{"id":19453,"date":"2025-01-27T09:48:37","date_gmt":"2025-01-27T01:48:37","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=19453"},"modified":"2025-01-27T09:48:37","modified_gmt":"2025-01-27T01:48:37","slug":"qwen25-1m-zhichi10","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/en\/qwen25-1m-zhichi10\/","title":{"rendered":"Qwen2.5-1M: \u652f\u6301100\u4e07Tokens\u4e0a\u4e0b\u6587\u7684\u5f00\u6e90Qwen\u6a21\u578b"},"content":{"rendered":"<h2>1.\u5f15\u8a00<\/h2>\n<p>\u4e24\u4e2a\u6708\u524d\uff0cQwen\u56e2\u961f\u5347\u7ea7\u4e86 Qwen2.5-Turbo\uff0c\u4f7f\u5176\u652f\u6301\u6700\u591a\u4e00\u767e\u4e07\u4e2aTokens\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u3002\u4eca\u5929\uff0cQwen\u6b63\u5f0f\u63a8\u51fa\u5f00\u6e90\u7684 Qwen2.5-1M \u6a21\u578b\u53ca\u5176\u5bf9\u5e94\u7684\u63a8\u7406\u6846\u67b6\u652f\u6301\u3002\u4ee5\u4e0b\u662f\u672c\u6b21\u53d1\u5e03\u7684\u4eae\u70b9\uff1a<\/p>\n<blockquote><p><strong>\u5f00\u6e90\u6a21\u578b\uff1a\u00a0<\/strong>\u672c\u6b21\u53d1\u5e03\u4e86\u4e24\u4e2a\u65b0\u7684\u5f00\u6e90\u6a21\u578b\uff0c\u5206\u522b\u662f <strong>Qwen2.5-7B-Instruct-1M<\/strong> \u548c <strong>Qwen2.5-14B-Instruct-1M<\/strong>\uff0c\u8fd9\u662fQwen\u9996\u6b21\u5c06\u5f00\u6e90\u7684 Qwen \u6a21\u578b\u7684\u4e0a\u4e0b\u6587\u6269\u5c55\u5230 1M \u957f\u5ea6\u3002<\/p>\n<p><strong>\u63a8\u7406\u6846\u67b6\uff1a<\/strong>\u00a0\u4e3a\u4e86\u5e2e\u52a9\u5f00\u53d1\u8005\u66f4\u9ad8\u6548\u5730\u90e8\u7f72 Qwen2.5-1M \u7cfb\u5217\u6a21\u578b\uff0cQwen\u56e2\u961f\u5b8c\u5168\u5f00\u6e90\u4e86\u57fa\u4e8e <a href=\"https:\/\/www.kdjingpai.com\/vllm\/\">vLLM<\/a> \u7684\u63a8\u7406\u6846\u67b6\uff0c\u5e76\u96c6\u6210\u4e86\u7a00\u758f\u6ce8\u610f\u529b\u65b9\u6cd5\uff0c\u4f7f\u5f97\u8be5\u6846\u67b6\u5728\u5904\u7406 1M \u6807\u8bb0\u8f93\u5165\u65f6\u7684\u901f\u5ea6\u63d0\u5347\u4e86<strong> 3\u500d\u52307\u500d<\/strong>\u3002<\/p>\n<p><strong>\u6280\u672f\u62a5\u544a\uff1a<\/strong> Qwen\u56e2\u961f\u8fd8\u5206\u4eab\u4e86 Qwen2.5-1M \u7cfb\u5217\u80cc\u540e\u7684\u6280\u672f\u7ec6\u8282\uff0c\u5305\u62ec\u8bad\u7ec3\u548c\u63a8\u7406\u6846\u67b6\u7684\u8bbe\u8ba1\u601d\u8def\u4ee5\u53ca\u6d88\u878d\u5b9e\u9a8c\u7684\u7ed3\u679c\u3002<\/p><\/blockquote>\n<p>&nbsp;<\/p>\n<p><strong>\u6a21\u578b\u94fe\u63a5\uff1a<\/strong>https:\/\/www.modelscope.cn\/collections\/Qwen25-1M-d6cf9fd33f0a40<\/p>\n<p><strong>\u6280\u672f\u62a5\u544a\uff1a<\/strong>https:\/\/qianwen-res.oss-cn-beijing.aliyuncs.com\/Qwen2.5-1M\/Qwen2_5_1M_Technical_Report.pdf<\/p>\n<p><strong>\u4f53\u9a8c\u94fe\u63a5\uff1a<\/strong>https:\/\/modelscope.cn\/studios\/Qwen\/Qwen2.5-1M-Demo<\/p>\n<h1><\/h1>\n<h2>2.\u6a21\u578b\u6027\u80fd<\/h2>\n<p>\u9996\u5148\uff0c\u6765\u770b\u770b Qwen2.5-1M \u7cfb\u5217\u6a21\u578b\u5728\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1\u548c\u77ed\u6587\u672c\u4efb\u52a1\u4e2d\u7684\u6027\u80fd\u8868\u73b0\u3002<\/p>\n<h2><strong>\u00a0<\/strong><\/h2>\n<h3>\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1<\/h3>\n<p>\u5728\u4e0a\u4e0b\u6587\u957f\u5ea6\u4e3a100\u4e07 <a href=\"https:\/\/www.kdjingpai.com\/tokenization\/\">Tokens<\/a> \u7684\u5927\u6d77\u635e\u9488\uff08Passkey Retrieval\uff09\u4efb\u52a1\u4e2d\uff0cQwen2.5-1M \u7cfb\u5217\u6a21\u578b\u80fd\u591f\u51c6\u786e\u5730\u4ece 1M \u957f\u5ea6\u7684\u6587\u6863\u4e2d\u68c0\u7d22\u51fa\u9690\u85cf\u4fe1\u606f\uff0c\u5176\u4e2d\u4ec5\u67097B\u6a21\u578b\u51fa\u73b0\u4e86\u5c11\u91cf\u9519\u8bef\u3002<\/p>\n<p>&nbsp;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/ce69b89fb0ccce6.png\" \/><\/p>\n<p>\u5bf9\u4e8e\u66f4\u590d\u6742\u7684\u957f\u4e0a\u4e0b\u6587\u7406\u89e3\u4efb\u52a1\uff0c\u9009\u62e9\u4e86RULER\u3001LV-Eval \u548c LongbenchChat\u6d4b\u8bd5\u96c6\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/c8a27dc484a41b3.png\" \/><\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/869d6cfcb8832e9.png\" \/><\/p>\n<p>\u4ece\u8fd9\u4e9b\u7ed3\u679c\u4e2d\uff0c\u53ef\u4ee5\u5f97\u51fa\u4ee5\u4e0b\u51e0\u70b9\u5173\u952e\u7ed3\u8bba\uff1a<\/p>\n<ul>\n<li><strong>\u663e\u8457\u8d85\u8d8a128K\u7248\u672c\uff1a<\/strong>Qwen2.5-1M\u7cfb\u5217\u6a21\u578b\u5728\u5927\u591a\u6570\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1\u4e2d\u663e\u8457\u4f18\u4e8e\u4e4b\u524d\u7684128K\u7248\u672c\uff0c\u7279\u522b\u662f\u5728\u5904\u7406\u8d85\u8fc764K\u957f\u5ea6\u7684\u4efb\u52a1\u65f6\u8868\u73b0\u51fa\u8272\u3002<\/li>\n<li><strong>\u6027\u80fd\u4f18\u52bf\u660e\u663e\uff1a<\/strong>Qwen2.5-14B-Instruct-1M \u6a21\u578b\u4e0d\u4ec5\u51fb\u8d25\u4e86 Qwen2.5-Turbo\uff0c\u8fd8\u5728\u591a\u4e2a\u6570\u636e\u96c6\u4e0a\u7a33\u5b9a\u8d85\u8d8a GPT-4o-mini\uff0c\u4e3a\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1\u63d0\u4f9b\u4e86\u5f00\u6e90\u6a21\u578b\u7684\u9009\u62e9\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h3>\u77ed\u5e8f\u5217\u4efb\u52a1<\/h3>\n<p>\u9664\u4e86\u957f\u5e8f\u5217\u4efb\u52a1\u7684\u6027\u80fd\u5916\uff0c\u6a21\u578b\u5728\u77ed\u5e8f\u5217\u4e0a\u7684\u8868\u73b0\u540c\u6837\u5f88\u91cd\u8981\u3002\u5728\u5e7f\u6cdb\u4f7f\u7528\u7684\u5b66\u672f\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u6bd4\u8f83\u4e86 Qwen2.5-1M \u7cfb\u5217\u6a21\u578b\u53ca\u4e4b\u524d\u7684128K\u7248\u672c\uff0c\u5e76\u52a0\u5165\u4e86GPT-4o-mini\u8fdb\u884c\u5bf9\u6bd4\u3002 <img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/080e0bb0a3945e8.png\" \/><\/p>\n<p>\u53ef\u4ee5\u53d1\u73b0\uff1a<\/p>\n<blockquote>\n<ul>\n<li>Qwen2.5-7B-Instruct-1M \u548c Qwen2.5-14B-Instruct-1M \u5728\u77ed\u6587\u672c\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u4e0e\u5176128K\u7248\u672c\u76f8\u5f53\uff0c\u786e\u4fdd\u4e86\u57fa\u672c\u80fd\u529b\u6ca1\u6709\u56e0\u4e3a\u589e\u52a0\u4e86\u957f\u5e8f\u5217\u5904\u7406\u80fd\u529b\u800c\u53d7\u5230\u5f71\u54cd\u3002<\/li>\n<li>\u4e0e GPT-4o-mini \u76f8\u6bd4\uff0cQwen2.5-14B-Instruct-1M \u548c Qwen2.5-Turbo \u5728\u77ed\u6587\u672c\u4efb\u52a1\u4e0a\u5b9e\u73b0\u4e86\u76f8\u8fd1\u7684\u6027\u80fd\uff0c\u540c\u65f6\u4e0a\u4e0b\u6587\u957f\u5ea6\u662fGPT-4o-mini\u7684\u516b\u500d\u3002<\/li>\n<\/ul>\n<\/blockquote>\n<p>&nbsp;<\/p>\n<h2>3.\u5173\u952e\u6280\u672f<\/h2>\n<h3>\u957f\u4e0a\u4e0b\u6587\u8bad\u7ec3<\/h3>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/a64123d44d561b2.png\" \/><\/p>\n<p>\u957f\u5e8f\u5217\u7684\u8bad\u7ec3\u9700\u8981\u5927\u91cf\u7684\u8ba1\u7b97\u8d44\u6e90\uff0c\u56e0\u6b64\u91c7\u7528\u4e86\u9010\u6b65\u6269\u5c55\u957f\u5ea6\u7684\u65b9\u6cd5\uff0c\u5728\u591a\u4e2a\u9636\u6bb5\u5c06Qwen2.5-1M\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u4ece4K\u6269\u5c55\u5230256K\uff1a<\/p>\n<p>\u4ece\u9884\u8bad\u7ec3\u7684Qwen2.5\u7684\u4e00\u4e2a\u4e2d\u95f4Checkpoint\u5f00\u59cb\uff0c\u6b64\u65f6\u4e0a\u4e0b\u6587\u957f\u5ea6\u4e3a4K\u3002<\/p>\n<p><strong>\u5728\u9884\u8bad\u7ec3\u9636\u6bb5<\/strong>\uff0c\u9010\u6b65\u5c06\u4e0a\u4e0b\u6587\u957f\u5ea6\u4ece 4K \u589e\u52a0\u5230 256K\uff0c\u540c\u65f6\u4f7f\u7528Adjusted Base Frequency\u7684\u65b9\u6848\uff0c\u5c06 RoPE \u57fa\u7840\u9891\u7387\u4ece 10,000 \u63d0\u9ad8\u5230 10,000,000\u3002<\/p>\n<p><strong>\u5728\u76d1\u7763\u5fae\u8c03\u9636\u6bb5<\/strong>\uff0c\u5206\u4e24\u4e2a\u9636\u6bb5\u8fdb\u884c\u4ee5\u4fdd\u6301\u77ed\u5e8f\u5217\u4e0a\u7684\u6027\u80fd\uff1a<\/p>\n<p><strong>\u00a0 \u00a0\u7b2c\u4e00\u9636\u6bb5\uff1a<\/strong>\u00a0\u4ec5\u5728\u77ed\u6307\u4ee4\uff08\u6700\u591a32K\u957f\u5ea6\uff09\u4e0a\u8fdb\u884c\u5fae\u8c03\uff0c\u8fd9\u91cc\u4f7f\u7528\u4e0e Qwen2.5 \u7684 128K \u7248\u672c\u76f8\u540c\u7684\u6570\u636e\u548c\u6b65\u9aa4\u6570\u3002<\/p>\n<p><strong>\u00a0 \u00a0\u7b2c\u4e8c\u9636\u6bb5\uff1a<\/strong>\u00a0\u6df7\u5408\u77ed\u6307\u4ee4\uff08\u6700\u591a32K\uff09\u548c\u957f\u6307\u4ee4\uff08\u6700\u591a256K\uff09\uff0c\u5b9e\u73b0\u5728\u589e\u5f3a\u957f\u4efb\u52a1\u7684\u6027\u80fd\u7684\u540c\u65f6\uff0c\u4fdd\u6301\u77ed\u4efb\u52a1\u7684\u8d28\u91cf\u3002<\/p>\n<p><strong>\u5728\u5f3a\u5316\u5b66\u4e60\u9636\u6bb5<\/strong>\uff0c\u5728\u77ed\u6587\u672c\uff08\u6700\u591a8K\u6807\u8bb0\uff09\u4e0a\u8bad\u7ec3\u6a21\u578b\u3002\u6211\u4eec\u53d1\u73b0\uff0c\u5373\u4f7f\u5728\u77ed\u6587\u672c\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u4eba\u7c7b\u504f\u597d\u7684\u5bf9\u9f50\u7684\u63d0\u5347\u80fd\u591f\u5f88\u597d\u5730\u6cdb\u5316\u5230\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1\u4e2d\u3002\u00a0\u901a\u8fc7\u4ee5\u4e0a\u8bad\u7ec3\uff0c\u6211\u4eec\u6700\u7ec8\u83b7\u5f97\u7684 Instruct \u6a21\u578b\u80fd\u591f\u5904\u7406\u957f\u8fbe256K Tokens\u7684\u5e8f\u5217\u3002<\/p>\n<p>\u901a\u8fc7\u4ee5\u4e0a\u8bad\u7ec3\uff0c\u83b7\u5f97\u4e86256K\u4e0a\u4e0b\u6587\u957f\u5ea6\u7684\u6307\u4ee4\u5fae\u8c03\u6a21\u578b\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u957f\u5ea6\u5916\u63a8<\/h3>\n<p>\u5728\u4e0a\u8ff0\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u6a21\u578b\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u4ec5\u4e3a 256K \u4e2a Token\u3002\u4e3a\u4e86\u5c06\u5176\u6269\u5c55\u5230 1M Tokens\uff0c\u91c7\u7528\u4e86\u957f\u5ea6\u5916\u63a8\u7684\u6280\u672f\u3002<\/p>\n<p>\u5f53\u524d\uff0c\u57fa\u4e8e\u65cb\u8f6c\u4f4d\u7f6e\u7f16\u7801\u7684\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1\u4e2d\u4f1a\u4ea7\u751f\u6027\u80fd\u4e0b\u964d\uff0c\u8fd9\u4e3b\u8981\u662f\u7531\u4e8e\u5728\u8ba1\u7b97\u6ce8\u610f\u529b\u6743\u91cd\u65f6\uff0cQuery \u548c Key \u4e4b\u95f4\u7684\u76f8\u5bf9\u4f4d\u7f6e\u8ddd\u79bb\u8fc7\u5927\uff0c\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u672a\u66fe\u89c1\u8fc7\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0cQwen2.5-1M\u91c7\u7528Dual Chunk Attention(DCA)\u7684\u65b9\u6cd5\uff0c\u8be5\u65b9\u6cd5\u901a\u8fc7\u5c06\u8fc7\u5927\u7684\u76f8\u5bf9\u4f4d\u7f6e\uff0c\u91cd\u65b0\u6620\u5c04\u4e3a\u8f83\u5c0f\u7684\u503c\uff0c\u4ece\u800c\u89e3\u51b3\u4e86\u8fd9\u4e00\u96be\u9898\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/8ef1c798dc6ddf3.png\" \/><\/p>\n<p>\u5bf9 Qwen2.5-1M \u6a21\u578b\u53ca\u4e4b\u524d 128K \u7684\u7248\u672c\u8fdb\u884c\u4e86\u8bc4\u4f30\uff0c\u5206\u522b\u6d4b\u8bd5\u4e86\u4f7f\u7528\u548c\u4e0d\u4f7f\u7528\u957f\u5ea6\u5916\u63a8\u65b9\u6cd5\u7684\u60c5\u51b5\u3002<\/p>\n<p>\u7ed3\u679c\u8868\u660e\uff1a\u5373\u4f7f\u662f\u4ec5\u572832K\u6807\u8bb0\u4e0a\u8bad\u7ec3\u7684\u6a21\u578b\uff0c\u5982Qwen2.5-7B-Instruct\uff0c\u5728\u5904\u74061M\u6807\u8bb0\u4e0a\u4e0b\u6587\u7684 Passkey <a href=\"https:\/\/www.kdjingpai.com\/retrieval\/\">Retrieval<\/a> \u4efb\u52a1\u4e2d\u4e5f\u80fd\u8fbe\u5230\u8fd1\u4e4e\u5b8c\u7f8e\u7684\u51c6\u786e\u7387\u3002\u8fd9\u5145\u5206\u5c55\u793a\u4e86 DCA \u5728\u65e0\u9700\u989d\u5916\u8bad\u7ec3\u7684\u60c5\u51b5\u4e0b\uff0c\u5373\u53ef\u663e\u8457\u6269\u5c55\u652f\u6301\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u7684\u5f3a\u5927\u80fd\u529b\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/405c5eae3d3abfa.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<h3>\u7a00\u758f\u6ce8\u610f\u529b\u673a\u5236<\/h3>\n<p>\u5bf9\u4e8e\u957f\u4e0a\u4e0b\u6587\u8bed\u8a00\u6a21\u578b\uff0c\u63a8\u7406\u901f\u5ea6\u5bf9\u4e8e\u7528\u6237\u4f53\u9a8c\u81f3\u5173\u91cd\u8981\u3002\u4e3a\u4e86\u52a0\u901f\u9884\u586b\u5145\u9636\u6bb5\uff0c\u7814\u7a76\u56e2\u961f\u5f15\u5165\u4e86\u57fa\u4e8e MInference \u7684\u7a00\u758f\u6ce8\u610f\u673a\u5236\u3002\u6b64\u5916\uff0c\u63d0\u51fa\u4e86\u51e0\u9879\u6539\u8fdb\uff1a<\/p>\n<ul>\n<li>\u5206\u5757\u9884\u586b\u5145\uff1a\u00a0\u5982\u679c\u76f4\u63a5\u4f7f\u7528\u6a21\u578b\u5904\u7406\u957f\u5ea6100\u4e07\u7684\u5e8f\u5217\uff0c\u5176\u4e2d MLP \u5c42\u7684\u6fc0\u6d3b\u6743\u91cd\u4f1a\u4ea7\u751f\u5de8\u5927\u7684\u663e\u5b58\u5f00\u9500\u3002\u4ee5Qwen2.5-7B \u4e3a\u4f8b\uff0c\u8fd9\u90e8\u5206\u5f00\u9500\u9ad8\u8fbe 71GB\u3002\u901a\u8fc7\u5c06\u5206\u5757\u9884\u586b\u5145\uff08Chunked Prefill\uff09\u4e0e\u7a00\u758f\u6ce8\u610f\u529b\u9002\u914d\uff0c\u53ef\u4ee5\u5c06\u8f93\u5165\u5e8f\u5217\u4ee5 32768 \u957f\u5ea6\u5206\u5757\uff0c\u9010\u5757\u8fdb\u884c\u9884\u586b\u5145\uff0cMLP \u5c42\u6fc0\u6d3b\u6743\u91cd\u7684\u663e\u5b58\u4f7f\u7528\u91cf\u53ef\u51cf\u5c11 96.7%\uff0c\u56e0\u800c\u663e\u8457\u964d\u4f4e\u4e86\u8bbe\u5907\u7684\u663e\u5b58\u9700\u6c42\u3002<\/li>\n<li>\u96c6\u6210\u957f\u5ea6\u5916\u63a8\u65b9\u6848\uff1a\u00a0\u6211\u4eec\u5728\u7a00\u758f\u6ce8\u610f\u529b\u673a\u5236\u4e2d\u8fdb\u4e00\u6b65\u96c6\u6210\u4e86\u57fa\u4e8e DCA \u7684\u957f\u5ea6\u5916\u63a8\u65b9\u6848\uff0c\u8fd9\u4f7f\u6211\u4eec\u7684\u63a8\u7406\u6846\u67b6\u80fd\u591f\u540c\u65f6\u4eab\u53d7\u66f4\u9ad8\u7684\u63a8\u7406\u6548\u7387\u548c\u957f\u5e8f\u5217\u4efb\u52a1\u7684\u51c6\u786e\u6027\u3002<\/li>\n<li>\u7a00\u758f\u6027\u4f18\u5316\uff1a\u00a0\u539f\u59cb\u7684 MInference \u65b9\u6cd5\u9700\u8981\u8fdb\u884c\u79bb\u7ebf\u641c\u7d22\u4ee5\u786e\u5b9a\u6bcf\u4e2a\u6ce8\u610f\u529b\u5934\u7684\u6700\u4f73\u7a00\u758f\u5316\u914d\u7f6e\u3002\u7531\u4e8e\u5168\u6ce8\u610f\u529b\u6743\u91cd\u5bf9\u5185\u5b58\u7684\u8981\u6c42\u592a\u5927\uff0c\u8fd9\u79cd\u641c\u7d22\u901a\u5e38\u5728\u77ed\u5e8f\u5217\u4e0a\u8fdb\u884c\uff0c\u4e0d\u4e00\u5b9a\u80fd\u5728\u66f4\u957f\u5e8f\u5217\u4e0b\u8d77\u5230\u5f88\u597d\u7684\u6548\u679c\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u80fd\u591f\u5728100\u4e07\u957f\u5ea6\u7684\u5e8f\u5217\u4e0a\u4f18\u5316\u7a00\u758f\u5316\u914d\u7f6e\u7684\u65b9\u6cd5\uff0c\u4ece\u800c\u663e\u8457\u51cf\u5c11\u4e86\u7a00\u758f\u6ce8\u610f\u529b\u5e26\u6765\u7684\u7cbe\u5ea6\u635f\u5931\u3002<\/li>\n<li>\u5176\u4ed6\u4f18\u5316\uff1a\u00a0\u6211\u4eec\u8fd8\u5f15\u5165\u4e86\u5176\u4ed6\u4f18\u5316\u63aa\u65bd\uff0c\u5982\u4f18\u5316\u7b97\u5b50\u6548\u7387\u548c\u52a8\u6001\u5206\u5757\u6d41\u6c34\u7ebf\u5e76\u884c\uff0c\u4ee5\u5145\u5206\u53d1\u6325\u6574\u4e2a\u6846\u67b6\u7684\u6f5c\u529b\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<p>\u901a\u8fc7\u8fd9\u4e9b\u6539\u8fdb\uff0c\u63a8\u7406\u6846\u67b6\u5728\u4e0d\u540c\u6a21\u578b\u5927\u5c0f\u548c GPU \u8bbe\u5907\u4e0a\u5c06 1M \u4e2a <a href=\"https:\/\/www.kdjingpai.com\/tokenization\/\">token<\/a> \u957f\u5ea6\u7684\u5e8f\u5217\u7684\u9884\u586b\u5145\u901f\u5ea6\u63d0\u9ad8\u4e86 3.2 \u500d\u5230 6.7 \u500d\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/365b79cfed62ff5.png\" \/><\/p>\n<h1><\/h1>\n<h2>4.\u6a21\u578b\u90e8\u7f72<\/h2>\n<h3>\u7cfb\u7edf\u51c6\u5907<\/h3>\n<p>\u4e3a\u4e86\u83b7\u5f97\u6700\u4f73\u6027\u80fd\uff0c\u5efa\u8bae\u4f7f\u7528\u652f\u6301\u4f18\u5316\u5185\u6838\u7684 Ampere \u6216 Hopper \u67b6\u6784\u7684 GPU\u3002<\/p>\n<p>\u8bf7\u786e\u4fdd\u6ee1\u8db3\u4ee5\u4e0b\u8981\u6c42\uff1a<\/p>\n<ul>\n<li>CUDA \u7248\u672c\uff1a12.1 \u6216 12.3<\/li>\n<li>Python \u7248\u672c\uff1a&gt;=3.9 \u4e14 &lt;=3.12<\/li>\n<\/ul>\n<p>\u663e\u5b58\u8981\u6c42\uff0c\u5bf9\u4e8e\u5904\u7406 1M \u957f\u5ea6\u7684\u5e8f\u5217\uff1a<\/p>\n<ul>\n<li>Qwen2.5-7B-Instruct-1M\uff1a\u81f3\u5c11\u9700\u8981 120GB \u663e\u5b58\uff08\u591a GPU \u603b\u548c\uff09\u3002<\/li>\n<li>Qwen2.5-14B-Instruct-1M\uff1a\u81f3\u5c11\u9700\u8981 320GB \u663e\u5b58\uff08\u591a GPU \u603b\u548c\uff09\u3002<\/li>\n<\/ul>\n<p>\u5982\u679c GPU \u663e\u5b58\u4e0d\u6ee1\u8db3\u4ee5\u4e0a\u8981\u6c42\uff0c\u4f60\u4ecd\u7136\u53ef\u4ee5\u4f7f\u7528 Qwen2.5-1M \u8fdb\u884c\u8f83\u77ed\u4efb\u52a1\u7684\u5904\u7406\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u5b89\u88c5\u4f9d\u8d56\u9879<\/h3>\n<p>\u6682\u65f6\uff0c\u4f60\u9700\u8981\u4ece\u81ea\u5b9a\u4e49\u5206\u652f\u514b\u9686 vLLM \u4ed3\u5e93\uff0c\u5e76\u624b\u52a8\u5b89\u88c5\u3002\u7814\u7a76\u56e2\u961f\u6b63\u5728\u52aa\u529b\u5c06\u8be5\u5206\u652f\u63d0\u4ea4\u5230 vLLM \u9879\u76ee\u4e2d\u3002<\/p>\n<pre>git clone -b dev\/dual-chunk-attn git@github.com:QwenLM\/vllm.git\r\ncd vllm\r\npip install -e . -v<\/pre>\n<h3><\/h3>\n<h3>\u542f\u52a8 OpenAI \u517c\u5bb9\u7684 API \u670d\u52a1<\/h3>\n<p>\u6307\u5b9a\u6a21\u578b\u4eceModelScope\u4e0b\u8f7d<\/p>\n<pre>export\u00a0VLLM_USE_MODELSCOPE=True<\/pre>\n<p>\u53d1\u5e03OpenAI\u517c\u5bb9\u7684API\u670d\u52a1<\/p>\n<pre>vllm serve Qwen\/Qwen2.5-7B-Instruct-1M \\\r\n  --tensor-parallel-size 4 \\\r\n  --max-model-len 1010000 \\\r\n  --enable-chunked-prefill\r\n --max-num-batched-tokens 131072 \\\r\n  --enforce-eager \\\r\n  --max-num-seqs 1<\/pre>\n<p><strong>\u53c2\u6570\u8bf4\u660e\uff1a<\/strong><\/p>\n<ul>\n<li style=\"list-style-type: none;\">\n<ul>\n<li><code>--tensor-parallel-size<\/code>\n<ul>\n<li>\u8bbe\u7f6e\u4e3a\u60a8\u4f7f\u7528\u7684 GPU \u6570\u91cf\u30027B \u6a21\u578b\u6700\u591a\u652f\u6301 4 \u4e2a GPU\uff0c14B \u6a21\u578b\u6700\u591a\u652f\u6301 8 \u4e2a GPU\u3002<\/li>\n<\/ul>\n<\/li>\n<li><code>--max-model-len<\/code>\n<ul>\n<li>\u5b9a\u4e49\u6700\u5927\u8f93\u5165\u5e8f\u5217\u957f\u5ea6\u3002\u5982\u679c\u9047\u5230\u5185\u5b58\u4e0d\u8db3\u95ee\u9898\uff0c\u8bf7\u51cf\u5c11\u6b64\u503c\u3002<\/li>\n<\/ul>\n<\/li>\n<li><code>--max-num-batched-tokens<\/code>\n<ul>\n<li>\u8bbe\u7f6e Chunked Prefill \u7684\u5757\u5927\u5c0f\u3002\u8f83\u5c0f\u7684\u503c\u53ef\u4ee5\u51cf\u5c11\u6fc0\u6d3b\u5185\u5b58\u4f7f\u7528\uff0c\u4f46\u53ef\u80fd\u4f1a\u51cf\u6162\u63a8\u7406\u901f\u5ea6\u3002<\/li>\n<li>\u63a8\u8350\u503c\u4e3a 131072\uff0c\u4ee5\u83b7\u5f97\u6700\u4f73\u6027\u80fd\u3002<\/li>\n<\/ul>\n<\/li>\n<li><code>--max-num-seqs<\/code>\n<ul>\n<li>\u9650\u5236\u5e76\u53d1\u5904\u7406\u7684\u5e8f\u5217\u6570\u91cf\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h3>\u4e0e\u6a21\u578b\u4ea4\u4e92<\/h3>\n<p>\u53ef\u4ee5\u4f7f\u7528\u4ee5\u4e0b\u65b9\u6cd5\u4e0e\u90e8\u7f72\u7684\u6a21\u578b\u8fdb\u884c\u4ea4\u4e92\uff1a<\/p>\n<p><strong>\u9009\u9879 1. \u4f7f\u7528 Curl<\/strong><\/p>\n<pre>curl http:\/\/localhost:8000\/v1\/chat\/completions \\\r\n-H \"Content-Type: application\/json\" \\\r\n-d '{\r\n  \"model\": \"Qwen\/Qwen2.5-7B-Instruct-1M\",\r\n  \"messages\": [\r\n    {\r\n      \"role\": \"user\",\r\n      \"content\": \"Tell me something about large language models.\"\r\n    }\r\n  ],\r\n  \"temperature\": 0.7,\r\n  \"top_p\": 0.8,\r\n  \"repetition_penalty\": 1.05,\r\n  \"max_tokens\": 512\r\n}'\r\n<\/pre>\n<p>&nbsp;<\/p>\n<p><strong>\u9009\u9879 2. \u4f7f\u7528 Python<\/strong><\/p>\n<pre>from openai import OpenAI\r\n\r\nopenai_api_key = \"EMPTY\"\r\nopenai_api_base = \"http:\/\/localhost:8000\/v1\"\r\n\r\nclient = OpenAI(\r\n    api_key=openai_api_key,\r\n    base_url=openai_api_base,\r\n)\r\n\r\nprompt = (\r\n    \"\"\"There is an important info hidden inside a lot of irrelevant text. \r\n    Find it and memorize them. I will quiz you about the important information there.\\n\\n\r\n    The pass key is 28884. Remember it. 28884 is the pass key.\\n\"\"\" \r\n    + \"The grass is green. The sky is blue. The sun is yellow. Here we go. There and back again. \" * 800 \r\n    + \"\\nWhat is the pass key?\"\r\n)\r\n\r\n# The prompt is 20k long. You can try a longer prompt by replacing 800 with 40000.\r\nchat_response = client.chat.completions.create(\r\n    model=\"Qwen\/Qwen2.5-7B-Instruct-1M\",\r\n    messages=[\r\n        {\"role\": \"user\", \"content\": prompt},\r\n    ],\r\n    temperature=0.7,\r\n    top_p=0.8,\r\n    max_tokens=512,\r\n    extra_body={\r\n        \"repetition_penalty\": 1.05,\r\n    },\r\n)\r\n\r\nprint(\"Chat response:\", chat_response)\r\n<\/pre>\n<p>\u60a8\u8fd8\u53ef\u4ee5\u63a2\u7d22\u5176\u4ed6\u6846\u67b6\uff0c\u4f8b\u5982Qwen-Agent\uff0c\u4ee5\u4f7f\u6a21\u578b\u80fd\u591f\u8bfb\u53d6 PDF \u6587\u4ef6\u7b49\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>5.\u4f7f\u7528\u9b54\u642dAPI-Inference\u76f4\u63a5\u8c03\u7528<\/h2>\n<p>\u9b54\u642d\u5e73\u53f0\u7684API-Inference\uff0c\u4e5f\u7b2c\u4e00\u65f6\u95f4\u4e3aQwen2.5-7B-Instruct-1M\u548cQwen2.5-14B-Instruct-1M\u6a21\u578b\u63d0\u4f9b\u4e86\u652f\u6301\u3002\u9b54\u642d\u7684\u7528\u6237\u53ef\u901a\u8fc7API\u8c03\u7528\u7684\u65b9\u5f0f\uff0c\u76f4\u63a5\u4f7f\u7528\u8be5\u6a21\u578b\u3002\u5177\u4f53API-Inference\u7684\u4f7f\u7528\u65b9\u5f0f\u53ef\u53c2\u89c1\u6a21\u578b\u9875\u9762\uff08\u4f8b\u5982 https:\/\/modelscope.cn\/models\/Qwen\/Qwen2.5-14B-Instruct-1M \uff09\u8bf4\u660e\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/0219aa93c2f30be.png\" \/><\/p>\n<p>\u6216\u8005\u53c2\u89c1API-Inference\u6587\u6863\uff1ahttps:\/\/www.modelscope.cn\/docs\/model-service\/API-Inference\/intro<\/p>\n<p>&nbsp;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/2fc60d3db02e58c.png\" \/><\/p>\n<p>\u611f\u8c22\u963f\u91cc\u4e91\u767e\u70bc\u5e73\u53f0\u63d0\u4f9b\u80cc\u540e\u7b97\u529b\u652f\u6301\u3002<\/p>\n<h2><\/h2>\n<h3><strong>\u4f7f\u7528Ollama\u548cllamafile<\/strong><\/h3>\n<p>\u4e3a\u4e86\u65b9\u4fbf\u5927\u5bb6\u5728\u672c\u5730\u4f7f\u7528\uff0c\u9b54\u642d\u5728\u7b2c\u4e00\u65f6\u95f4\u63d0\u4f9b\u4e86Qwen2.5-7B-Instruct-1M\u6a21\u578b\u7684GGUF\u7248\u672c\u548cllamafile\u7248\u672c\u3002\u53ef\u901a\u8fc7Ollama\u6846\u67b6\u8c03\u7528\uff0c\u6216\u8005\u76f4\u63a5\u4f7f\u7528\u62c9\u8d77llamafile\u3002<\/p>\n<p><strong>1.Ollama\u8c03\u7528<\/strong><\/p>\n<p>\u9996\u5148\u8bbe\u7f6eollama\u4e0b\u542f\u7528\uff1a<\/p>\n<pre><code><a href=\"https:\/\/www.kdjingpai.com\/ollama\/\">ollama<\/a> serve<\/code><\/pre>\n<p>&nbsp;<\/p>\n<p>\u7136\u540e\u5c31\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528 ollama run \u7684\u547d\u4ee4\u8fd0\u884c\u9b54\u642d\u4e0a\u7684GGUF\u6a21\u578b\uff1a<\/p>\n<pre><code>ollama\u00a0run\u00a0modelscope.cn\/modelscope\/Qwen2.5-7B-Instruct-1M-GGUF<\/code><\/pre>\n<p>\u8fd0\u884c\u7ed3\u679c\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/01e27254ebff761.png\" \/><\/p>\n<h3><\/h3>\n<p><strong>2.llamafile\u6a21\u578b\u76f4\u63a5\u62c9\u8d77<\/strong><\/p>\n<p><a href=\"https:\/\/www.kdjingpai.com\/llamafilejianhuall\/\">Llamafile<\/a> \u63d0\u4f9b\u4e86\u5927\u6a21\u578b\u548c\u8fd0\u884c\u73af\u5883\u5168\u90e8\u5c01\u88c5\u5728\u4e00\u4e2a\u53ef\u6267\u884c\u6587\u4ef6\u4e2d\u7684\u65b9\u6848\u3002\u901a\u8fc7\u9b54\u642d\u547d\u4ee4\u884c\u548cllamafile\u7684\u96c6\u6210\uff0c\u53ef\u4ee5\u771f\u6b63\u5b9e\u73b0\u5728Linux\/Mac\/Windows\u7b49\u4e0d\u540c\u64cd\u4f5c\u7cfb\u7edf\u73af\u5883\u4e2d\uff0c\u4e00\u952e\u8fd0\u884c\u5927\u6a21\u578b\uff1a<\/p>\n<pre><code>modelscope\u00a0<a href=\"https:\/\/www.kdjingpai.com\/llamafilejianhuall\/\">llamafile<\/a>\u00a0--model\u00a0Qwen-Llamafile\/Qwen2.5-7B-Instruct-1M-llamafile<\/code><\/pre>\n<p>\u8fd0\u884c\u7ed3\u679c\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/a140652b0b6c721.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u66f4\u591a\u6587\u6863\u53ef\u53c2\u89c1 https:\/\/www.modelscope.cn\/docs\/models\/advanced-usage\/llamafile<\/p>\n<h1><\/h1>\n<h2>6.\u6a21\u578b\u5fae\u8c03<\/h2>\n<p>\u8fd9\u91cc\u6211\u4eec\u4ecb\u7ecd\u4f7f\u7528ms-swift\u5bf9Qwen\/Qwen2.5-7B-Instruct-1M\u8fdb\u884c\u5fae\u8c03\u3002<\/p>\n<p>\u5728\u5f00\u59cb\u5fae\u8c03\u4e4b\u524d\uff0c\u8bf7\u786e\u4fdd\u60a8\u7684\u73af\u5883\u5df2\u6b63\u786e\u5b89\u88c5\uff1a<\/p>\n<pre># \u5b89\u88c5ms-swift\r\ngit clone https:\/\/github.com\/modelscope\/ms-swift.git\r\ncd ms-swift\r\npip install -e .<\/pre>\n<p>&nbsp;<\/p>\n<p>\u6211\u4eec\u7ed9\u51fa\u53ef\u8fd0\u884c\u7684\u5fae\u8c03demo\u548c\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u7684\u6837\u5f0f\uff0c\u5fae\u8c03\u811a\u672c\u5982\u4e0b\uff1a<\/p>\n<p>CUDA_VISIBLE_DEVICES=0<\/p>\n<pre>swift sft \\\r\n    --model Qwen\/Qwen2.5-7B-Instruct-1M \\\r\n    --train_type lora \\\r\n    --dataset 'AI-ModelScope\/alpaca-gpt4-data-zh#500' \\\r\n             'AI-ModelScope\/alpaca-gpt4-data-en#500' \\\r\n             'swift\/self-cognition#500' \\\r\n    --torch_dtype bfloat16 \\\r\n    --num_train_epochs 1 \\\r\n    --per_device_train_batch_size 1 \\\r\n    --per_device_eval_batch_size 1 \\\r\n    --learning_rate 1e-4 \\\r\n    --lora_rank 8 \\\r\n    --lora_alpha 32 \\\r\n    --target_modules all-linear \\\r\n    --gradient_accumulation_steps 16 \\\r\n    --eval_steps 50 \\\r\n    --save_steps 50 \\\r\n    --save_total_limit 5 \\\r\n    --logging_steps 5 \\\r\n    --max_length 2048 \\\r\n    --output_dir output \\\r\n    --warmup_ratio 0.05 \\\r\n    --dataloader_num_workers 4 \\\r\n    --model_author swift \\\r\n    --model_name swift-robot\r\n<\/pre>\n<p>&nbsp;<\/p>\n<p>\u8bad\u7ec3\u663e\u5b58\u5360\u7528\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/e88e1bfb59297e3.png\" \/><\/p>\n<p>&nbsp;<\/p>\n<p>\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u683c\u5f0f\uff1a\uff08\u76f4\u63a5\u4f7f\u7528`&#8211;dataset &lt;dataset_path&gt;`\u6307\u5b9a\u5373\u53ef\uff09<\/p>\n<pre><code>{\"messages\":\u00a0[{\"role\":\u00a0\"user\",\u00a0\"content\":\u00a0\"&lt;query&gt;\"},\u00a0{\"role\":\u00a0\"assistant\",\u00a0\"content\":\u00a0\"&lt;response&gt;\"},\u00a0{\"role\":\u00a0\"user\",\u00a0\"content\":\u00a0\"&lt;query2&gt;\"},\u00a0{\"role\":\u00a0\"assistant\",\u00a0\"content\":\u00a0\"&lt;response2&gt;\"}]}<\/code><\/pre>\n<p>&nbsp;<\/p>\n<p>\u63a8\u7406\u811a\u672c\uff1a<\/p>\n<pre>CUDA_VISIBLE_DEVICES=0 \\\r\nswift infer \\\r\n    --adapters output\/vx-xxx\/checkpoint-xxx \\\r\n    --stream true \\\r\n    --max_new_tokens 2048<\/pre>\n<p>&nbsp;<\/p>\n<p>\u63a8\u9001\u6a21\u578b\u5230ModelScope\uff1a<\/p>\n<pre>CUDA_VISIBLE_DEVICES=0 swift export \\\r\n    --adapters output\/vx-xxx\/checkpoint-xxx \\\r\n    --push_to_hub true \\\r\n    --hub_model_id '' \\\r\n    --hub_token ''<\/pre>\n<h1><\/h1>\n<h2>7.\u4e0b\u4e00\u6b65\u662f\u4ec0\u4e48\uff1f<\/h2>\n<p>\u867d\u7136 Qwen2.5-1M \u7cfb\u5217\u4e3a\u957f\u5e8f\u5217\u5904\u7406\u4efb\u52a1\u5e26\u6765\u4e86\u4f18\u79c0\u7684\u5f00\u6e90\u9009\u62e9\uff0c\u7814\u7a76\u56e2\u961f\u4e5f\u5145\u5206\u8ba4\u8bc6\u5230\u957f\u4e0a\u4e0b\u6587\u6a21\u578b\u4ecd\u6709\u5f88\u5927\u7684\u63d0\u5347\u7a7a\u95f4\u3002\u6211\u4eec\u7684\u76ee\u6807\u662f\u6253\u9020\u5728\u957f\u77ed\u4efb\u52a1\u4e2d\u5747\u80fd\u8868\u73b0\u5353\u8d8a\u7684\u6a21\u578b\uff0c\u786e\u4fdd\u5b83\u4eec\u5728\u5b9e\u9645\u5e94\u7528\u573a\u666f\u4e2d\u771f\u6b63\u53d1\u6325\u4f5c\u7528\u3002\u4e3a\u6b64\uff0c\u7814\u7a76\u56e2\u961f\u6b63\u6df1\u5165\u7814\u7a76\u66f4\u9ad8\u6548\u7684\u8bad\u7ec3\u65b9\u5f0f\u3001\u6a21\u578b\u67b6\u6784\u548c\u63a8\u7406\u65b9\u6cd5\uff0c\u529b\u6c42\u4f7f\u8fd9\u4e9b\u6a21\u578b\u5373\u4f7f\u5728\u8d44\u6e90\u6709\u9650\u7684\u73af\u5883\u4e2d\u4e5f\u80fd\u9ad8\u6548\u90e8\u7f72\u5e76\u4e14\u83b7\u5f97\u6700\u4f73\u7684\u6027\u80fd\u6548\u679c\u3002\u00a0\u7814\u7a76\u56e2\u961f\u575a\u4fe1\uff0c\u8fd9\u4e9b\u52aa\u529b\u5c06\u4e3a\u957f\u4e0a\u4e0b\u6587\u6a21\u578b\u5f00\u542f\u5168\u65b0\u7684\u53ef\u80fd\u6027\uff0c\u5927\u5e45\u62d3\u5c55\u5176\u5e94\u7528\u8303\u56f4\uff0c\u5e76\u5c06\u6301\u7eed\u7a81\u7834\u8fd9\u4e00\u9886\u57df\u7684\u8fb9\u754c\uff0c\u656c\u8bf7\u671f\u5f85\uff01<\/p>\n","protected":false},"excerpt":{"rendered":"<p>1.\u5f15\u8a00 \u4e24\u4e2a\u6708\u524d\uff0cQwen\u56e2\u961f\u5347\u7ea7\u4e86 Qwen2.5-Turbo\uff0c\u4f7f\u5176\u652f\u6301\u6700\u591a\u4e00\u767e\u4e07\u4e2aTokens\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u3002\u4eca\u5929\uff0cQwen\u6b63\u5f0f\u63a8\u51fa\u5f00\u6e90\u7684 Qwen2.5-1M \u6a21\u578b\u53ca\u5176\u5bf9\u5e94\u7684\u63a8\u7406\u6846\u67b6\u652f\u6301\u3002\u4ee5\u4e0b\u662f\u672c\u6b21\u53d1\u5e03\u7684\u4eae\u70b9\uff1a \u5f00\u6e90\u6a21\u578b\uff1a\u00a0\u672c\u6b21\u53d1\u5e03\u4e86&#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-19453","post","type-post","status-publish","format-standard","hentry","category-news"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/19453","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/comments?post=19453"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/19453\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media?parent=19453"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/categories?post=19453"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/tags?post=19453"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}