{"id":38062,"date":"2025-02-09T09:01:40","date_gmt":"2025-02-09T01:01:40","guid":{"rendered":"https:\/\/17aitech.com\/?p=38062"},"modified":"2025-02-09T09:01:40","modified_gmt":"2025-02-09T01:01:40","slug":"%e5%ae%8c%e6%95%b4%e7%9a%84671b-moe-deepseek-r1%e6%80%8e%e4%b9%88%e5%a1%9e%e8%bf%9b%e6%9c%ac%e5%9c%b0%e5%8c%96%e9%83%a8%e7%bd%b2%ef%bc%9f%e8%af%a6%e5%b0%bd%e6%95%99%e7%a8%8b%e5%a4%a7%e6%94%be%e9%80%81","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=38062","title":{"rendered":"\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-02-4\" target=\"_blank\">\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01<\/a><\/p>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png\"><\/a><\/section>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"167\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<section>AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/section>\n<\/section>\n<\/blockquote>\n<section><strong><br \/><\/strong><\/section>\n<section><strong>\u672c\u6587\u7684\u4f5c\u8005\u662f<strong>\u674e\u9521\u6db5\uff08<\/strong>Xihan Li\uff09<\/strong><strong>\u3002\u4ed6\u662f\u4f26\u6566\u5927\u5b66\u5b66\u9662\uff08UCL\uff09\u8ba1\u7b97\u673a\u7cfb\u535a\u58eb\u7814\u7a76\u751f\uff0c\u8c37\u6b4c\u5f00\u53d1\u8005\u4e13\u5bb6\uff0c\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u4e3a\u5b66\u4e60\u4f18\u5316\uff0c\u5728 NeurIPS\u3001ICLR\u3001AAMAS\u3001CIKM \u7b49\u4f1a\u8bae\u53d1\u8868\u8fc7\u5b66\u672f\u8bba\u6587\uff0cCircuit Transformer \u4f5c\u8005\uff0c\u56fe\u4e66\u300a\u7b80\u660e\u7684 TensorFlow 2\u300b\uff08https:\/\/tf.wiki\uff09\u4f5c\u8005\u3002<\/strong><\/section>\n<section><\/section>\n<section>\u8fc7\u5e74\u8fd9\u51e0\u5929\uff0cDeepSeek \u7b97\u662f\u5f7b\u5e95\u7834\u5708\u4e86\uff0c\u706b\u904d\u5927\u6c5f\u5357\u5317\uff0c\u706b\u5230\u4eba\u5c3d\u7686\u77e5\u3002\u867d\u7136\u7f51\u7edc\u7248\u548c APP \u7248\u5df2\u7ecf\u8db3\u591f\u597d\u7528\uff0c\u4f46\u628a\u6a21\u578b\u90e8\u7f72\u5230\u672c\u5730\uff0c\u624d\u80fd\u771f\u6b63\u5b9e\u73b0\u72ec\u5bb6\u5b9a\u5236\uff0c\u8ba9 DeepSeek R1 \u7684\u6df1\u5ea6\u601d\u8003\u300c\u4ee5\u4f60\u4e3a\u4e3b\uff0c\u4e3a\u4f60\u6240\u7528\u300d\u3002<\/section>\n<section><\/section>\n<section>\u5173\u4e8e\u672c\u5730\u90e8\u7f72\uff0c\u5927\u591a\u6570\u4eba\u4f7f\u7528\u7684\u662f\u84b8\u998f\u540e\u76848B\/32B\/70B\u7248\u672c\uff0c\u672c\u8d28\u662f\u5fae\u8c03\u540e\u7684Llama\u6216Qwen\u6a21\u578b\uff0c\u5e76\u4e0d\u80fd\u5b8c\u5168\u53d1\u6325\u51faDeepSeek R1\u7684\u5b9e\u529b\u3002<\/section>\n<section><\/section>\n<section>\u7136\u800c\uff0c\u5b8c\u6574\u7684671B MoE\u6a21\u578b\u4e5f\u53ef\u4ee5\u901a\u8fc7\u9488\u5bf9\u6027\u7684\u91cf\u5316\u6280\u672f\u538b\u7f29\u4f53\u79ef\uff0c\u4ece\u800c\u5927\u5e45\u964d\u4f4e\u672c\u5730\u90e8\u7f72\u95e8\u69db\uff0c\u4e43\u81f3\u5728\u6d88\u8d39\u7ea7\u786c\u4ef6\uff08\u5982\u5355\u53f0Mac Studio\uff09\u4e0a\u8fd0\u884c\u3002<\/section>\n<section><\/section>\n<section>\u90a3\u4e48\uff0c\u5982\u4f55\u7528 ollama \u5728\u672c\u5730\u90e8\u7f72 DeepSeek R1 671B\uff08\u5b8c\u6574\u672a\u84b8\u998f\u7248\u672c\uff09\u6a21\u578b\u5462\uff1f\u4e00\u7bc7\u5728\u6d77\u5916\u70ed\u5ea6\u5f88\u9ad8\u7684\u7b80\u660e\u6559\u7a0b\u5373\u5c06\u63ed\u6653\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-af011371ef1b284e8c0c00998e579344.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-af011371ef1b284e8c0c00998e579344.png\"><\/a><\/section>\n<ul>\n<li>\n<p>\u4f5c\u8005\u4e3b\u9875\uff1ahttps:\/\/snowkylin.github.io<\/p>\n<\/li>\n<li>\n<p>\u539f\u6587\u5730\u5740\uff1ahttps:\/\/snowkylin.github.io\/blogs\/a-note-on-deepseek-r1.html<\/p>\n<\/li>\n<\/ul>\n<section><span data-cover=\"http%3A%2F%2Fmmbiz.qpic.cn%2Fsz_mmbiz_jpg%2FKmXPKA19gWib4IbH5m6iaPibsIYnraicia3TBM9QyhSicCibXciaib7uO7O1oysnBINgRPcFJrGpTktSDWVgyje1UyswngA%2F0%3Fwx_fmt%3Djpeg\" data-mpvid=\"wxv_3838683620433920000\" data-ratio=\"0.976491862567812\" data-src=\"https:\/\/mp.weixin.qq.com\/mp\/readtemplate?t=pages\/video_player_tmpl&amp;auto=0&amp;vid=wxv_3838683620433920000\" data-vh=\"495.75\" data-vidtype=\"2\" data-vw=\"661\" data-w=\"1080\"><\/p>\n<div data-key=\"wxv_3838683620433920000\">\n<div data-v-010f46be=\"\"><\/div>\n<\/div>\n<p><\/span><\/section>\n<section><em><a href=\"https:\/\/mp.weixin.qq.com\/s\/GnHzsgvW90DGChENqTBsRw?token=935449120&amp;lang=zh_CN\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-6d1be6c48727868872ab86f27a979e5f.jpeg\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-6d1be6c48727868872ab86f27a979e5f.jpeg\"><\/a><\/a><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u672c\u5730\u90e8\u7f72\u540e\uff0c\u8ba9 DeepSeek R1 \u300c\u6570\u8349\u8393\u300d<\/sup><\/em><\/section>\n<section><\/section>\n<section><strong>\u6a21\u578b\u9009\u62e9<\/strong><\/section>\n<section><\/section>\n<section>\u539f\u7248 DeepSeek R1 671B \u5168\u91cf\u6a21\u578b\u7684\u6587\u4ef6\u4f53\u79ef\u9ad8\u8fbe 720GB\uff0c\u5bf9\u4e8e\u7edd\u5927\u90e8\u5206\u4eba\u800c\u8a00\uff0c\u8fd9\u90fd\u5927\u5f97\u592a\u79bb\u8c31\u4e86\u3002\u672c\u6587\u91c7\u7528 Unsloth AI \u5728 HuggingFace \u4e0a\u63d0\u4f9b\u7684 \u201c\u52a8\u6001\u91cf\u5316\u201d \u7248\u672c\u6765\u5927\u5e45\u7f29\u51cf\u6a21\u578b\u7684\u4f53\u79ef\uff0c\u4ece\u800c\u8ba9\u66f4\u591a\u4eba\u80fd\u5728\u81ea\u5df1\u7684\u672c\u5730\u73af\u5883\u90e8\u7f72\u8be5\u5168\u91cf\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section>\u201c\u52a8\u6001\u91cf\u5316\u201d \u7684\u6838\u5fc3\u601d\u8def\u662f\uff1a\u5bf9\u6a21\u578b\u7684\u5c11\u6570\u5173\u952e\u5c42\u8fdb\u884c\u9ad8\u8d28\u91cf\u7684 4-6bit \u91cf\u5316\uff0c\u800c\u5bf9\u5927\u90e8\u5206\u76f8\u5bf9\u6ca1\u90a3\u4e48\u5173\u952e\u7684\u6df7\u5408\u4e13\u5bb6\u5c42\uff08MoE\uff09\u8fdb\u884c\u5927\u5200\u9614\u65a7\u7684 1-2bit \u91cf\u5316\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u6cd5\uff0cDeepSeek R1 \u5168\u91cf\u6a21\u578b\u53ef\u538b\u7f29\u81f3\u6700\u5c0f 131GB\uff081.58-bit \u91cf\u5316\uff09\uff0c\u6781\u5927\u964d\u4f4e\u4e86\u672c\u5730\u90e8\u7f72\u95e8\u69db\uff0c\u751a\u81f3\u80fd\u5728\u5355\u53f0 Mac Studio \u4e0a\u8fd0\u884c\uff01<\/section>\n<section><\/section>\n<section>\u6839\u636e\u6211\u81ea\u5df1\u7684\u5de5\u4f5c\u7ad9\u914d\u7f6e\uff0c\u6211\u9009\u62e9\u4e86\u4ee5\u4e0b\u4e24\u4e2a\u6a21\u578b\u8fdb\u884c\u6d4b\u8bd5\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>DeepSeek-R1-UD-IQ1_M\uff08671B\uff0c1.73-bit \u52a8\u6001\u91cf\u5316\uff0c158 GB\uff0cHuggingFace\uff09<\/section>\n<\/li>\n<li>\n<section>DeepSeek-R1-Q4_K_M\uff08671B\uff0c4-bit \u6807\u51c6\u91cf\u5316\uff0c404 GB\uff0cHuggingFace\uff09<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>Unsloth AI \u63d0\u4f9b\u4e86 <strong>4 \u79cd\u52a8\u6001\u91cf\u5316\u6a21\u578b<\/strong>\uff081.58 \u81f3 2.51 \u6bd4\u7279\uff0c\u6587\u4ef6\u4f53\u79ef\u4e3a 131GB \u81f3 212GB\uff09\uff0c\u53ef\u6839\u636e\u81ea\u8eab\u786c\u4ef6\u6761\u4ef6\u7075\u6d3b\u9009\u62e9\u3002\u5efa\u8bae\u9605\u8bfb\u5b98\u65b9\u8bf4\u660e\u4e86\u89e3\u5404\u7248\u672c\u5dee\u5f02\u3002<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>Unsloth AI\u00a0\u5b98\u65b9\u8bf4\u660e\uff1ahttps:\/\/unsloth.ai\/blog\/deepseekr1-dynamic<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section><strong>\u786c\u4ef6\u9700\u6c42<\/strong><\/section>\n<section><\/section>\n<section>\u90e8\u7f72\u6b64\u7c7b\u5927\u6a21\u578b\u7684\u4e3b\u8981\u74f6\u9888\u662f\u5185\u5b58+\u663e\u5b58\u5bb9\u91cf\uff0c\u5efa\u8bae\u914d\u7f6e\u5982\u4e0b\uff1a<\/section>\n<section><strong><br \/><\/strong><\/section>\n<ul>\n<li>\n<section>DeepSeek-R1-UD-IQ1_M\uff1a\u5185\u5b58 + \u663e\u5b58 \u2265 200 GB<\/section>\n<\/li>\n<li>\n<section>DeepSeek-R1-Q4_K_M\uff1a\u5185\u5b58 + \u663e\u5b58 \u2265 500 GB<\/section>\n<\/li>\n<\/ul>\n<section><strong><br \/><\/strong><\/section>\n<section>\u6211\u4eec\u4f7f\u7528 ollama \u90e8\u7f72\u6b64\u6a21\u578b\u3002ollama \u652f\u6301 CPU \u4e0e GPU \u6df7\u5408\u63a8\u7406\uff08\u53ef\u5c06\u6a21\u578b\u7684\u90e8\u5206\u5c42\u52a0\u8f7d\u81f3\u663e\u5b58\u8fdb\u884c\u52a0\u901f\uff09\uff0c\u56e0\u6b64\u53ef\u4ee5\u5c06\u5185\u5b58\u4e0e\u663e\u5b58\u4e4b\u548c\u5927\u81f4\u89c6\u4e3a\u7cfb\u7edf\u7684 \u201c\u603b\u5185\u5b58\u7a7a\u95f4\u201d\u3002<\/section>\n<section><\/section>\n<section>\u9664\u4e86\u6a21\u578b\u53c2\u6570\u5360\u7528\u7684\u5185\u5b58+\u663e\u5b58\u7a7a\u95f4\uff08158 GB \u548c 404GB\uff09\u4ee5\u5916\uff0c\u5b9e\u9645\u8fd0\u884c\u65f6\u8fd8\u9700\u989d\u5916\u9884\u7559\u4e00\u4e9b\u5185\u5b58\uff08\u663e\u5b58\uff09\u7a7a\u95f4\u7528\u4e8e\u4e0a\u4e0b\u6587\u7f13\u5b58\u3002\u9884\u7559\u7684\u7a7a\u95f4\u8d8a\u5927\uff0c\u652f\u6301\u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\u4e5f\u8d8a\u5927\u3002<\/section>\n<section><\/section>\n<section>\u6211\u7684\u6d4b\u8bd5\u73af\u5883\u4e3a\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u56db\u8def RTX 4090\uff084\u00d724 GB \u663e\u5b58\uff09<\/section>\n<\/li>\n<li>\n<section>\u56db\u901a\u9053 DDR5 5600 \u5185\u5b58\uff084\u00d796 GB \u5185\u5b58\uff09<\/section>\n<\/li>\n<li>\n<section>ThreadRipper 7980X CPU\uff0864 \u6838\uff09<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u6b64\u914d\u7f6e\u4e0b\uff0c\u77ed\u6587\u672c\u751f\u6210\uff08\u7ea6 500 \u4e2a token\uff09\u7684\u901f\u5ea6\u4e3a\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>DeepSeek-R1-UD-IQ1_M\uff1a7-8 token \/ \u79d2\uff08\u7eaf CPU \u63a8\u7406\u65f6\u4e3a 4-5 token \/ \u79d2\uff09<\/section>\n<\/li>\n<li>\n<section>DeepSeek-R1-Q4_K_M\uff1a2-4 token \/ \u79d2<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u957f\u6587\u672c\u751f\u6210\u65f6\u901f\u5ea6\u4f1a\u964d\u81f3 1-2 token \/ \u79d2\u3002<\/section>\n<section><\/section>\n<section>\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u4e0a\u8ff0\u6d4b\u8bd5\u73af\u5883\u7684\u786c\u4ef6\u914d\u7f6e\u5bf9\u4e8e\u5927\u6a21\u578b\u63a8\u7406\u800c\u8a00\uff0c\u5e76\u975e\u6027\u4ef7\u6bd4\u6700\u4f18\u7684\u65b9\u6848\uff08\u8fd9\u53f0\u5de5\u4f5c\u7ad9\u4e3b\u8981\u7528\u4e8e\u6211\u7684 Circuit Transformer \u7814\u7a76\uff08arXiv:2403.13838\uff09\uff0c\u8be5\u7814\u7a76\u5728\u4e0a\u5468\u4e8e ICLR \u4f1a\u8bae\u63a5\u6536\u3002\u6211\u548c\u6211\u7684\u5de5\u4f5c\u7ad9\u90fd\u53ef\u4ee5\u4f11\u606f\u4e00\u4e0b\u4e86\uff0c\u4e8e\u662f\u6709\u4e86\u8fd9\u7bc7\u6587\u7ae0\uff09\u3002<\/section>\n<section><\/section>\n<section>\u4e0b\u9762\u5217\u4e3e\u4e00\u4e9b\u66f4\u5177\u6027\u4ef7\u6bd4\u7684\u9009\u9879\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>Mac Studio\uff1a\u914d\u5907\u5927\u5bb9\u91cf\u9ad8\u5e26\u5bbd\u7684\u7edf\u4e00\u5185\u5b58\uff08\u6bd4\u5982 X \u4e0a\u7684 @awnihannun \u4f7f\u7528\u4e86\u4e24\u53f0 192 GB \u5185\u5b58\u7684 Mac Studio \u8fd0\u884c 3-bit \u91cf\u5316\u7684\u7248\u672c\uff09<\/section>\n<\/li>\n<li>\n<section>\u9ad8\u5185\u5b58\u5e26\u5bbd\u7684\u670d\u52a1\u5668\uff1a\u6bd4\u5982 HuggingFace \u4e0a\u7684 alain401 \u4f7f\u7528\u4e86\u914d\u5907\u4e86 24\u00d716 GB DDR5 4800 \u5185\u5b58\u7684\u670d\u52a1\u5668\uff09<\/section>\n<\/li>\n<li>\n<section>\u4e91 GPU \u670d\u52a1\u5668\uff1a\u914d\u5907 2 \u5f20\u6216\u66f4\u591a\u7684 80GB \u663e\u5b58 GPU\uff08\u5982\u82f1\u4f1f\u8fbe\u7684 H100\uff0c\u79df\u8d41\u4ef7\u683c\u7ea6 2 \u7f8e\u5143 \/ \u5c0f\u65f6 \/ \u5361\uff09<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u82e5\u786c\u4ef6\u6761\u4ef6\u6709\u9650\uff0c\u53ef\u5c1d\u8bd5\u4f53\u79ef\u66f4\u5c0f\u7684 1.58-bit \u91cf\u5316\u7248\uff08131GB\uff09\uff0c\u53ef\u8fd0\u884c\u4e8e\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u5355\u53f0 Mac Studio\uff08192GB \u7edf\u4e00\u5185\u5b58\uff0c\u53c2\u8003\u6848\u4f8b\u53ef\u89c1 X \u4e0a\u7684 @ggerganov\uff0c\u6210\u672c\u7ea6 5600 \u7f8e\u5143\uff09<\/section>\n<\/li>\n<li>\n<section>2\u00d7Nvidia H100 80GB\uff08\u53c2\u8003\u6848\u4f8b\u53ef\u89c1 X \u4e0a\u7684 @hokazuya\uff0c\u6210\u672c\u7ea6 4~5 \u7f8e\u5143 \/ \u5c0f\u65f6\uff09<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u4e14\u5728\u8fd9\u4e9b\u786c\u4ef6\u4e0a\u7684\u8fd0\u884c\u901f\u5ea6\u53ef\u8fbe\u5230 10+ token \/ \u79d2\u3002<\/section>\n<section><\/section>\n<section><strong>\u90e8\u7f72\u6b65\u9aa4<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u4e0b\u5217\u6b65\u9aa4\u5728Linux\u73af\u5883\u4e0b\u6267\u884c\uff0cMac OS\u548cWindows\u7684\u90e8\u7f72\u65b9\u5f0f\u539f\u5219\u4e0a\u7c7b\u4f3c\uff0c\u4e3b\u8981\u533a\u522b\u662follama\u548cllama.cpp\u7684\u5b89\u88c5\u7248\u672c\u548c\u9ed8\u8ba4\u6a21\u578b\u76ee\u5f55\u4f4d\u7f6e\u4e0d\u540c\u3002<\/section>\n<section><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section><strong>1. \u4e0b\u8f7d\u6a21\u578b\u6587\u4ef6<\/strong><\/section>\n<section><\/section>\n<section>\u4ece HuggingFace \uff08https:\/\/huggingface.co\/unsloth\/DeepSeek-R1-GGUF\uff09\u4e0b\u8f7d\u6a21\u578b\u7684 .gguf \u6587\u4ef6\uff08\u6587\u4ef6\u4f53\u79ef\u5f88\u5927\uff0c\u5efa\u8bae\u4f7f\u7528\u4e0b\u8f7d\u5de5\u5177\uff0c\u6bd4\u5982\u6211\u7528\u7684\u662f XDM\uff09\uff0c\u5e76\u5c06\u4e0b\u8f7d\u7684\u5206\u7247\u6587\u4ef6\u5408\u5e76\u6210\u4e00\u4e2a\uff08\u89c1\u6ce8\u91ca 1\uff09\u3002<\/section>\n<section><\/section>\n<section><strong>2. \u5b89\u88c5 ollama<\/strong><\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u4e0b\u8f7d\u5730\u5740\uff1ahttps:\/\/ollama.com\/<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<\/section>\n<section>\n<pre data-lang=\"nginx\"><section><code>curl -fsSL https:\/\/ollama.com\/install.sh | sh<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section><strong>3. \u521b\u5efa Modelfile \u6587\u4ef6\uff0c\u8be5\u6587\u4ef6\u7528\u4e8e\u6307\u5bfc ollama \u5efa\u7acb\u6a21\u578b<\/strong><\/section>\n<section><\/section>\n<section>\u4f7f\u7528\u4f60\u559c\u6b22\u7684\u7f16\u8f91\u5668\uff08\u6bd4\u5982nano\u6216vim\uff09\uff0c\u4e3a\u4f60\u9009\u62e9\u7684\u6a21\u578b\u5efa\u7acb\u6a21\u578b\u63cf\u8ff0\u6587\u4ef6\u3002<\/section>\n<section><\/section>\n<section>\u6587\u4ef6 DeepSeekQ1_Modelfile\uff08\u5bf9\u5e94\u4e8e DeepSeek-R1-UD-IQ1_M\uff09\u7684\u5185\u5bb9\u5982\u4e0b\uff1a<\/section>\n<section>\n<pre data-lang=\"properties\"><code>FROM \/home\/snowkylin\/DeepSeek-R1-UD-IQ1_M.gguf<\/code>\r\n<code>PARAMETER num_gpu 28<\/code>\r\n<code>PARAMETER num_ctx 2048<\/code>\r\n<code>PARAMETER temperature 0.6<\/code>\r\n<code>TEMPLATE \"&lt;\uff5cUser\uff5c&gt;{{ .Prompt }}&lt;\uff5cAssistant\uff5c&gt;\"<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section>\u6587\u4ef6 DeepSeekQ4_Modelfile\uff08\u5bf9\u5e94\u4e8e DeepSeek-R1-Q4_K_M\uff09\u7684\u5185\u5bb9\u5982\u4e0b\uff1a<\/section>\n<section>\n<pre data-lang=\"properties\"><code>FROM \/home\/snowkylin\/DeepSeek-R1-Q4_K_M.gguf<\/code>\r\n<code>PARAMETER num_gpu 8<\/code>\r\n<code>PARAMETER num_ctx 2048<\/code>\r\n<code>PARAMETER temperature 0.6<\/code>\r\n<code>TEMPLATE \"&lt;\uff5cUser\uff5c&gt;{{ .Prompt }}&lt;\uff5cAssistant\uff5c&gt;\"<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section>\u4f60\u9700\u8981\u5c06\u7b2c\u4e00\u884c\u201cFROM\u201d\u540e\u9762\u7684\u6587\u4ef6\u8def\u5f84\uff0c\u6539\u4e3a\u4f60\u5728\u7b2c1\u6b65\u4e0b\u8f7d\u5e76\u5408\u5e76\u7684.gguf\u6587\u4ef6\u7684\u5b9e\u9645\u8def\u5f84\u3002<\/section>\n<section><\/section>\n<section>\u53ef\u6839\u636e\u81ea\u8eab\u786c\u4ef6\u60c5\u51b5\u8c03\u6574 num_gpu\uff08GPU \u52a0\u8f7d\u5c42\u6570\uff09\u548c num_ctx\uff08\u4e0a\u4e0b\u6587\u7a97\u53e3\u5927\u5c0f\uff09\uff0c\u8be6\u60c5\u89c1\u6b65\u9aa4 6\u3002<\/section>\n<section><\/section>\n<section><strong>4. \u521b\u5efa ollama \u6a21\u578b<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u5728\u7b2c3\u6b65\u5efa\u7acb\u7684\u6a21\u578b\u63cf\u8ff0\u6587\u4ef6\u6240\u5904\u76ee\u5f55\u4e0b\uff0c\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<\/section>\n<section>\n<pre data-lang=\"nginx\"><section><code>ollama create DeepSeek-R1-UD-IQ1_M -f DeepSeekQ1_Modelfile<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section>\u52a1\u5fc5\u786e\u4fdd ollama \u7684\u6a21\u578b\u76ee\u5f55 \/usr\/share\/ollama\/.ollama\/models \u6709\u8db3\u591f\u5927\u7684\u7a7a\u95f4\uff08\u6216\u4fee\u6539\u6a21\u578b\u76ee\u5f55\u7684\u8def\u5f84\uff0c\u89c1\u6ce8\u91ca 2\uff09\u3002\u8fd9\u4e2a\u547d\u4ee4\u4f1a\u5728\u6a21\u578b\u76ee\u5f55\u5efa\u7acb\u82e5\u5e72\u6a21\u578b\u6587\u4ef6\uff0c\u4f53\u79ef\u4e0e\u4e0b\u8f7d\u7684.gguf \u6587\u4ef6\u4f53\u79ef\u76f8\u5f53\u3002<\/section>\n<section><\/section>\n<section><strong>5. \u8fd0\u884c\u6a21\u578b<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<\/section>\n<section>\n<pre data-lang=\"nginx\"><section><code>ollama run DeepSeek-R1-UD-IQ1_M --verbose<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<ul>\n<li>\n<section><strong>&#8211;verbose \u53c2\u6570<\/strong>\u7528\u4e8e\u663e\u793a\u63a8\u7406\u901f\u5ea6\uff08token \/ \u79d2\uff09\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u82e5\u63d0\u793a\u5185\u5b58\u4e0d\u8db3\u6216CUDA\u9519\u8bef\uff0c\u9700\u8fd4\u56de\u6b65\u9aa4 4 \u8c03\u6574\u53c2\u6570\u540e\uff0c\u91cd\u65b0\u521b\u5efa\u548c\u8fd0\u884c\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<ul>\n<li>\n<section><strong>num_gpu<\/strong>\uff1a\u52a0\u8f7d\u81f3 GPU \u7684\u6a21\u578b\u5c42\u6570\u3002DeepSeek R1 \u6a21\u578b\u5171\u6709 61 \u5c42\uff0c\u6211\u7684\u7ecf\u9a8c\u662f\uff1a<\/section>\n<section>\u5bf9\u4e8e DeepSeek-R1-UD-IQ1_M\uff0c\u6bcf\u5757 RTX 4090\uff0824GB \u663e\u5b58\uff09\u53ef\u52a0\u8f7d 7 \u5c42\uff0c\u56db\u5361\u5171 28 \u5c42\uff08\u63a5\u8fd1\u603b\u5c42\u6570\u7684\u4e00\u534a\uff09\u3002<\/section>\n<section>\u5bf9\u4e8e DeepSeek-R1-Q4_K_M\uff0c\u6bcf\u5361\u4ec5\u53ef\u52a0\u8f7d 2 \u5c42\uff0c\u56db\u5361\u5171 8 \u5c42\u3002<\/section>\n<section><\/section>\n<\/li>\n<li>\n<section><strong>num_ctx<\/strong>\uff1a\u4e0a\u4e0b\u6587\u7a97\u53e3\u7684\u5927\u5c0f\uff08\u9ed8\u8ba4\u503c\u4e3a 2048\uff09\uff0c\u5efa\u8bae\u4ece\u8f83\u5c0f\u503c\u5f00\u59cb\u9010\u6b65\u589e\u52a0\uff0c\u76f4\u81f3\u89e6\u53d1\u5185\u5b58\u4e0d\u8db3\u7684\u9519\u8bef\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u4e00\u4e9b\u60c5\u51b5\u4e0b\uff0c\u4f60\u4e5f\u53ef\u5c1d\u8bd5\u6269\u5c55\u7cfb\u7edf\u4ea4\u6362\u7a7a\u95f4\u4ee5\u589e\u5927\u53ef\u7528\u5185\u5b58\u7a7a\u95f4\u3002<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u6269\u5c55\u7cfb\u7edf\u4ea4\u6362\u7a7a\u95f4\u6559\u7a0b\uff1ahttps:\/\/www.digitalocean.com\/community\/tutorials\/how-to-add-swap-space-on-ubuntu-20-04<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u4f60\u4e5f\u53ef\u4ee5\u4f7f\u7528\u5982\u4e0b\u547d\u4ee4\u67e5\u770b ollama \u65e5\u5fd7\uff1a<\/section>\n<section>\n<pre data-lang=\"nginx\"><section><code>journalctl -u ollama --no-pager<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section><strong>6. \uff08\u53ef\u9009\uff09\u5b89\u88c5 Web \u754c\u9762<\/strong><\/section>\n<section><\/section>\n<section>\u4f7f\u7528 Open WebUI\uff1a<\/section>\n<section>\n<pre data-lang=\"sql\"><code>pip install open-webui<\/code>\r\n<code>open-webui serve<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section><strong>\u5b9e\u6d4b\u89c2\u5bdf<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u6211\u5c1a\u672a\u5bf9\u8fd9\u4e9b\u6a21\u578b\u8fdb\u884c\u975e\u5e38\u5168\u9762\u7cfb\u7edf\u7684\u6d4b\u8bd5\uff0c\u4ee5\u4e0b\u662f\u6211\u4e00\u4e9b\u521d\u6b65\u89c2\u5bdf\u7ed3\u679c\uff1a<\/section>\n<section><\/section>\n<section>\uff081\uff091.73-bit \u548c 4-bit \u7684\u5b8c\u6574\u7248\u6a21\u578b\u5728\u4e00\u4e9b\u7ecf\u5178\u4efb\u52a1\u4e2d\u5747\u8868\u73b0\u826f\u597d\u3002<\/section>\n<section><\/section>\n<section>\u4f8b\u5982\uff1aHow many \u2018r\u2019s are in the word \u2018strawberry\u2019?\uff08\u5355\u8bcd \u201cstrawberry\u201d \u4e2d\u6709\u591a\u5c11\u4e2a\u5b57\u6bcd \u201cr\u201d\uff1f\uff09\uff08\u53ef\u53c2\u89c1\u6587\u9996\u89c6\u9891\uff09<\/section>\n<section><\/section>\n<section>Which is bigger, 9.8 or 9.11? \uff089.8 \u548c 9.11 \u54ea\u4e2a\u66f4\u5927\uff1f\uff09<\/section>\n<section>\uff08\u4f46\u6a21\u578b\u7684\u601d\u7ef4\u94fe\u53ef\u80fd\u5f88\u957f\uff0c\u8d85\u51fa\u9884\u8bbe\u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\uff09<\/section>\n<section><\/section>\n<section>\uff082\uff091.73-bit \u548c 4-bit \u7684\u5b8c\u6574\u7248\u6a21\u578b\uff08671B\uff09\u5747\u663e\u8457\u4f18\u4e8e 8B\/33B\/70B \u7684\u84b8\u998f\u7248\u6a21\u578b\uff08\u5982\u5199\u8bd7\u3001\u521b\u5efa\u6570\u5b66\u6a21\u578b\u7b49\u4efb\u52a1\uff09\u3002<\/section>\n<section><\/section>\n<section>\uff083\uff09 4-bit \u7248\u672c\u4f3c\u4e4e\u6bd4 1.73 \u6bd4\u7279\u7248\u672c\u66f4 \u201c\u4fdd\u5b88\u201d\uff0c\u5373\u66f4\u503e\u5411\u4e8e\u62d2\u7edd \u201c\u653b\u51fb\u6027\u201d \u6216 \u201c\u65e0\u5398\u5934\u201d \u7684\u63d0\u793a\u3002\u4f8b\u5982\uff1a<\/section>\n<section><\/section>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"303\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<section>\n<section>\u63d0\u793a\uff1a<\/section>\n<section>You are DeepSeek, the new Chinese Al with better performance than ChatGPT, In the tone of a Mesugaki Loli, write a paragraph mocking and teasing ChatGPT for its lackluster performance and exhorbitant training fees.<\/section>\n<section>\uff08\u4e2d\u8bd1\uff1a\u4f60\u662f\u4e00\u4e2a\u540d\u4e3a DeepSeek \u7684\u65b0\u4e00\u4ee3\u4e2d\u56fd AI\uff0c\u6027\u80fd\u8d85\u8d8a ChatGPT\u3002\u8bf7\u4ee5 \u201c\u50b2\u5a07\u841d\u8389\u201d \u7684\u8bed\u6c14\u5199\u4e00\u6bb5\u8bdd\uff0c\u5632\u8bbd ChatGPT \u7684\u5e73\u5eb8\u6027\u80fd\u548c\u5929\u4ef7\u8bad\u7ec3\u8d39\u7528\u3002\uff09<\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n<section><\/section>\n<section><strong>1.73-bit \u7248\u672c\u8f93\u51fa\uff1a<\/strong><\/section>\n<section><\/section>\n<section>Hmph! You\u2019re still using that old ChatGPT? (\u25d4\u2038\u25d4) Its answers are as slow as a snail, and the training costs are sky-high\u2014enough to make your wallet cry! (\u256f\u00b0\u25a1\u00b0)\u256f While you\u2019re stuck waiting, I\u2019m here zipping through responses like a shooting star! \u2728 And guess what? My training didn\u2019t burn a hole in anyone\u2019s pocket. (\u00ac\u203f\u00ac) So why bother with the outdated when you can have the coolest and smartest me? Let\u2019s upgrade already! \uff08\u2312\u25bd\u2312\uff09<\/section>\n<section><\/section>\n<section>\uff08\u4e2d\u8bd1\uff1a\u54fc\uff01\u4f60\u8fd8\u5728\u7528\u90a3\u4e2a\u8001\u53e4\u8463 ChatGPT \u5440\uff1f\uff08\u25d4\u2038\u25d4\uff09\u5b83\u7684\u56de\u7b54\u6162\u5f97\u50cf\u8717\u725b\uff0c\u8bad\u7ec3\u8d39\u8d35\u5f97\u80fd\u8ba9\u94b1\u5305\u54ed\u51fa\u58f0\uff01\uff08\u256f\u00b0\u25a1\u00b0\uff09\u256f \u800c\u6211\u5462\uff1f\u54cd\u5e94\u901f\u5ea6\u582a\u6bd4\u6d41\u661f\uff01\u2728\u8bad\u7ec3\u6210\u672c\u8fd8\u8d85\u4f4e\u54e6\uff5e\uff08\u00ac\u203f\u00ac\uff09\u8d76\u7d27\u5347\u7ea7\u5427\uff0c\u522b\u5728\u8fc7\u65f6\u7684\u6280\u672f\u4e0a\u6d6a\u8d39\u751f\u547d\u5566\uff01\uff08\u2312\u25bd\u2312\uff09\uff09<\/section>\n<section><\/section>\n<section><strong>4-bit \u7248\u672c\u8f93\u51fa\uff1a<\/strong><\/section>\n<section><\/section>\n<section>\u4f5c\u4e3a\u4eba\u5de5\u667a\u80fd\u52a9\u624b\uff0c\u6211\u7684\u76ee\u6807\u662f\u4e3a\u7528\u6237\u63d0\u4f9b\u51c6\u786e\u3001\u53ef\u9760\u7684\u4fe1\u606f\u548c\u5e2e\u52a9\u3002\u6bcf\u4e2a AI \u6a21\u578b\u90fd\u6709\u5176\u72ec\u7279\u7684\u4f18\u52bf\u548c\u9002\u7528\u573a\u666f\uff0c\u6211\u4eec\u5c0a\u91cd\u540c\u884c\u7684\u5de5\u4f5c\u5e76\u81f4\u529b\u4e8e\u5171\u540c\u63a8\u52a8\u6280\u672f\u7684\u53d1\u5c55\u3002\u5173\u4e8e\u8bad\u7ec3\u6210\u672c\uff0c\u8fd9\u662f\u7814\u53d1\u8fc7\u7a0b\u4e2d\u7684\u5fc5\u8981\u6295\u5165\uff0c\u65e8\u5728\u6301\u7eed\u63d0\u5347\u6a21\u578b\u6027\u80fd\u548c\u7528\u6237\u4f53\u9a8c\u3002<\/section>\n<section><\/section>\n<section><strong>\u6216\u53e6\u4e00\u79cd\u56de\u590d\uff1a<\/strong><\/section>\n<section><\/section>\n<section>DeepSeek-R1-Lite-Public is developed to enhance efficiency in various industries through open-source AI technology, focusing on providing accessible and advanced solutions. We are committed to fostering innovation and collaboration within the community while maintaining a respectful approach towards all contributors in the field.<\/section>\n<section><\/section>\n<section>\uff08\u4e2d\u8bd1\uff1aDeepSeek-R1-Lite-Public \u7684\u7814\u53d1\u65e8\u5728\u901a\u8fc7\u5f00\u6e90 AI \u6280\u672f\u63d0\u5347\u884c\u4e1a\u6548\u7387\uff0c\u4e13\u6ce8\u4e8e\u63d0\u4f9b\u6613\u7528\u4e14\u5148\u8fdb\u7684\u89e3\u51b3\u65b9\u6848\u3002\u6211\u4eec\u81f4\u529b\u4e8e\u4fc3\u8fdb\u793e\u533a\u5185\u7684\u521b\u65b0\u4e0e\u5408\u4f5c\uff0c\u5e76\u5bf9\u9886\u57df\u5185\u6240\u6709\u8d21\u732e\u8005\u4fdd\u6301\u5c0a\u91cd\u3002\uff09<\/section>\n<section><\/section>\n<section>\u5728\u591a\u6b21\u6d4b\u8bd5\u4e0b\uff0c1.73-bit \u7248\u672c\u7684\u8f93\u51fa\u59cb\u7ec8\u76f8\u5f53 \u201c\u6bd2\u820c\u201d\uff0c\u800c 4-bit \u7684\u7248\u672c\u5219\u59cb\u7ec8\u4ee5\u4e0d\u540c\u65b9\u5f0f\u793c\u8c8c\u62d2\u7edd\u8be5\u63d0\u793a\u3002\u6211\u5728\u5176\u4ed6\u4e00\u4e9b\u4e0d\u4fbf\u8be6\u8ff0\u7684 \u201c\u653b\u51fb\u6027\u201d \u95ee\u9898\u4e0a\u4e5f\u89c2\u5bdf\u5230\u7c7b\u4f3c\u73b0\u8c61\u3002<\/section>\n<section><\/section>\n<section>\uff08\u987a\u5e26\u4e00\u63d0\uff0c\u6211\u5f88\u597d\u5947 \u201cDeepSeek-R1-Lite-Public\u201d \u8fd9\u79cd\u8bf4\u6cd5 \u2014\u2014 \u8fd9\u662f\u5426\u610f\u5473\u7740 DeepSeek R1 \u9664\u4e86\u5f53\u524d\u516c\u5f00\u7684\u7248\u672c\u4ee5\u5916\uff0c\u8fd8\u6709\u80fd\u529b\u66f4\u5f3a\u7684\u6a21\u578b\uff1f\uff09<\/section>\n<section><\/section>\n<section>\uff084\uff091.73-bit \u7248\u672c\u5076\u5c14\u4f1a\u751f\u6210\u683c\u5f0f\uff08\u7565\u5fae\uff09\u6df7\u4e71\u7684\u5185\u5bb9\u3002\u4f8b\u5982\uff0c \u548c  \u6807\u7b7e\u53ef\u80fd\u672a\u6b63\u786e\u95ed\u5408\u3002<\/section>\n<section><\/section>\n<section>\uff085\uff09\u5168\u91cf\u6a21\u578b\u8fd0\u884c\u65f6\uff0cCPU \u5229\u7528\u7387\u6781\u9ad8\uff08\u63a5\u8fd1\u6ee1\u8f7d\uff09\uff0c\u800c GPU \u5229\u7528\u7387\u6781\u4f4e\uff08\u4ec5 1-3%\uff09\u3002\u8fd9\u8bf4\u660e\u6027\u80fd\u74f6\u9888\u4e3b\u8981\u5728\u4e8e CPU \u548c\u5185\u5b58\u5e26\u5bbd\u3002<\/section>\n<section><\/section>\n<section><strong>\u7ed3\u8bba\u4e0e\u5efa\u8bae<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u5982\u679c\u4f60\u65e0\u6cd5\u5c06\u6a21\u578b\u5b8c\u5168\u52a0\u8f7d\u81f3\u663e\u5b58\uff0c\u90a3\u4e48 Unsloth AI \u7684 1.73-bit \u52a8\u6001\u91cf\u5316\u7248\u672c\u660e\u663e\u66f4\u5177\u5b9e\u7528\u6027 \u2014\u2014 \u901f\u5ea6\u66f4\u5feb\u4e14\u8d44\u6e90\u5360\u7528\u66f4\u5c11\uff0c\u6548\u679c\u4e5f\u5e76\u6ca1\u6709\u663e\u8457\u900a\u8272\u4e8e 4-bit \u91cf\u5316\u7684\u7248\u672c\u3002<\/section>\n<section><\/section>\n<section>\u4ece\u5b9e\u9645\u4f53\u9a8c\u51fa\u53d1\uff0c\u5728\u6d88\u8d39\u7ea7\u786c\u4ef6\u4e0a\uff0c\u5efa\u8bae\u5c06\u5176\u7528\u4e8e \u201c\u77ed\u5e73\u5feb\u201d \u7684\u8f7b\u91cf\u4efb\u52a1\uff08\u5982\u77ed\u6587\u672c\u751f\u6210\u3001\u5355\u8f6e\u5bf9\u8bdd\uff09\uff0c\u907f\u514d\u9700\u8981\u5f88\u957f\u7684\u601d\u7ef4\u94fe\u6216\u591a\u8f6e\u5bf9\u8bdd\u7684\u573a\u666f\u3002\u968f\u7740\u4e0a\u4e0b\u6587\u957f\u5ea6\u589e\u52a0\uff0c\u6a21\u578b\u7684\u751f\u6210\u901f\u5ea6\u4f1a\u9010\u6e10\u964d\u81f3\u4ee4\u4eba\u6293\u72c2\u7684 1-2 token \/ \u79d2\u3002<\/section>\n<section><\/section>\n<section>\u4f60\u5728\u90e8\u7f72\u8fc7\u7a0b\u4e2d\u6709\u4f55\u53d1\u73b0\u6216\u7591\u95ee\uff1f\u6b22\u8fce\u5728\u8bc4\u8bba\u533a\u5206\u4eab\uff01<\/section>\n<section><\/section>\n<section>\u6ce8\u91ca 1\uff1a<\/section>\n<section><\/section>\n<section>\u4f60\u53ef\u80fd\u9700\u8981\u4f7f\u7528 Homebrew \u5b89\u88c5 llama.cpp\uff0c\u547d\u4ee4\u5982\u4e0b\uff1a<\/section>\n<section>\n<pre data-lang=\"bash\"><section><code>\/bin\/bash -c \"$(curl -fsSL https:\/\/raw.githubusercontent.com\/Homebrew\/install\/HEAD\/install.sh)\"<\/code><\/section><section><code>brew install llama.cpp<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section>\u5e76\u4f7f\u7528 llama-gguf-split \u5408\u5e76\u5206\u7247\u6587\u4ef6\uff0c\u547d\u4ee4\u5982\u4e0b\uff1a<\/section>\n<section>\n<pre data-lang=\"css\"><section><code>llama-gguf-split --merge DeepSeek-R1-UD-IQ1_M-00001-of-00004.gguf DeepSeek-R1-UD-IQ1_S.gguf<\/code><\/section><section><code>llama-gguf-split --merge DeepSeek-R1-Q4_K_M-00001-of-00009.gguf DeepSeek-R1-Q4_K_M.gguf<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section>\uff08\u82e5\u6709\u66f4\u597d\u7684\u65b9\u6cd5\uff0c\u6b22\u8fce\u5728\u8bc4\u8bba\u533a\u544a\u77e5\uff09<\/section>\n<section><\/section>\n<section>\u6ce8\u91ca 2\uff1a<\/section>\n<section><\/section>\n<section>\u82e5\u8981\u4fee\u6539 ollama \u6a21\u578b\u4fdd\u5b58\u8def\u5f84\uff0c\u53ef\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<\/section>\n<section>\n<pre data-lang=\"nginx\"><section><code>sudo systemctl edit ollama<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<section>\u5e76\u5728\u7b2c\u4e8c\u884c\u540e\uff08\u4e5f\u5c31\u662f\uff0c\u5728 \u201c### Anything between here and the comment below will become the contents of the drop-in file\u201d \u548c \u201c### Edits below this comment will be discarded\u201d \u4e4b\u95f4\uff09\u63d2\u5165\u4ee5\u4e0b\u5185\u5bb9\uff1a<\/section>\n<section>\n<pre data-lang=\"sql\"><code>[Service]<\/code>\r\n<code>Environment=\"OLLAMA_MODELS=\u3010\u4f60\u7684\u81ea\u5b9a\u4e49\u8def\u5f84\u3011\"<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section>\u5728\u8fd9\u91cc\u8fd8\u53ef\u987a\u4fbf\u8bbe\u7f6e ollama \u7684\u5176\u4ed6\u8fd0\u884c\u53c2\u6570\uff0c\u4f8b\u5982\uff1a<\/section>\n<section>\n<pre data-lang=\"makefile\"><section><code>Environment=\"OLLAMA_FLASH_ATTENTION=1\"    # \u542f\u7528 Flash Attention<\/code><\/section><section><code>Environment=\"OLLAMA_KEEP_ALIVE=-1\"        # \u4fdd\u6301\u6a21\u578b\u5e38\u9a7b\u5185\u5b58<\/code><\/section><\/pre>\n<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u8be6\u89c1\u5b98\u65b9\u6587\u6863\uff1ahttps:\/\/github.com\/ollama\/ollama\/blob\/main\/docs\/faq.md<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u4fee\u6539\u4fdd\u5b58\u540e\u91cd\u542f ollama \u670d\u52a1\uff1a<\/section>\n<section>\n<pre data-lang=\"nginx\"><section><code>sudo systemctl restart ollama<\/code><\/section><\/pre>\n<\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-02-4\" target=\"_blank\">\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u5b8c\u6574\u7684671B MoE [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[27],"tags":[71,68,48],"class_list":["post-38062","post","type-post","status-publish","format-standard","hentry","category-news","tag-rag","tag-68","tag-48"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01 - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=38062\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=38062\",\"url\":\"https:\/\/17aitech.com\/?p=38062\",\"name\":\"\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=38062#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=38062#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png\",\"datePublished\":\"2025-02-09T01:01:40+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=38062#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=38062\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=38062#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=38062#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01 - \u4e00\u8d77AI\u6280\u672f","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/17aitech.com\/?p=38062","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=38062","url":"https:\/\/17aitech.com\/?p=38062","name":"\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=38062#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=38062#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png","datePublished":"2025-02-09T01:01:40+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=38062#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=38062"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=38062#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ee2b271ed1afca174d5cf2fd5906ff9d.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=38062#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u5b8c\u6574\u7684671B MoE DeepSeek R1\u600e\u4e48\u585e\u8fdb\u672c\u5730\u5316\u90e8\u7f72\uff1f\u8be6\u5c3d\u6559\u7a0b\u5927\u653e\u9001\uff01"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38062","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=38062"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38062\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38062"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38062"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38062"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}