{"id":19632,"date":"2025-01-29T10:58:52","date_gmt":"2025-01-29T02:58:52","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=19632"},"modified":"2025-01-29T10:58:52","modified_gmt":"2025-01-29T02:58:52","slug":"deepseek-r1-overthinker","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/ja\/deepseek-r1-overthinker\/","title":{"rendered":"R1 Overthinker\uff1a\u5f3a\u5236 DeepSeek R1 \u6a21\u578b\u5ef6\u957f\u601d\u8003\u65f6\u95f4"},"content":{"rendered":"<p><a href=\"https:\/\/www.kdjingpai.com\/deepseek-chatshena\/\">DeepSeek<\/a> R1 Overthinker \u662f\u4e00\u4e2a\u4e13\u4e3a\u589e\u5f3a <a href=\"https:\/\/www.kdjingpai.com\/deepseek-r1nenglixiang\/\">DeepSeek R1<\/a> \u6a21\u578b\u601d\u8003\u6df1\u5ea6\u800c\u8bbe\u8ba1\u7684\u5de5\u5177\u3002\u901a\u8fc7\u5ef6\u957f\u6a21\u578b\u7684\u63a8\u7406\u8fc7\u7a0b\uff0c\u8be5\u5de5\u5177\u80fd\u591f\u4f7f\u6a21\u578b\u8fdb\u884c\u66f4\u6df1\u5165\u7684\u601d\u8003\uff0c\u4ece\u800c\u63d0\u9ad8\u5176\u56de\u7b54\u7684\u8d28\u91cf\u548c\u51c6\u786e\u6027\u3002\u8be5\u5de5\u5177\u5229\u7528 <a href=\"https:\/\/www.kdjingpai.com\/unsloth\/\">unsloth<\/a> \u4f18\u5316\u6a21\u578b\uff0c\u652f\u6301\u65e0\u9650\u4e0a\u4e0b\u6587\u957f\u5ea6\uff08\u53d6\u51b3\u4e8e\u53ef\u7528\u7684 VRAM\uff09\u3002\u7528\u6237\u53ef\u4ee5\u81ea\u5b9a\u4e49\u63a8\u7406\u6269\u5c55\u548c\u601d\u8003\u9608\u503c\uff0c\u7cbe\u7ec6\u63a7\u5236\u6a21\u578b\u53c2\u6570\uff08\u5982\u6e29\u5ea6\u3001top-p \u7b49\uff09\uff0c\u5e76\u5b9e\u65f6\u8ddf\u8e2a\u601d\u8003\u8fc7\u7a0b\u4e2d\u7684 <a href=\"https:\/\/www.kdjingpai.com\/tokenization\/\">token<\/a> \u8ba1\u6570\u3002DeepSeek R1 Overthinker \u9002\u7528\u4e8e\u591a\u79cd VRAM \u914d\u7f6e\uff0c\u5e76\u652f\u6301\u591a\u79cd\u6a21\u578b\u5927\u5c0f\uff08\u4ece 1.5B \u5230 70B \u53c2\u6570\uff09\u3002<\/p>\n<blockquote><p>\u9879\u76ee\u539f\u7406\u4e0d\u65ad\u68c0\u6d4b\u8f93\u51fa\uff0c\u628a&lt;\/thinking&gt;\u66ff\u6362\u4e3a\u4e00\u53e5 rethinking \u7684\u8bdd\uff0c\u8be5\u9879\u76ee\u7684\u7075\u611f\u6765\u6e90\u4e8e\uff1a<a href=\"https:\/\/gist.github.com\/vgel\/8a2497dc45b1ded33287fa7bb6cc1adc\">r1_overthinker<\/a>\u00a0\u3002<\/p><\/blockquote>\n<p><img decoding=\"async\" title=\"DeepSeek R1 Overthinker\uff1a\u5f3a\u5236 DeepSeek R1 \u84b8\u998f\u6a21\u578b\u5ef6\u957f\u601d\u8003\u65f6\u95f4-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/d6f6544b63473bc.jpg\" alt=\"DeepSeek R1 Overthinker\uff1a\u5f3a\u5236 DeepSeek R1 \u84b8\u998f\u6a21\u578b\u5ef6\u957f\u601d\u8003\u65f6\u95f4-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u5f3a\u5236\u6a21\u578b\u8fdb\u884c\u66f4\u957f\u65f6\u95f4\u548c\u66f4\u6df1\u5165\u7684\u601d\u8003<\/li>\n<li>\u81ea\u5b9a\u4e49\u63a8\u7406\u6269\u5c55\u548c\u601d\u8003\u9608\u503c<\/li>\n<li>\u7cbe\u7ec6\u63a7\u5236\u6a21\u578b\u53c2\u6570\uff08\u6e29\u5ea6\u3001top-p \u7b49\uff09<\/li>\n<li>\u5b9e\u65f6\u53ef\u89c1\u7684\u601d\u8003\u8fc7\u7a0b\u548c token \u8ba1\u6570\u8ddf\u8e2a<\/li>\n<li>\u652f\u6301 LaTeX \u6570\u5b66\u8868\u8fbe\u5f0f<\/li>\n<li>\u4f18\u5316\u9002\u7528\u4e8e\u5404\u79cd VRAM \u914d\u7f6e<\/li>\n<li>\u652f\u6301\u591a\u79cd\u6a21\u578b\u5927\u5c0f\uff081.5B \u5230 70B \u53c2\u6570\uff09<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<ol>\n<li>\u8fd0\u884c <a href=\"https:\/\/colab.research.google.com\/github\/qunash\/r1-overthinker\/blob\/master\/r1_overthinker.ipynb\">Google Colab<\/a>\u3002<\/li>\n<\/ol>\n<h3>\u4f7f\u7528\u6307\u5357<\/h3>\n<ol>\n<li><strong>\u542f\u52a8\u5e94\u7528<\/strong>\uff1a\n<ul>\n<li>\u8fd0\u884c\u00a0<code>python r1_overthinker.py<\/code>\u00a0\u542f\u52a8\u5e94\u7528\u7a0b\u5e8f\u3002<\/li>\n<li>\u5e94\u7528\u7a0b\u5e8f\u542f\u52a8\u540e\uff0c\u4f1a\u52a0\u8f7d DeepSeek R1 \u6a21\u578b\uff0c\u5e76\u51c6\u5907\u5904\u7406\u8f93\u5165\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u8f93\u5165\u5904\u7406<\/strong>\uff1a\n<ul>\n<li>\u8f93\u5165\u53e5\u5b50\u6216\u6bb5\u843d\uff0c\u5e94\u7528\u7a0b\u5e8f\u5c06\u901a\u8fc7\u6269\u5c55\u63a8\u7406\u8fc7\u7a0b\u8fdb\u884c\u5904\u7406\u3002<\/li>\n<li>\u5e94\u7528\u7a0b\u5e8f\u4f1a\u68c0\u6d4b\u6a21\u578b\u4f55\u65f6\u8bd5\u56fe\u7ed3\u675f\u601d\u8003\uff0c\u5e76\u7528\u9f13\u52b1\u989d\u5916\u63a8\u7406\u7684\u63d0\u793a\u8fdb\u884c\u66ff\u6362\uff0c\u76f4\u5230\u8fbe\u5230\u7528\u6237\u6307\u5b9a\u7684\u601d\u8003\u65f6\u95f4\u9608\u503c\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u81ea\u5b9a\u4e49\u601d\u8003\u65f6\u95f4<\/strong>\uff1a\n<ul>\n<li>\u7528\u6237\u53ef\u4ee5\u5728\u542f\u52a8\u5e94\u7528\u7a0b\u5e8f\u65f6\u6307\u5b9a\u601d\u8003\u65f6\u95f4\u9608\u503c\uff0c\u4f8b\u5982\uff1a\n<pre><code>python r1_overthinker.py --min-think-time 10\r\n<\/code><\/pre>\n<\/li>\n<li>\u4ee5\u4e0a\u547d\u4ee4\u5c06\u8bbe\u7f6e\u6a21\u578b\u9700\u8981\u601d\u8003\u7684\u6700\u5c0f\u65f6\u95f4\u4e3a 10 \u79d2\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u67e5\u770b\u7ed3\u679c<\/strong>\uff1a\n<ul>\n<li>\u5904\u7406\u5b8c\u6210\u540e\uff0c\u5e94\u7528\u7a0b\u5e8f\u5c06\u8f93\u51fa\u6269\u5c55\u63a8\u7406\u540e\u7684\u7ed3\u679c\uff0c\u7528\u6237\u53ef\u4ee5\u67e5\u770b\u6a21\u578b\u66f4\u6df1\u5165\u7684\u63a8\u7406\u8fc7\u7a0b\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h3>\u8be6\u7ec6\u64cd\u4f5c\u6d41\u7a0b<\/h3>\n<ol>\n<li><strong>\u521d\u59cb\u5316\u8bbe\u7f6e<\/strong>\uff1a\n<ul>\n<li>\u9996\u6b21\u4f7f\u7528\u65f6\uff0c\u786e\u4fdd\u5df2\u6b63\u786e\u5b89\u88c5\u6240\u6709\u4f9d\u8d56\u9879\uff0c\u5e76\u8fdb\u884c\u521d\u59cb\u5316\u8bbe\u7f6e\u3002<\/li>\n<li>\u68c0\u67e5 VRAM \u53ef\u7528\u60c5\u51b5\uff0c\u4ee5\u786e\u4fdd\u5904\u7406\u8f83\u5927\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u8f93\u5165\u6587\u672c\u5904\u7406<\/strong>\uff1a\n<ul>\n<li>\u8f93\u5165\u9700\u8981\u5904\u7406\u7684\u6587\u672c\uff0c\u5e94\u7528\u7a0b\u5e8f\u5c06\u81ea\u52a8\u68c0\u6d4b\u5e76\u6269\u5c55\u6a21\u578b\u7684\u63a8\u7406\u8fc7\u7a0b\u3002<\/li>\n<li>\u5904\u7406\u8fc7\u7a0b\u4e2d\uff0c\u7528\u6237\u53ef\u4ee5\u5b9e\u65f6\u67e5\u770b\u6a21\u578b\u7684\u63a8\u7406\u8fdb\u5c55\u548c\u5f53\u524d\u601d\u8003\u72b6\u6001\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u81ea\u5b9a\u4e49\u8bbe\u7f6e<\/strong>\uff1a\n<ul>\n<li>\u7528\u6237\u53ef\u4ee5\u6839\u636e\u9700\u8981\u8c03\u6574\u601d\u8003\u65f6\u95f4\u9608\u503c\u548c\u5176\u4ed6\u53c2\u6570\uff0c\u4ee5\u83b7\u5f97\u6700\u4f73\u7ed3\u679c\u3002<\/li>\n<li>\u5e94\u7528\u7a0b\u5e8f\u652f\u6301\u591a\u79cd\u81ea\u5b9a\u4e49\u8bbe\u7f6e\uff0c\u7528\u6237\u53ef\u4ee5\u7075\u6d3b\u8c03\u6574\u4ee5\u9002\u5e94\u4e0d\u540c\u7684\u4f7f\u7528\u573a\u666f\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u7ed3\u679c\u8f93\u51fa<\/strong>\uff1a\n<ul>\n<li>\u5904\u7406\u5b8c\u6210\u540e\uff0c\u5e94\u7528\u7a0b\u5e8f\u5c06\u8f93\u51fa\u5b8c\u6574\u7684\u63a8\u7406\u7ed3\u679c\uff0c\u7528\u6237\u53ef\u4ee5\u67e5\u770b\u548c\u5206\u6790\u6a21\u578b\u7684\u601d\u8003\u8fc7\u7a0b\u3002<\/li>\n<li>\u7ed3\u679c\u8f93\u51fa\u652f\u6301\u591a\u79cd\u683c\u5f0f\uff0c\u7528\u6237\u53ef\u4ee5\u6839\u636e\u9700\u8981\u8fdb\u884c\u4fdd\u5b58\u548c\u8fdb\u4e00\u6b65\u5904\u7406\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>DeepSeek R1 Overthinker \u662f\u4e00\u4e2a\u4e13\u4e3a\u589e\u5f3a DeepSeek R1 \u6a21\u578b\u601d\u8003\u6df1\u5ea6\u800c\u8bbe\u8ba1\u7684\u5de5\u5177\u3002\u901a\u8fc7\u5ef6\u957f\u6a21\u578b\u7684\u63a8\u7406\u8fc7\u7a0b\uff0c\u8be5\u5de5\u5177\u80fd\u591f\u4f7f\u6a21\u578b\u8fdb\u884c\u66f4\u6df1\u5165\u7684\u601d\u8003\uff0c\u4ece\u800c\u63d0\u9ad8\u5176\u56de\u7b54\u7684\u8d28\u91cf\u548c\u51c6\u786e\u6027\u3002\u8be5\u5de5\u5177\u5229\u7528 unsloth \u4f18\u5316\u6a21\u578b\uff0c&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61742,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230],"class_list":["post-19632","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/19632","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/comments?post=19632"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/19632\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media\/61742"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media?parent=19632"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/categories?post=19632"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/tags?post=19632"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}