProofNet\uff09\uff0c\u6309\u7167\u6807\u51c6\u4f7f\u7528 k = {1, 2, 4, 8, 16, 32}\uff1b\u5bf9\u4e8e\u8f83\u5c0f\u7684\u6570\u636e\u96c6\uff082024 \u5e74\u7f8e\u56fd\u6570\u5b66\u9080\u8bf7\u8d5b\u30012025 \u5e74\u7f8e\u56fd\u6570\u5b66\u9080\u8bf7\u8d5b\u30012023 \u5e74\u7f8e\u56fd\u6570\u5b66\u7ade\u8d5b\uff09\uff0ck = {1, 2, 4, 8, 16, 32, 64}\uff1b\u5bf9\u4e8e\u8f83\u5927\u7684\u6570\u636e\u96c6\uff08\u5965\u6797\u5339\u514b\u7ade\u8d5b\u57fa\u51c6\u6d4b\u8bd5\u3001\u5b9e\u65f6\u7f16\u7801\u57fa\u51c6\u6d4b\u8bd5\uff09\uff0ck = {1, 2, 4, 8, 16}\u3002\u5bf9\u4e8e\u5f62\u5f0f\u5b9a\u7406\u8bc1\u660e\u57fa\u51c6\u6d4b\u8bd5\uff0c\u300c\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@32\uff09\u300d \u662f\u6807\u51c6\u6307\u6807\uff0c\u800c\u5bf9\u4e8e\u6570\u5b66\u548c\u7f16\u7a0b\u57fa\u51c6\u6d4b\u8bd5\uff0c\u6700\u5e38\u7528\u7684\u662f\u300c\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u300d\uff08\u5373\u51c6\u786e\u7387\uff09\u3002<\/p>\n<p><strong>\u5b9e\u9a8c\u7ed3\u679c<\/strong><\/p>\n<p>\u5728\u672a\u63a7\u5236 token \u9884\u7b97\u7684\u60c5\u51b5\u4e0b\u5bf9 Thinking\u3001NoThinking \u4e0e Qwen Instruct \u8fdb\u884c\u5bf9\u6bd4<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s3.51cto.com\/oss\/202504\/20\/99aaadd26f90fdd1eb16858ed00faf8dc74390.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u9996\u5148\uff0c\u5728 MiniF2F \u548c ProofNet \u4e0a\uff0cNoThinking \u5728\u6240\u6709 k \u503c\u4e0a\u7684\u8868\u73b0\u4e0eThinking \u76f8\u5f53\uff08\u4e24\u8005\u90fd\u663e\u8457\u4f18\u4e8e Qwen-Instruct\uff09\u3002\u8003\u8651\u5230 NoThinking \u4f7f\u7528\u7684 token \u6bd4 Thinking \u5c11 3.3\u20133.7 \u500d\uff0c\u8fd9\u4e00\u7ed3\u679c\u5c24\u5176\u4ee4\u4eba\u60ca\u8bb6\u3002\u5728\u5176\u4ed6\u6570\u636e\u96c6\u4e0a\uff0c\u7ed3\u679c\u5219\u66f4\u4e3a\u590d\u6742\u3002\u5728 k=1 \u65f6\uff0cNoThinking \u843d\u540e\u4e8e Thinking\uff0c\u4f46\u968f\u7740 k \u7684\u589e\u52a0\uff0c\u5dee\u8ddd\u9010\u6e10\u7f29\u5c0f\u3002<\/p>\n<p>\u5728\u6240\u6709\u6570\u636e\u96c6\u4e2d\uff0c\u5f53 k \u503c\u6700\u5927\u65f6\uff0cNoThinking\u00a0\u7684\u8868\u73b0\u4e0e\u00a0Thinking\u00a0\u76f8\u5f53\uff0c\u4f46 token \u4f7f\u7528\u91cf\u5374\u6bd4\u00a0Thinking\u00a0\u5c11 2.0\u20135.1 \u500d\u3002\u5728 AIME24\u3001AIME25 \u548c LiveCodeBench \u4e0a\uff0cThinking\u00a0\u548c\u00a0NoThinking\u00a0\u90fd\u663e\u8457\u4f18\u4e8e Qwen-Instruct\u3002\u7136\u800c\uff0c\u5728 AMC23 \u548c OlympiadBench \u4e0a\uff0cQwen-Instruct \u4e5f\u7f29\u5c0f\u4e86\u4e0e Thinking \u548c NoThinking \u7684\u5dee\u8ddd\u3002<\/p>\n<p>\u5728\u00a0token \u9884\u7b97\u63a7\u5236\u4e0b\u7684\u60c5\u51b5\u4e0b\u5bf9 Thinking \u548c NoThinking \u8fdb\u884c\u5bf9\u6bd4<\/p>\n<p>\u9664 MiniF2F \u548c ProofNet \u5916\uff0cNoThinking \u5728\u5176\u4ed6\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\u867d\u7a0d\u900a\u4e8e Thinking\uff0c\u4f46\u5176 token \u6d88\u8017\u91cf\u4e5f\u663e\u8457\u66f4\u4f4e\u3002<\/p>\n<p>\u56e0\u6b64\uff0c\u63a5\u4e0b\u6765\u7ee7\u7eed\u901a\u8fc7\u9884\u7b97\u7ea6\u675f\u65b9\u6cd5\uff0c\u5728\u76f8\u8fd1 token \u9884\u7b97\u4e0b\u6bd4\u8f83\u4e24\u8005\u7684\u6027\u80fd\u8868\u73b0\u3002<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s4.51cto.com\/oss\/202504\/20\/459e68e864d6a985e4a45854d641801022d7e4.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u5982\u56fe 5 \u6240\u793a\uff0c\u5f53 token \u4f7f\u7528\u91cf\u53d7\u5230\u63a7\u5236\u65f6\uff0cNoThinking \u65b9\u6cd5\u901a\u5e38\u4f18\u4e8e Thinking \u65b9\u6cd5\u3002\u7279\u522b\u662f\u5728\u4f4e\u9884\u7b97\u8bbe\u7f6e\u4e0b\uff08\u4f8b\u5982\uff0c\u4f7f\u7528\u7684 token \u6570\u5c11\u4e8e\u7ea6 3000 \u4e2a\uff09\uff0c\u5728\u6240\u6709\u7684 k \u503c\u60c5\u51b5\u4e0b\uff0cNoThinking \u65b9\u6cd5\u59cb\u7ec8\u6bd4 Thinking \u65b9\u6cd5\u53d6\u5f97\u66f4\u597d\u7684\u7ed3\u679c\uff0c\u5e76\u4e14\u968f\u7740 k \u503c\u7684\u589e\u52a0\uff0c\u6027\u80fd\u5dee\u8ddd\u4f1a\u8fdb\u4e00\u6b65\u6269\u5927\u3002\u5f53 token \u4f7f\u7528\u91cf\u8f83\u9ad8\u65f6\uff08\u4f8b\u5982\uff0c\u5927\u7ea6 3500 \u4e2a token\uff09\uff0c\u5728\u5355\u6837\u672c\u51c6\u786e\u7387 (pass@1\uff09\u65b9\u9762\uff0cThinking \u65b9\u6cd5\u7684\u8868\u73b0\u4f18\u4e8e NoThinking \u65b9\u6cd5\uff0c\u4f46\u4ece k = 2 \u5f00\u59cb\uff0cNoThinking \u65b9\u6cd5\u7684\u8868\u73b0\u5c31\u8fc5\u901f\u8d85\u8fc7\u4e86 Thinking \u65b9\u6cd5\u3002<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s2.51cto.com\/oss\/202504\/20\/29ef857773a6120920d37765efa021c74ae0d7.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 6 \u901a\u8fc7\u5c06 token \u4f7f\u7528\u91cf\u7ed8\u5236\u5728\u6a2a\u8f74\u4e0a\uff0c\u8fdb\u4e00\u6b65\u8bf4\u660e\u4e86\u8fd9\u4e9b\u7ed3\u679c\uff0c\u540c\u65f6\u6bd4\u8f83\u4e86\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u548c\u5728\u53ef\u7528\u6700\u5927 k \u503c\u4e0b\u7684\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\u3002\u5728\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\u65b9\u9762\uff0cNoThinking \u65b9\u6cd5\u5728\u6574\u4e2a\u9884\u7b97\u8303\u56f4\u5185\u59cb\u7ec8\u65e0\u4e00\u4f8b\u5916\u5730\u4f18\u4e8e Thinking \u65b9\u6cd5\u3002\u5bf9\u4e8e\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u800c\u8a00\uff0cNoThinking \u65b9\u6cd5\u5728\u4f4e\u9884\u7b97\u60c5\u51b5\u4e0b\u8868\u73b0\u66f4\u597d\uff0c\u800c\u5728\u9ad8\u9884\u7b97\u60c5\u51b5\u4e0b\u8868\u73b0\u8f83\u5dee\u3002\u5b9e\u65f6\u7f16\u7801\u57fa\u51c6\u6d4b\u8bd5\u662f\u4e2a\u4f8b\u5916\uff0c\u5728\u8be5\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u5373\u4f7f\u662f\u5728\u4f4e\u9884\u7b97\u60c5\u51b5\u4e0b\uff0cThinking \u65b9\u6cd5\u5728\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u65b9\u9762\u4e5f\u4f18\u4e8e NoThinking \u65b9\u6cd5\u3002\u8fd9\u5f88\u53ef\u80fd\u662f\u56e0\u4e3a\u5728\u5b9e\u65f6\u7f16\u7801\u57fa\u51c6\u6d4b\u8bd5\u4e0a\uff0c\u7981\u7528\u601d\u8003\u6a21\u5757\u5e76\u4e0d\u80fd\u663e\u8457\u51cf\u5c11 token \u4f7f\u7528\u91cf\u3002<\/p>\n<p>\u4e3a\u4e86\u964d\u4f4e\u6570\u636e\u6c61\u67d3\u7684\u98ce\u9669\uff0c\u5b9e\u9a8c\u8fd8\u7eb3\u5165\u4e86\u65b0\u53d1\u5e03\u7684 \u00a0AIME 2025 \uff0c\u8fd9\u4e9b\u6570\u636e\u4e0d\u592a\u53ef\u80fd\u51fa\u73b0\u5728\u73b0\u6709\u6a21\u578b\u7684\u9884\u8bad\u7ec3\u6570\u636e\u4e2d\u3002\u91cd\u8981\u7684\u662f\uff0c\u5728\u65b0\u7684\u57fa\u51c6\u6d4b\u8bd5\u548c\u5df2\u6709\u7684\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u90fd\u5f97\u5230\u4e86\u4e00\u81f4\u7684\u7ed3\u679c\uff0c\u8fd9\u8868\u660e\u7814\u7a76\u6240\u89c2\u5bdf\u5230\u7684\u8d8b\u52bf\u5e76\u975e\u662f\u6a21\u578b\u8bb0\u5fc6\u7684\u4ea7\u7269\uff0c\u800c\u662f\u53cd\u6620\u4e86\u6a21\u578b\u5177\u6709\u53ef\u6cdb\u5316\u7684\u884c\u4e3a\u8868\u73b0\u3002<\/p>\n<p><strong>\u589e\u5927 k \u503c\u5bf9 NoThinking \u65b9\u6cd5\u6027\u80fd\u7684\u5f71\u54cd<\/strong><\/p>\n<p>\u8be5\u56e2\u961f\u7814\u7a76\u4e86\u968f\u7740 k \u503c\u589e\u52a0\uff0c\u6240\u89c2\u5bdf\u5230\u7684 NoThinking \u65b9\u6cd5\u6027\u80fd\u53d8\u5316\u7684\u6f5c\u5728\u539f\u56e0\uff0c\u4ed6\u4eec\u91cd\u70b9\u5173\u6ce8\u4e86\u751f\u6210\u7b54\u6848\u7684\u591a\u6837\u6027\u3002\u8fd9\u662f\u901a\u8fc7\u8ba1\u7b97\u6bcf\u4e2a\u95ee\u9898\u7684\u7b54\u6848\u5206\u5e03\u7684\u71b5\u6765\u8861\u91cf\u7684\u3002\u5177\u4f53\u800c\u8a00\uff0c\u5bf9\u4e8e\u7b54\u6848\u5206\u5e03\u4e3a<img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s7.51cto.com\/oss\/202504\/20\/216961381e512811eeb453bac8d9770aac734c.webp\" alt=\"image.png\" data-type=\"inline\" \/>\u7684\u4e00\u4e2a\u95ee\u9898\uff0c\u71b5\u7684\u5b9a\u4e49\u4e3a\uff1a<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s7.51cto.com\/oss\/202504\/20\/a8b887222db12e611460489adb48ff55b4f24e.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u5176\u4e2d p_i \u662f\u7b2c i \u4e2a\u72ec\u7279\u7b54\u6848\u7684\u7ecf\u9a8c\u6982\u7387\u3002\u7136\u540e\uff0c\u4f7f\u7528\u6240\u6709\u95ee\u9898\u7684\u71b5\u7684\u5747\u503c\u548c\u6807\u51c6\u5dee\u6765\u603b\u7ed3\u591a\u6837\u6027\u3002\u5747\u503c\u71b5\u8d8a\u9ad8\u8868\u660e\u603b\u4f53\u591a\u6837\u6027\u8d8a\u5927\uff0c\u800c\u6807\u51c6\u5dee\u8d8a\u4f4e\u5219\u610f\u5473\u7740\u5404\u4e2a\u95ee\u9898\u4e4b\u95f4\u7684\u591a\u6837\u6027\u66f4\u4e3a\u4e00\u81f4\u3002\u8fd9\u4e9b\u5206\u6790\u57fa\u4e8e\u56fe 5 \u4e2d Thinking \u65b9\u6cd5\u4e0e NoThinking \u65b9\u6cd5\u7684\u5bf9\u6bd4\u60c5\u51b5\uff0c\u4f46\u4e0d\u5305\u62ec\u7f3a\u5c11\u786e\u5207\u7b54\u6848\u7684\u5b9e\u65f6\u7f16\u7801\u57fa\u51c6\u6d4b\u8bd5\u3002<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s4.51cto.com\/oss\/202504\/20\/a6abb1f755d50cb8c159120919962d5f647f98.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u4ece\u8868 1 \u53ef\u4ee5\u770b\u5230\uff0c\u5c31\u5e73\u5747\u591a\u6837\u6027\u800c\u8a00\uff0c\u4e24\u79cd\u6a21\u5f0f\u90fd\u6ca1\u6709\u59cb\u7ec8\u5982\u4e00\u7684\u4f18\u52bf\u3002\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\uff0c NoThinking \u6a21\u5f0f\u5f97\u51fa\u7684\u5e73\u5747\u71b5\u66f4\u9ad8\uff1b\u800c\u5728\u53e6\u4e00\u4e9b\u60c5\u51b5\u4e0b\uff0c Thinking \u00a0\u6a21\u5f0f\u7684\u5e73\u5747\u71b5\u66f4\u9ad8\u3002\u7136\u800c\uff0c NoThinking \u6a21\u5f0f\u5728\u5404\u4e2a\u95ee\u9898\u4e0a\u59cb\u7ec8\u8868\u73b0\u51fa\u66f4\u4f4e\u7684\u65b9\u5dee\uff0c\u8fd9\u8868\u660e NoThinking \u6a21\u5f0f\u751f\u6210\u7684\u7b54\u6848\u5728\u4e0d\u540c\u793a\u4f8b\u4e4b\u95f4\u5177\u6709\u66f4\u5747\u5300\u7684\u591a\u6837\u6027\u3002\u7814\u7a76\u8005\u4eec\u63a8\u6d4b\uff0c\u8fd9\u79cd\u591a\u6837\u6027\u4e00\u81f4\u6027\u7684\u63d0\u9ad8\u53ef\u80fd\u662f\u968f\u7740 k \u503c\u589e\u52a0\uff0c NoThinking \u6a21\u5f0f\u5728\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\u4e0a\u8868\u73b0\u63d0\u5347\u7684\u4e00\u4e2a\u539f\u56e0\uff0c\u5c3d\u7ba1\u4ec5\u9760\u591a\u6837\u6027\u5e76\u4e0d\u80fd\u5b8c\u5168\u89e3\u91ca\u6027\u80fd\u5dee\u5f02\u3002<\/p>\n<h4>NoThinking \u65b9\u6cd5\u4f7f\u6d4b\u8bd5\u9636\u6bb5\u7684\u5e76\u884c\u8ba1\u7b97\u66f4\u52a0\u9ad8\u6548<\/h4>\n<p><strong>\u5e76\u884c scaling \u4e0e\u987a\u5e8f scaling<\/strong><\/p>\n<p>\u5e76\u884c scaling \u672c\u8d28\u4e0a\u80fd\u591f\u5b9e\u73b0\u4f4e\u5ef6\u8fdf\uff0c\u56e0\u4e3a\u591a\u4e2a\u6a21\u578b\u8c03\u7528\u53ef\u4ee5\u540c\u65f6\u6267\u884c \u2014\u2014 \u65e0\u8bba\u662f\u901a\u8fc7\u5e94\u7528\u7a0b\u5e8f\u7f16\u7a0b\u63a5\u53e3\u8c03\u7528\u8fd8\u662f\u672c\u5730\u6a21\u578b\u670d\u52a1\u6765\u5b9e\u73b0\u3002\u8fd9\u53ef\u4ee5\u901a\u8fc7\u591a GPU \u8bbe\u7f6e\u6216\u8005\u5728\u5355\u4e2a GPU \u4e0a\u8fdb\u884c\u6279\u5904\u7406\u6765\u8fbe\u6210\uff0c\u4e0e\u987a\u5e8f scaling \u76f8\u6bd4\uff0c\u8fd9\u79cd\u65b9\u5f0f\u80fd\u591f\u5b9e\u73b0\u66f4\u9ad8\u7684 GPU \u5229\u7528\u7387\u3002\u603b\u4f53\u5ef6\u8fdf\u7531\u5355\u4e2a\u6700\u957f\u7684\u751f\u6210\u65f6\u95f4\u51b3\u5b9a\u3002<\/p>\n<p>\u9274\u4e8e\u5b9e\u9a8c\u53d1\u73b0 NoThinking \u65b9\u6cd5\u5728\u4f4e\u9884\u7b97\u60c5\u51b5\u4e0b\u80fd\u751f\u6210\u66f4\u51c6\u786e\u7684\u89e3\u51b3\u65b9\u6848\uff0c\u5e76\u4e14\u968f\u7740 k \u503c\u7684\u589e\u52a0\uff0c\u5728\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\u65b9\u9762\u7684\u6548\u679c\u8d8a\u6765\u8d8a\u597d\u3002\u8fd9\u8bc1\u660e\u4e86\uff0c\u5f53 NoThinking \u65b9\u6cd5\u4e0e\u7b80\u5355\u7684 \u300c\u4ece N \u4e2a\u4e2d\u9009\u6700\u4f73\uff08Best-of-N\uff09\u300d\u65b9\u6cd5\u76f8\u7ed3\u5408\u65f6\uff0c\u91c7\u7528\u5e76\u884c\u91c7\u6837\u7684 NoThinking \u65b9\u6cd5\u80fd\u663e\u8457\u63d0\u9ad8\u51c6\u786e\u7387\u3002\u5728\u5ef6\u8fdf\u76f8\u5f53\u7684\u60c5\u51b5\u4e0b\uff0c\u5b83\u7684\u8868\u73b0\u4f18\u4e8e\u5176\u4ed6\u65b9\u6cd5\uff0c\u6bd4\u5982\u91c7\u7528\u5f3a\u5236\u9884\u7b97\u548c\u5e76\u884c\u91c7\u6837\u7684 Thinking \u65b9\u6cd5\u3002\u800c\u4e14\uff0c\u5c3d\u7ba1\u5176\u4ea7\u751f\u7684\u5ef6\u8fdf\u8981\u4f4e\u5f97\u591a\uff0c\u4f46\u5b83\u751a\u81f3\u5728\u987a\u5e8f scaling \u7684\u60c5\u51b5\u4e0b\u8d85\u8fc7\u4e86\u5b8c\u6574 Thinking \u65b9\u6cd5\uff08\u5373\u4e0d\u91c7\u7528\u5f3a\u5236\u9884\u7b97\u7684 Thinking \u65b9\u6cd5\uff09\u7684\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u6027\u80fd\u3002<\/p>\n<p><strong>\u7ed3\u679c<\/strong><\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s6.51cto.com\/oss\/202504\/20\/8728151442b01cc72ad6646415ab3c7353f2fa.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 7 \u4e2d\u5c55\u793a\u4e86 Thinking \u65b9\u6cd5\u548c NoThinking \u65b9\u6cd5\u5728\u6240\u6709\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u7684\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u7ed3\u679c\u3002\u5355\u4e2a\u91c7\u6837\u54cd\u5e94\u7684\u6027\u80fd\u88ab\u79f0\u4e3a\u65e0\u5e76\u884c scaling \u60c5\u51b5\u4e0b\u7684\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\uff0c\u800c\u5bf9\u591a\u4e2a\u6837\u672c\u8fdb\u884c \u300c\u4ece N \u4e2a\u4e2d\u9009\u6700\u4f73\u300d\u9009\u62e9\u540e\u7684\u51c6\u786e\u7387\u5219\u88ab\u89c6\u4e3a\u6709\u5e76\u884c scaling \u60c5\u51b5\u4e0b\u7684\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u3002\u5bf9\u4e8e\u6ca1\u6709\u9a8c\u8bc1\u5668\u7684\u4efb\u52a1\uff0c\u5728\u56fe\u4e2d\u4f7f\u7528\u57fa\u4e8e\u7f6e\u4fe1\u5ea6\u7684\u7ed3\u679c\uff0c\u5e76\u5728\u8868 2 \u4e2d\u7ed9\u51fa\u9009\u5b9a\u5b9e\u9a8c\u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\u3002\u8be5\u8868\u6bd4\u8f83\u4e86\u4e0a\u8ff0\u8ba8\u8bba\u7684\u300c\u4ece N \u4e2a\u4e2d\u9009\u6700\u4f73\u300d\u65b9\u6cd5\u3002\u57fa\u4e8e\u7f6e\u4fe1\u5ea6\u7684\u9009\u62e9\u65b9\u6cd5\u901a\u5e38\u4f18\u4e8e\u591a\u6570\u6295\u7968\u6cd5\u3002\u8fd8\u62a5\u544a\u4e86\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\uff0c\u5c06\u5176\u4f5c\u4e3a\u4f7f\u7528\u5e76\u884c scaling \u65f6\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u7684\u4e0a\u9650\u3002<\/p>\n<p>NoThinking \u65b9\u6cd5\u4e0e\u5e76\u884c scaling \u76f8\u7ed3\u5408\uff0c\u4e3a\u4f20\u7edf\u7684\u987a\u5e8f\u65b9\u6cd5\u63d0\u4f9b\u4e86\u4e00\u79cd\u9ad8\u6548\u7684\u66ff\u4ee3\u65b9\u6848\uff0c\u80fd\u591f\u5728\u663e\u8457\u964d\u4f4e\u5ef6\u8fdf\u548c token \u4f7f\u7528\u91cf\u7684\u60c5\u51b5\u4e0b\uff0c\u8fbe\u5230\u76f8\u4f3c\u751a\u81f3\u66f4\u597d\u7684\u51c6\u786e\u7387\u3002\u5982\u56fe 7 \u7684\u524d\u4e24\u4e2a\u56fe\u6240\u793a\uff0cNoThinking \u65b9\u6cd5\u5b9e\u73b0\u4e86\u4e0e Thinking \u65b9\u6cd5\u76f8\u5f53\u751a\u81f3\u66f4\u9ad8\u7684\u6027\u80fd\uff0c\u540c\u65f6\u5ef6\u8fdf\u8981\u4f4e\u5f97\u591a\u3002\u5728\u6ca1\u6709\u5e76\u884c scaling \u7684\u60c5\u51b5\u4e0b\uff0cNoThinking \u65b9\u6cd5\u5728\u51c6\u786e\u7387\u4e0a\u4e0e Thinking \u65b9\u6cd5\u76f8\u8fd1\uff0c\u800c\u5ef6\u8fdf\u4ec5\u4e3a\u540e\u8005\u7684\u4e00\u5c0f\u90e8\u5206\u3002<\/p>\n<p><img decoding=\"async\" title=\"image.png\" src=\"https:\/\/s3.51cto.com\/oss\/202504\/20\/a31214701e31f3e739d076a24d0d78ab63b4f2.webp\" alt=\"image.png\" data-type=\"inline\" \/><\/p>\n<p>\u5982\u679c\u6709\u4e00\u4e2a\u5b8c\u7f8e\u7684\u9a8c\u8bc1\u5668\u53ef\u7528\uff0c\u90a3\u4e48\u4ece k \u4e2a\u91c7\u6837\u54cd\u5e94\u4e2d\u9009\u62e9\u6700\u4f73\u7684\u4e00\u4e2a\u5c31\u80fd\u5b9e\u73b0 pass@k \u51c6\u786e\u5ea6\u3002\u5f53\u4e0e\u5e76\u884c scaling \u7ed3\u5408\u4f7f\u7528\u65f6\uff0cNoThinking \u65b9\u6cd5\u5728\u51c6\u786e\u7387\u4e0a\u4e0e\u4e0d\u91c7\u7528\u5f3a\u5236\u9884\u7b97\u4e14\u4e0d\u8fdb\u884c\u5e76\u884c scaling \u7684 Thinking \u65b9\u6cd5\uff08\u8fd9\u662f\u5177\u6709\u4ee3\u8868\u6027\u7684\u987a\u5e8f scaling \u57fa\u7ebf\uff09\u76f8\u5f53\uff0c\u540c\u65f6\u5c06\u5ef6\u8fdf\u964d\u4f4e\u5230\u539f\u6765\u7684\u4e03\u5206\u4e4b\u4e00\u3002\u6b64\u5916\uff0c\u5728 MiniF2F \u548c ProofNet \u8fd9\u4e24\u4e2a\u6570\u636e\u96c6\u4e0a\uff0cNoThinking \u65b9\u6cd5\u4f7f\u7528\u7684\u8f93\u51fa token \u6570\u91cf\u51cf\u5c11\u4e86\u56db\u5206\u4e4b\u4e09\uff0c\u5374\u5b9e\u73b0\u4e86\u76f8\u540c\u7684\u51c6\u786e\u7387\uff0c\u8fd9\u51f8\u663e\u4e86\u5b83\u7684\u8ba1\u7b97\u6548\u7387\u3002\u8fd9\u4e9b\u7ed3\u679c\u5f3a\u8c03\u4e86\u5728\u6709\u9a8c\u8bc1\u5668\u53ef\u7528\u7684\u60c5\u51b5\u4e0b\uff0c\u5e76\u884c\u91c7\u6837\u7684\u6709\u6548\u6027\u3002<\/p>\n<p>\u5f53 NoThinking \u65b9\u6cd5\u4e0e\u5e76\u884c scaling \u4ee5\u53ca\u57fa\u4e8e\u7f6e\u4fe1\u5ea6\u7684\u9009\u62e9\u65b9\u6cd5\u76f8\u7ed3\u5408\u65f6\uff0c\u5728\u5927\u591a\u6570\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u5b83\u5728\u4f4e token \u9884\u7b97\u7684\u60c5\u51b5\u4e0b\u59cb\u7ec8\u4f18\u4e8e Thinking \u65b9\u6cd5\u3002\u56fe 7\uff08\u6700\u540e\u4e94\u4e2a\u56fe\uff09\u5c55\u793a\u4e86\u57fa\u4e8e\u7f6e\u4fe1\u5ea6\u9009\u62e9\u65b9\u6cd5\u5728\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u7684\u7ed3\u679c\uff0c\u6bd4\u8f83\u4e86\u5728\u53d7\u63a7 token \u4f7f\u7528\u91cf\u60c5\u51b5\u4e0b Thinking \u65b9\u6cd5\u548c NoThinking \u65b9\u6cd5\u7684\u8868\u73b0\u3002<\/p>\n<p>\u5173\u6ce8\u4f4e\u9884\u7b97\u60c5\u51b5\u6709\u4e24\u4e2a\u539f\u56e0\uff1a\uff081\uff09\u8fd9\u7b26\u5408\u6211\u4eec\u5bf9\u9ad8\u6548\u63a8\u7406\u7684\u4e3b\u8981\u7814\u7a76\u5174\u8da3\uff1b\uff082\uff09\u5982\u679c\u5c06\u6700\u5927 token \u6570\u8bbe\u7f6e\u5f97\u8fc7\u9ad8\uff0c\u901a\u5e38\u4f1a\u5bfc\u81f4\u8f93\u51fa\u5185\u5bb9\u8fc7\u957f\u4e14\u4e0d\u8fde\u8d2f\uff08\u300c\u80e1\u8a00\u4e71\u8bed\u300d\uff09\uff0c\u8fd9\u4f1a\u589e\u52a0\u5ef6\u8fdf\u5e76\u964d\u4f4e\u6bd4\u8f83\u7684\u4ef7\u503c\u3002<\/p>\n<p>\u6b63\u5982\u9884\u671f\u7684\u90a3\u6837\uff0c\u5e76\u884c scaling \u63d0\u9ad8\u4e86 Thinking \u65b9\u6cd5\u548c NoThinking \u65b9\u6cd5\u7684\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u6027\u80fd\u3002\u7136\u800c\uff0c\u5728\u6240\u6709\u6570\u5b66\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cNoThinking \u65b9\u6cd5\u59cb\u7ec8\u5904\u4e8e\u5e15\u7d2f\u6258\u6700\u4f18\u8fb9\u754c\u7684\u4e3b\u5bfc\u5730\u4f4d\u3002<\/p>\n<p>\u5728\u91c7\u7528\u5e76\u884c scaling \u7684 Thinking \u65b9\u6cd5\u65b9\u9762\uff0cNoThinking \u65b9\u6cd5\u5c55\u73b0\u51fa\u4e86\u66f4\u4f18\u7684\u51c6\u786e\u7387\u4e0e\u9884\u7b97\u4e4b\u95f4\u7684\u6743\u8861\u3002\u5728 AMC 2023 \u548c OlympiadBench \u57fa\u51c6\u4e0a\uff0c\u65e0\u8bba\u662f\u5426\u4f7f\u7528\u5e76\u884c scaling\uff0cNoThinking \u65b9\u6cd5\u7684\u8868\u73b0\u59cb\u7ec8\u4f18\u4e8e Thinking \u65b9\u6cd5\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u5373\u4f7f\u4e0e\u5b8c\u6574\u7684 Thinking \u65b9\u6cd5\uff08\u4e0d\u91c7\u7528\u5f3a\u5236\u9884\u7b97\u7684 Thinking \u65b9\u6cd5\uff09\u76f8\u6bd4\uff0cNoThinking \u65b9\u6cd5\u5728\u5c06\u5ef6\u8fdf\u964d\u4f4e\u5230\u539f\u6765\u7684\u4e5d\u5206\u4e4b\u4e00\u7684\u540c\u65f6\uff0c\u8fd8\u5b9e\u73b0\u4e86\u66f4\u9ad8\u7684\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u5f97\u5206\uff0855.79 \u6bd4 54.1\uff09\u3002<\/p>\n<p>NoThinking \u65b9\u6cd5\u5728 LiveCodeBench \u4e0a\u7684\u6548\u679c\u8f83\u5dee\uff0c\u8be5\u57fa\u51c6\u6d4b\u8bd5\u4f3c\u4e4e\u662f\u4e2a\u4f8b\u5916\u60c5\u51b5\u3002\u8fd9\u53ef\u80fd\u662f\u56e0\u4e3a\u57fa\u4e8e\u7f6e\u4fe1\u5ea6\u7684\u9009\u62e9\u65b9\u6cd5\u5728\u7f16\u7801\u4efb\u52a1\u4e2d\u5b58\u5728\u5c40\u9650\u6027\uff0c\u5728\u6ca1\u6709\u5b8c\u5168\u5339\u914d\u8f93\u51fa\u7684\u60c5\u51b5\u4e0b\uff0c\u6295\u7968\u7b56\u7565\u65e0\u6cd5\u5e94\u7528\u3002\u5728\u8fd9\u4e9b\u60c5\u51b5\u4e0b\uff0c\u53ea\u80fd\u9000\u800c\u6c42\u5176\u6b21\uff0c\u9009\u62e9\u7f6e\u4fe1\u5ea6\u6700\u9ad8\u7684\u7b54\u6848\uff0c\u800c\u8fd9\u79cd\u65b9\u5f0f\u7684\u53ef\u9760\u6027\u8f83\u4f4e\u3002\u5982\u8868 2 \u6240\u793a\uff0c\u4e0e\u5728\u53ef\u5e94\u7528\u6295\u7968\u7b56\u7565\u7684\u4efb\u52a1\u4e2d\u57fa\u4e8e\u6295\u7968\u7684\u65b9\u6cd5\u76f8\u6bd4\uff0c\u8fd9\u79cd\u65b9\u6cd5\u7684\u8868\u73b0\u4e00\u76f4\u8f83\u5dee\uff08\u901a\u5e38\u5dee\u8ddd\u5f88\u5927\uff09\u3002\u603b\u4f53\u800c\u8a00\uff0c\u8fd9\u4e9b\u7ed3\u679c\u51f8\u663e\u4e86\u5728\u65e0\u9a8c\u8bc1\u5668\u7684\u60c5\u51b5\u4e0b\uff0c\u5f53 NoThinking \u65b9\u6cd5\u4e0e\u5e76\u884c\u91c7\u6837\u4ee5\u53ca\u5f3a\u5927\u7684\u9009\u62e9\u7b56\u7565\u76f8\u7ed3\u5408\u65f6\u7684\u6709\u6548\u6027\u3002<\/p>\n<p>\u968f\u7740 k \u503c\u7684\u589e\u52a0\uff0cNoThinking \u65b9\u6cd5\u5728\u591a\u6837\u672c\u51c6\u786e\u7387\uff08pass@k\uff09\u65b9\u9762\u4ee4\u4eba\u60ca\u559c\u7684\u8868\u73b0\u53ef\u4ee5\u901a\u8fc7\u5e76\u884c scaling \u5f97\u5230\u8fdb\u4e00\u6b65\u5229\u7528\uff0c\u4ece\u800c\u5728\u76f8\u4f3c\u751a\u81f3\u663e\u8457\u66f4\u4f4e\u7684\u5ef6\u8fdf\uff08\u6700\u591a\u53ef\u964d\u4f4e\u81f3\u539f\u6765\u7684\u4e5d\u5206\u4e4b\u4e00\uff09\u60c5\u51b5\u4e0b\uff0c\u63d0\u5347\u5355\u6837\u672c\u51c6\u786e\u7387\uff08pass@1\uff09\u7684\u7ed3\u679c\u3002\u5bf9\u4e8e\u914d\u5907\u4e86\u5b8c\u7f8e\u9a8c\u8bc1\u5668\u7684\u4efb\u52a1\uff0c\u8fd9\u79cd\u65b9\u6cd5\u8fd8\u80fd\u5728\u8fbe\u5230\u76f8\u4f3c\u6216\u66f4\u9ad8\u51c6\u786e\u7387\u7684\u540c\u65f6\uff0c\u5c06 token \u7684\u603b\u4f7f\u7528\u91cf\u51cf\u5c11\u591a\u8fbe\u56db\u5206\u4e4b\u4e09\u3002<\/p>\n<h4>\u603b\u7ed3<\/h4>\n<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u751f\u6210\u89e3\u7b54\u4e4b\u524d\u4f1a\u4ea7\u751f\u5197\u957f\u7684\u601d\u8003\u8fc7\u7a0b\uff0c\u8fd9\u79cd\u65b9\u5f0f\u5728\u63a8\u7406\u4efb\u52a1\u4e0a\u5df2\u7ecf\u53d6\u5f97\u4e86\u5f88\u597d\u7684\u6210\u679c\u3002\u8be5\u7814\u7a76\u5bf9\u8fd9\u4e00\u8fc7\u7a0b\u7684\u5fc5\u8981\u6027\u63d0\u51fa\u4e86\u8d28\u7591\uff0c\u4e3a\u6b64\u5f15\u5165\u4e86 NoThinking \u65b9\u6cd5\u3002<\/p>\n<p>\u8fd9\u662f\u4e00\u79cd\u7b80\u5355\u800c\u6709\u6548\u7684\u63d0\u793a\u7b56\u7565\uff0c\u80fd\u591f\u7ed5\u8fc7\u663e\u5f0f\u7684\u601d\u8003\u8fc7\u7a0b\u3002\u5b9e\u9a8c\u8bc1\u660e\uff0c\u540c\u6837\u7684\u6a21\u578b\u5728\u6ca1\u6709\u5197\u957f\u601d\u7ef4\u94fe\u7684\u60c5\u51b5\u4e0b\uff0c\u968f\u7740 pass@k \u4e2d k \u503c\u7684\u589e\u52a0\uff0c\u5176\u8868\u73b0\u53ef\u4ee5\u4e0e Thinking \u65b9\u6cd5\u76f8\u5f53\uff0c\u751a\u81f3\u4f18\u4e8e Thinking \u65b9\u6cd5\uff0c\u540c\u65f6\u6240\u4f7f\u7528\u7684 token \u8981\u5c11\u5f97\u591a\u3002<\/p>\n<p>\u5728 token \u9884\u7b97\u76f8\u5f53\u7684\u60c5\u51b5\u4e0b\uff0c\u5bf9\u4e8e\u5927\u591a\u6570 k \u503c\uff0cNoThinking \u65b9\u6cd5\u7684\u8868\u73b0\u59cb\u7ec8\u4f18\u4e8e\u4f20\u7edf\u7684 Thinking \u7ed3\u679c\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u7814\u7a76\u8fd8\u53d1\u73b0\uff0cNoThinking \u65b9\u6cd5\u53ef\u4ee5\u4e0e \u300c\u4ece N \u4e2a\u4e2d\u9009\u6700\u4f73\u300d\u7684\u9009\u62e9\u65b9\u6cd5\u76f8\u7ed3\u5408\uff0c\u4ece\u800c\u5728\u51c6\u786e\u7387\u548c\u5ef6\u8fdf\u7684\u6743\u8861\u65b9\u9762\uff0c\u53d6\u5f97\u6bd4\u6807\u51c6 Thinking \u65b9\u6cd5\u66f4\u597d\u7684\u6548\u679c\u3002<\/p>\n<p>\u7814\u7a76\u8005\u8868\u793a\uff1a\u300c\u6211\u4eec\u5e0c\u671b\u8fd9\u4e2a\u7814\u7a76\u80fd\u591f\u4fc3\u4f7f\u4eba\u4eec\u91cd\u65b0\u5ba1\u89c6\u5197\u957f\u601d\u8003\u8fc7\u7a0b\u7684\u5fc5\u8981\u6027\uff0c\u540c\u65f6\u4e3a\u5728\u4f4e\u9884\u7b97\u548c\u4f4e\u5ef6\u8fdf\u7684\u60c5\u51b5\u4e0b\u5b9e\u73b0\u5f3a\u5927\u7684\u63a8\u7406\u6027\u80fd\uff0c\u63d0\u4f9b\u4e00\u4e2a\u6781\u5177\u7ade\u4e89\u529b\u7684\u53c2\u8003\u3002\u300d<\/p>\n<p>\u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_25509\" class=\"pvc_stats total_only  \" data-element-id=\"25509\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u751f\u6210\u89e3\u7b54\u4e4b\u524d\u4f1a\u4ea7\u751f\u5197\u957f\u7684\u601d\u8003\u8fc7\u7a0b\uff0c\u8fd9\u79cd\u65b9\u5f0f\u5728\u63a8\u7406\u4efb\u52a1\u4e0a\u5df2\u7ecf\u53d6\u5f97\u4e86\u5f88\u597d\u7684\u6210\u679c\u3002\u8be5\u7814\u7a76\u5bf9\u8fd9\u4e00\u8fc7\u7a0b\u7684\u5fc5 [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_25509\" class=\"pvc_stats total_only  \" data-element-id=\"25509\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":56,"featured_media":25511,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,23,20,80],"tags":[124,968,1486],"class_list":["post-25509","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-23","category-20","category-80","tag-ai","tag-968","tag-1486"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/25509","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=25509"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/25509\/revisions"}],"predecessor-version":[{"id":25512,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/25509\/revisions\/25512"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/25511"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=25509"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=25509"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=25509"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}