{"id":16538,"date":"2024-12-27T10:41:49","date_gmt":"2024-12-27T02:41:49","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=16538"},"modified":"2024-12-27T10:43:41","modified_gmt":"2024-12-27T02:43:41","slug":"deepseek-fabule-v","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/deepseek-fabule-v\/","title":{"rendered":"DeepSeek \u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u5f00\u6e90\u7248\u672c\uff0c\u73b0\u9636\u6bb5\u62e5\u6709\uff08\u56fd\u4ea7\uff09\u6700\u5f3a\u4ee3\u7801\u80fd\u529b"},"content":{"rendered":"<blockquote><p>DeepSeek-V3\u662f\u4e00\u6b3e\u5f3a\u5927\u7684\u6df7\u5408\u4e13\u5bb6\uff08Mixture-of-Experts, MoE\uff09\u8bed\u8a00\u6a21\u578b\uff0c\u62e5\u67096710\u4ebf\u603b\u53c2\u6570\uff0c\u9488\u5bf9\u6bcf\u4e2atoken\u6fc0\u6d3b37\u4ebf\u53c2\u6570\u3002\u8be5\u6a21\u578b\u91c7\u7528\u4e86\u4e00\u79cd\u521b\u65b0\u7684\u591a\u5934\u6f5c\u5728\u6ce8\u610f\u529b\uff08Multi-head Latent Attention, MLA\uff09\u67b6\u6784\uff0c\u4ee5\u53ca\u7ecf\u8fc7\u5145\u5206\u9a8c\u8bc1\u7684DeepSeekMoE\u67b6\u6784\u3002CogAgent\u5b9e\u65bd\u4e86\u65e0\u8f85\u52a9\u635f\u5931\u7684\u8d1f\u8f7d\u5747\u8861\u7b56\u7565\uff0c\u5e76\u63d0\u51fa\u4e86\u591atoken\u9884\u6d4b\u8bad\u7ec3\u76ee\u6807\uff0c\u4ece\u800c\u663e\u8457\u63d0\u5347\u6a21\u578b\u6027\u80fd\u3002\u5b83\u57281480\u4e07\u591a\u6837\u7684\u9ad8\u8d28\u91cftoken\u4e0a\u8fdb\u884c\u4e86\u9884\u8bad\u7ec3\uff0c\u5e76\u7ecf\u8fc7\u76d1\u7763\u5fae\u8c03\u548c\u5f3a\u5316\u5b66\u4e60\u9636\u6bb5\uff0c\u4ece\u800c\u5145\u5206\u6316\u6398\u5176\u6f5c\u529b\u3002<\/p>\n<p>DeepSeek-V3\u5728\u8bf8\u591a\u6807\u51c6\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u4f18\u5f02\uff0c\u5c24\u5176\u5728\u6570\u5b66\u548c\u4ee3\u7801\u4efb\u52a1\u4e0a\u8868\u73b0\u7a81\u51fa\uff0c\u6210\u4e3a\u5f53\u524d\u6700\u5f3a\u7684\u5f00\u6e90\u57fa\u7840\u6a21\u578b\uff0c\u4e14\u8bad\u7ec3\u6210\u672c\u4f4e\u5ec9\uff0c\u5176\u5168\u7a0b\u8bad\u7ec3\u7684\u7a33\u5b9a\u6027\u4e5f\u5f97\u5230\u4e86\u9ad8\u5ea6\u8ba4\u53ef\u3002<\/p><\/blockquote>\n<p>\u6628\u65e5\uff0cDeepSeek\u5168\u65b0\u7cfb\u5217\u6a21\u578b DeepSeek-V3 \u9996\u4e2a\u7248\u672c\u4e0a\u7ebf\u5e76\u540c\u6b65\u5f00\u6e90\u3002\u767b\u5f55\u5b98\u7f51 chat.deepseek.com \u5373\u53ef\u4e0e\u6700\u65b0\u7248 V3 \u6a21\u578b\u5bf9\u8bdd\u3002API \u670d\u52a1\u5df2\u540c\u6b65\u66f4\u65b0\uff0c\u63a5\u53e3\u914d\u7f6e\u65e0\u9700\u6539\u52a8\u3002\u5f53\u524d\u7248\u672c\u7684 DeepSeek-V3 \u6682\u4e0d\u652f\u6301\u591a\u6a21\u6001\u8f93\u5165\u8f93\u51fa\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2><strong>\u6027\u80fd\u5bf9\u9f50\u6d77\u5916\u9886\u519b\u95ed\u6e90\u6a21\u578b<\/strong><\/h2>\n<p>DeepSeek-V3 \u4e3a\u81ea\u7814 MoE \u6a21\u578b\uff0c671B \u53c2\u6570\uff0c\u6fc0\u6d3b 37B\uff0c\u5728 14.8T <a href=\"https:\/\/www.kdjingpai.com\/en\/tokenization\/\">token<\/a> \u4e0a\u8fdb\u884c\u4e86\u9884\u8bad\u7ec3\u3002<\/p>\n<p>\u8bba\u6587\u94fe\u63a5\uff1a<\/p>\n<p>https:\/\/github.com\/deepseek-ai\/DeepSeek-V3\/blob\/main\/DeepSeek_V3.pdf<\/p>\n<p>DeepSeek-V3 \u591a\u9879\u8bc4\u6d4b\u6210\u7ee9\u8d85\u8d8a\u4e86 Qwen2.5-72B \u548c Llama-3.1-405B \u7b49\u5176\u4ed6\u5f00\u6e90\u6a21\u578b\uff0c\u5e76\u5728\u6027\u80fd\u4e0a\u548c\u4e16\u754c\u9876\u5c16\u7684\u95ed\u6e90\u6a21\u578b GPT-4o \u4ee5\u53ca Claude-3.5-Sonnet \u4e0d\u5206\u4f2f\u4ef2\u3002<br \/>\n<img decoding=\"async\" title=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/0a1b62ab6bf8b71.png\" alt=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" \/><\/p>\n<ul>\n<li><strong>\u767e\u79d1\u77e5\u8bc6<\/strong>\uff1aDeepSeek-V3 \u5728\u77e5\u8bc6\u7c7b\u4efb\u52a1\uff08MMLU, MMLU-Pro, GPQA, SimpleQA\uff09\u4e0a\u7684\u6c34\u5e73\u76f8\u6bd4\u524d\u4ee3 DeepSeek-V2.5 \u663e\u8457\u63d0\u5347\uff0c\u63a5\u8fd1\u5f53\u524d\u8868\u73b0\u6700\u597d\u7684\u6a21\u578b Claude-3.5-Sonnet-1022\u3002<\/li>\n<li><strong>\u957f\u6587\u672c<\/strong>\uff1a\u5728\u957f\u6587\u672c\u6d4b\u8bc4\u4e2d\uff0cDROP\u3001FRAMES \u548c LongBench v2 \u4e0a\uff0cDeepSeek-V3 \u5e73\u5747\u8868\u73b0\u8d85\u8d8a\u5176\u4ed6\u6a21\u578b\u3002<\/li>\n<li><strong>\u4ee3\u7801<\/strong>\uff1a<strong>DeepSeek-V3 \u5728\u7b97\u6cd5\u7c7b\u4ee3\u7801\u573a\u666f\uff08Codeforces\uff09\uff0c\u8fdc\u8fdc\u9886\u5148\u4e8e\u5e02\u9762\u4e0a\u5df2\u6709\u7684\u5168\u90e8\u975e o1 \u7c7b\u6a21\u578b<\/strong>\uff1b\u5e76\u5728\u5de5\u7a0b\u7c7b\u4ee3\u7801\u573a\u666f\uff08SWE-Bench Verified\uff09\u903c\u8fd1 Claude-3.5-Sonnet-1022\u3002<\/li>\n<li><strong>\u6570\u5b66<\/strong>\uff1a\u5728\u7f8e\u56fd\u6570\u5b66\u7ade\u8d5b\uff08AIME 2024, MATH\uff09\u548c\u5168\u56fd\u9ad8\u4e2d\u6570\u5b66\u8054\u8d5b\uff08CNMO 2024\uff09\u4e0a\uff0cDeepSeek-V3 \u5927\u5e45\u8d85\u8fc7\u4e86\u6240\u6709\u5f00\u6e90\u95ed\u6e90\u6a21\u578b\u3002<\/li>\n<li><strong>\u4e2d\u6587\u80fd\u529b<\/strong>\uff1aDeepSeek-V3 \u4e0e Qwen2.5-72B \u5728\u6559\u80b2\u7c7b\u6d4b\u8bc4 C-Eval \u548c\u4ee3\u8bcd\u6d88\u6b67\u7b49\u8bc4\u6d4b\u96c6\u4e0a\u8868\u73b0\u76f8\u8fd1\uff0c\u4f46\u5728\u4e8b\u5b9e\u77e5\u8bc6 C-SimpleQA \u4e0a\u66f4\u4e3a\u9886\u5148\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<p><img decoding=\"async\" title=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/530291920ee636f.jpg\" alt=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2><strong>\u751f\u6210\u901f\u5ea6\u63d0\u5347\u81f3 3 \u500d<\/strong><\/h2>\n<p>\u901a\u8fc7\u7b97\u6cd5\u548c\u5de5\u7a0b\u4e0a\u7684\u521b\u65b0\uff0cDeepSeek-V3 \u7684\u751f\u6210\u5410\u5b57\u901f\u5ea6\u4ece 20 TPS \u5927\u5e45\u63d0\u9ad8\u81f3 60 TPS\uff0c\u76f8\u6bd4 V2.5 \u6a21\u578b\u5b9e\u73b0\u4e86 3 \u500d\u7684\u63d0\u5347\uff0c\u4e3a\u7528\u6237\u5e26\u6765\u66f4\u52a0\u8fc5\u901f\u6d41\u7545\u7684\u4f7f\u7528\u4f53\u9a8c\u3002 <img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/ed7d4638a10304d.gif\" alt=\"\" \/><\/p>\n<h2><strong>API \u670d\u52a1\u4ef7\u683c\u8c03\u6574<\/strong><\/h2>\n<p>\u968f\u7740\u6027\u80fd\u66f4\u5f3a\u3001\u901f\u5ea6\u66f4\u5feb\u7684 DeepSeek-V3 \u66f4\u65b0\u4e0a\u7ebf\uff0c\u6211\u4eec\u7684\u6a21\u578b API \u670d\u52a1\u5b9a\u4ef7\u4e5f\u5c06\u8c03\u6574\u4e3a<strong>\u6bcf\u767e\u4e07\u8f93\u5165 tokens 0.5 \u5143\uff08\u7f13\u5b58\u547d\u4e2d\uff09\/ 2 \u5143\uff08\u7f13\u5b58\u672a\u547d\u4e2d\uff09\uff0c\u6bcf\u767e\u4e07\u8f93\u51fa tokens 8 \u5143<\/strong>\uff0c\u4ee5\u671f\u80fd\u591f\u6301\u7eed\u5730\u4e3a\u5927\u5bb6\u63d0\u4f9b\u66f4\u597d\u7684\u6a21\u578b\u670d\u52a1\u3002 <img decoding=\"async\" title=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/d89c12aa3fa06de.png\" alt=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" \/> \u4e0e\u6b64\u540c\u65f6\uff0c\u6211\u4eec\u51b3\u5b9a\u4e3a\u5168\u65b0\u6a21\u578b\u8bbe\u7f6e\u957f\u8fbe 45 \u5929\u7684\u4f18\u60e0\u4ef7\u683c\u4f53\u9a8c\u671f\uff1a\u5373\u65e5\u8d77\u81f3 2025 \u5e74 2 \u6708 8 \u65e5\uff0cDeepSeek-V3 \u7684 API \u670d\u52a1\u4ef7\u683c\u4ecd\u7136\u4f1a\u662f\u5927\u5bb6\u719f\u6089\u7684<strong>\u6bcf\u767e\u4e07\u8f93\u5165 tokens 0.1 \u5143\uff08\u7f13\u5b58\u547d\u4e2d\uff09\/ 1 \u5143\uff08\u7f13\u5b58\u672a\u547d\u4e2d\uff09\uff0c\u6bcf\u767e\u4e07\u8f93\u51fa tokens 2 \u5143<\/strong>\uff0c\u5df2\u7ecf\u6ce8\u518c\u7684\u8001\u7528\u6237\u548c\u5728\u6b64\u671f\u95f4\u5185\u6ce8\u518c\u7684\u65b0\u7528\u6237\u5747\u53ef\u4eab\u53d7\u4ee5\u4e0a\u4f18\u60e0\u4ef7\u683c\u3002 <img decoding=\"async\" title=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/12\/ace272ff87d4d5a.png\" alt=\"DeepSeek \u521a\u521a\u53d1\u5e03\u4e86 v3 \u6a21\u578b\u7684\u9996\u4e2a\u7248\u672c\uff0c\u5f00\u6e90\u4e0a\u7ebf\u3002-1\" \/><\/p>\n<h2><strong>\u5f00\u6e90\u6743\u91cd\u548c\u672c\u5730\u90e8\u7f72<\/strong><\/h2>\n<p>DeepSeek-V3 \u91c7\u7528 FP8 \u8bad\u7ec3\uff0c\u5e76\u5f00\u6e90\u4e86\u539f\u751f FP8 \u6743\u91cd\u3002\u5f97\u76ca\u4e8e\u5f00\u6e90\u793e\u533a\u7684\u652f\u6301\uff0cSGLang \u548c LMDeploy \u00a0\u7b2c\u4e00\u65f6\u95f4\u652f\u6301\u4e86 V3 \u6a21\u578b\u7684\u539f\u751f FP8 \u63a8\u7406\uff0c\u540c\u65f6 TensorRT-LLM \u548c MindIE \u5219\u5b9e\u73b0\u4e86 BF16 \u63a8\u7406\u3002\u6b64\u5916\uff0c\u4e3a\u65b9\u4fbf\u793e\u533a\u9002\u914d\u548c\u62d3\u5c55\u5e94\u7528\u573a\u666f\uff0c\u6211\u4eec\u63d0\u4f9b\u4e86\u4ece FP8 \u5230 BF16 \u7684\u8f6c\u6362\u811a\u672c\u3002<\/p>\n<p>\u6a21\u578b\u6743\u91cd\u4e0b\u8f7d\u548c\u66f4\u591a\u672c\u5730\u90e8\u7f72\u4fe1\u606f\u8bf7\u53c2\u8003\uff1a<\/p>\n<p>https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-V3-Base<\/p>\n","protected":false},"excerpt":{"rendered":"<p>DeepSeek-V3\u662f\u4e00\u6b3e\u5f3a\u5927\u7684\u6df7\u5408\u4e13\u5bb6\uff08Mixture-of-Experts, MoE\uff09\u8bed\u8a00\u6a21\u578b\uff0c\u62e5\u67096710\u4ebf\u603b\u53c2\u6570\uff0c\u9488\u5bf9\u6bcf\u4e2atoken\u6fc0\u6d3b37\u4ebf\u53c2\u6570\u3002\u8be5\u6a21\u578b\u91c7\u7528\u4e86\u4e00\u79cd\u521b\u65b0\u7684\u591a\u5934\u6f5c\u5728\u6ce8\u610f\u529b\uff08Multi-head Latent Atte&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[46],"tags":[],"class_list":["post-16538","post","type-post","status-publish","format-standard","hentry","category-news"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/16538","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/comments?post=16538"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/16538\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=16538"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=16538"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=16538"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}