{"id":34979,"date":"2026-08-10T14:59:49","date_gmt":"2026-08-10T06:59:49","guid":{"rendered":"https:\/\/www.varidata.com\/uncategorized-zh-cn\/how-to-configure-task-priorities-for-ai-training-jobs\/"},"modified":"2026-08-10T15:11:08","modified_gmt":"2026-08-10T07:11:08","slug":"how-to-configure-task-priorities-for-ai-training-jobs","status":"publish","type":"post","link":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/","title":{"rendered":"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7"},"content":{"rendered":"<p><a href=\"https:\/\/www.varidata.com\/zh-cn\/server\/los-angeles\/\" target=\"_blank\">AI \u57fa\u7840\u8bbe\u65bd<\/a>\u5de5\u7a0b\u5e08\u5fc5\u987b\u5728\u6ee1\u8db3\u5b9e\u65f6\u6267\u884c\u9700\u6c42\u7684\u540c\u65f6\u63d0\u5347\u6574\u4f53\u8fd0\u884c\u6548\u7387\u3002\u4f60\u53ef\u4ee5\u901a\u8fc7\u8c03\u6574\u5e95\u5c42\u7cfb\u7edf\u7684 CPU \u548c GPU \u8bbe\u7f6e\u3001\u5efa\u7acb\u4f18\u5148\u7ea7\u7c7b\u4ee5\u53ca\u5b9a\u4e49\u8c03\u5ea6\u5668\u961f\u5217\u89c4\u5219\uff0c\u6765\u4e3a <a href=\"https:\/\/www.varidata.com\/zh-cn\/blog\/esign-operate-ai-training-clusters-with-us-hosting\/\" target=\"_blank\">AI \u8bad\u7ec3<\/a>\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\u3002<\/p>\n<p>\u6070\u5f53\u7684 AI \u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u4ee5\u9632\u6b62\u591a\u8282\u70b9\u96c6\u7fa4\u4e2d\u7684 GPU \u8d44\u6e90\u9965\u997f\u3002\u5f53\u5de5\u4f5c\u8d1f\u8f7d\u5f00\u59cb\u6392\u961f\u65f6\uff0c\u6709\u6548\u7684\u4efb\u52a1\u7ba1\u7406\u4f1a\u5e94\u7528\u660e\u786e\u7684\u7b56\u7565\uff0c\u4ee5\u6700\u5927\u5316\u96c6\u7fa4\u751f\u4ea7\u529b\u548c\u7b97\u529b\u4ea7\u51fa\u3002\u8c03\u5ea6\u5668\u4f1a\u901a\u8fc7\u8bc4\u4f30\u4ee5\u4e0b\u5177\u4f53\u6807\u51c6\u6765\u51b3\u5b9a\u4efb\u52a1\u6267\u884c\u987a\u5e8f\uff1a<\/p>\n<ul>\n<li>\n<p>\u5de5\u4f5c\u8d1f\u8f7d\u4f18\u5148\u7ea7<\/p>\n<\/li>\n<li>\n<p>\u63d0\u4ea4\u65f6\u95f4\uff08FIFO\uff0c\u5148\u8fdb\u5148\u51fa\uff09<\/p>\n<\/li>\n<li>\n<p>\u670d\u52a1\u7ea7\u522b\u534f\u8bae\uff08SLA\uff09<\/p>\n<\/li>\n<li>\n<p>\u56e2\u961f\u4e4b\u95f4\u7684\u516c\u5e73\u5171\u4eab\u8c03\u5ea6<\/p>\n<\/li>\n<li>\n<p>\u8d44\u6e90\u53ef\u7528\u6027<\/p>\n<\/li>\n<\/ul>\n<p>\u91c7\u7528 AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u4ee5\u4f18\u5316\u73b0\u4ee3 AI \u7cfb\u7edf\u3002\u9ad8\u4f18\u5148\u7ea7 AI \u4efb\u52a1\u4f1a\u7acb\u5373\u83b7\u5f97\u786c\u4ef6\u8bbf\u95ee\u6743\uff0c\u800c\u4f4e\u4f18\u5148\u7ea7\u8bad\u7ec3\u4efb\u52a1\u5219\u4f1a\u4ee5\u5e73\u6ed1\u7684\u65b9\u5f0f\u8ba9\u51fa\u7b97\u529b\u8d44\u6e90\u3002<\/p>\n<h2>\u4e3a AI \u786c\u4ef6\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7<\/h2>\n<p>\u73b0\u4ee3\u9ad8\u6027\u80fd\u8ba1\u7b97\u96c6\u7fa4\u9700\u8981\u5bf9\u5e95\u5c42\u7cfb\u7edf\u8bbe\u7f6e\u8fdb\u884c\u76f4\u63a5\u63a7\u5236\u3002\u4f60\u53ef\u4ee5\u4e3a\u9ad8\u9700\u6c42\u786c\u4ef6\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ee5\u6d88\u9664\u6df1\u5ea6\u5b66\u4e60\u74f6\u9888\u3002\u6070\u5f53\u7684\u6267\u884c\u7ba1\u7406\u80fd\u591f\u4f18\u5316\u8d44\u6e90\u5206\u914d\u3001\u4fdd\u62a4\u7cfb\u7edf\u7a33\u5b9a\u6027\uff0c\u5e76\u63a8\u52a8\u6574\u4f53\u7ec4\u7ec7\u6548\u7387\u63d0\u5347\u3002<\/p>\n<h3>\u8c03\u6574\u7cfb\u7edf CPU \u548c GPU \u4f18\u5148\u7ea7<\/h3>\n<p>\u5185\u6838\u7ea7\u7cfb\u7edf\u8bbe\u7f6e\u4f1a\u76f4\u63a5\u5f71\u54cd\u8bad\u7ec3\u901f\u5ea6\u3002\u6807\u51c6\u64cd\u4f5c\u7cfb\u7edf\u8c03\u5ea6\u5668\u7ba1\u7406\u5904\u7406\u5668\u786c\u4ef6\u65f6\uff0c\u5e76\u4e0d\u4f1a\u8003\u8651 NVLink \u6216\u975e\u7edf\u4e00\u5185\u5b58\u8bbf\u95ee\uff08NUMA\uff09\u7b49\u7269\u7406\u4e92\u8fde\u62d3\u6251\u3002\u5ffd\u89c6\u62d3\u6251\u7ed3\u6784\u7684\u653e\u7f6e\u65b9\u5f0f\u4f1a\u8feb\u4f7f\u6570\u636e\u7a7f\u8d8a\u8f83\u8fdc\u7684\u5185\u5b58\u603b\u7ebf\uff0c\u4ece\u800c\u589e\u52a0\u4e3b\u673a\u5904\u7406\u5668\u4e0e\u52a0\u901f\u5668\u4e4b\u95f4\u7684\u901a\u4fe1\u5ef6\u8fdf\u3002\u901a\u8fc7\u5b9e\u65bd\u62d3\u6251\u611f\u77e5\u7684\u653e\u7f6e\u89c4\u5219\uff0c\u4f60\u53ef\u4ee5\u6d88\u9664\u8de8 NUMA \u5185\u5b58\u5e26\u6765\u7684\u6027\u80fd\u635f\u5931\u3002<\/p>\n<p>\u8c03\u6574\u5e95\u5c42\u5904\u7406\u5668\u8bbe\u7f6e\u53ef\u4ee5\u63d0\u5347\u786c\u4ef6\u541e\u5410\u91cf\u5e76\u589e\u5f3a\u4efb\u52a1\u6267\u884c\u6548\u7387\u3002\u901a\u8fc7 <code>nice<\/code> \u547d\u4ee4\u8fdb\u884c\u6807\u51c6\u4f18\u5148\u7ea7\u8c03\u6574\uff0c\u5728\u8f7b\u5ea6\u5171\u4eab\u8d1f\u8f7d\u4e0b\u901a\u5e38\u53ea\u80fd\u5e26\u6765 5\u201310% \u7684\u9002\u5ea6\u63d0\u5347\u3002\u64cd\u4f5c\u7cfb\u7edf\u8c03\u5ea6\u5668\u5f15\u5165\u7684\u5f00\u9500\u4f1a\u6539\u53d8\u6267\u884c\u901f\u5ea6\uff1a\u5728\u5e72\u51c0\u73af\u5883\u4e2d\u5f71\u54cd\u7ea6\u4e3a 1.2%\uff0c\u800c\u5728\u7ec4\u5408\u8d1f\u8f7d\u4e0b\u53ef\u9ad8\u8fbe 20.5%\u3002\u901a\u8fc7 <code>isolcpus<\/code> \u4e0e\u4e2d\u65ad\u8bf7\u6c42\u4eb2\u548c\u6027\uff08IRQ affinity\uff09\u5b9e\u73b0\u5b8c\u6574 CPU \u9694\u79bb\uff0c\u53ef\u4e3a\u5bf9\u5ef6\u8fdf\u654f\u611f\u7684\u4efb\u52a1\u5e26\u6765 15\u201320% \u7684\u63d0\u5347\u3002\u6b64\u5916\uff0c\u5904\u7406\u5668\u7ed1\u5b9a\u53ef\u5728\u5171\u4eab\u670d\u52a1\u5668\u4e0a\u5c06\u4e0a\u4e0b\u6587\u5207\u6362\u51cf\u5c11 96%\u3002<\/p>\n<div class=\"qc-default-table-wrapper \">\n<table style=\"min-width: 75px\">\n<colgroup>\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\"><\/colgroup>\n<tbody>\n<tr>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u4f18\u5316\u63aa\u65bd<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u89c2\u6d4b\u5230\u7684\u6548\u679c<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u9002\u7528\u573a\u666f<\/p>\n<\/th>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>CPU \u7ed1\u5b9a\uff08affinity\uff09<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u4e0a\u4e0b\u6587\u5207\u6362\u51cf\u5c11 96%\uff1b\u53ef\u6062\u590d\u5927\u90e8\u5206\u56e0\u8d44\u6e90\u5e72\u6270\u800c\u635f\u5931\u7684\u6027\u80fd\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u5728 CPU\u3001\u7f51\u7edc\u3001\u78c1\u76d8\u7b49\u9ad8\u5f3a\u5ea6\u6df7\u90e8\u8d1f\u8f7d\u4e0b\u8fd0\u884c\u7684\u5171\u4eab\u670d\u52a1\u5668\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u4f18\u5148\u7ea7\u8c03\u6574\uff08nice\uff09<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u6709\u5e2e\u52a9\uff0c\u4f46\u63d0\u5347\u6709\u9650\uff085\u201310%\uff09\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u8f7b\u5ea6\u6df7\u90e8\u8d1f\u8f7d\u4e0b\u7684\u5171\u4eab\u670d\u52a1\u5668\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u5b8c\u6574 CPU \u9694\u79bb\uff08isolcpus + IRQ affinity\uff09<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u63d0\u5347 15\u201320%\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u9ad8\u5f3a\u5ea6\u5171\u4eab\u670d\u52a1\u5668\u4e0a\u7684\u4f4e\u5ef6\u8fdf\u654f\u611f\u578b\u5de5\u4f5c\u8d1f\u8f7d\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u4e0d\u505a\u4f18\u5316<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u8c03\u5ea6\u5668\u5f71\u54cd\u8303\u56f4\u4ece 1.2%\uff08\u5e72\u51c0\u73af\u5883\uff09\u5230 20.5%\uff08\u7ec4\u5408\u8d1f\u8f7d\uff09\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u72ec\u7acb\u670d\u52a1\u5668\u6216\u5bf9\u5ef6\u8fdf\u4e0d\u654f\u611f\u7684\u4f5c\u4e1a\u3002<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<h3>\u5c06\u5de5\u4f5c\u8d1f\u8f7d\u5c42\u7ea7\u6620\u5c04\u5230\u786c\u4ef6\u9650\u5236<\/h3>\n<p>\u5c06\u5de5\u4f5c\u8d1f\u8f7d\u9700\u6c42\u5339\u914d\u5230\u660e\u786e\u7684\u786c\u4ef6\u5c42\u7ea7\uff0c\u53ef\u4ee5\u9632\u6b62\u7b97\u529b\u8d44\u6e90\u9965\u997f\u3002\u4f60\u5fc5\u987b\u5728\u5404\u7c7b\u786c\u4ef6\u8d44\u6e90\u4e0a\u914d\u7f6e\u4f18\u5148\u7ea7\u53c2\u6570\uff0c\u4ee5\u4f18\u5316\u6574\u4f53\u6267\u884c\u6d41\u7a0b\u3002\u6210\u529f\u7684\u786c\u4ef6\u6620\u5c04\u80fd\u591f\u4fdd\u62a4\u9ad8\u4f18\u5148\u7ea7\u5de5\u4f5c\u8d1f\u8f7d\uff0c\u5e76\u63d0\u5347\u6574\u4f53\u4e1a\u52a1\u751f\u4ea7\u6548\u7387\u3002<\/p>\n<ol>\n<li>\n<p>\u8bb0\u5f55\u6a21\u578b\u7279\u5f81\uff0c\u5305\u62ec\u53c2\u6570\u91cf\u3001\u6a21\u578b\u89c4\u6a21\u548c\u7cbe\u5ea6\u7ea7\u522b\u3002<\/p>\n<\/li>\n<li>\n<p>\u5b9a\u4e49\u660e\u786e\u7684\u6027\u80fd\u76ee\u6807\uff0c\u4f8b\u5982\u76ee\u6807 batch size\u3001\u5e8f\u5217\u957f\u5ea6\u3001\u6bcf\u79d2 token \u6570\u4ee5\u53ca\u5ef6\u8fdf\u76ee\u6807\u3002<\/p>\n<\/li>\n<li>\n<p>\u8ddf\u8e2a\u6570\u636e\u9700\u6c42\uff0c\u5c24\u5176\u662f\u6570\u636e\u96c6\u5927\u5c0f\u3001checkpoint \u6587\u4ef6\u5927\u5c0f\u4ee5\u53ca\u6570\u636e\u4fdd\u7559\u5468\u671f\u3002<\/p>\n<\/li>\n<li>\n<p>\u8bb0\u5f55\u5408\u89c4\u89c4\u5219\uff0c\u5305\u62ec\u9488\u5bf9\u654f\u611f\u8d44\u4ea7\u7684\u4e25\u683c\u6570\u636e\u9a7b\u7559\u9650\u5236\u3002<\/p>\n<\/li>\n<\/ol>\n<p>\u6070\u5f53\u7684\u786c\u4ef6\u5206\u5c42\u53ef\u6700\u5927\u5316\u590d\u6742 AI \u7cfb\u7edf\u4e2d\u7684\u8fd0\u884c\u6548\u7387\u3002\u8fd0\u884c\u53c2\u6570\u9ad8\u6548\u5fae\u8c03\u7684\u5c0f\u578b\u4efb\u52a1\uff0c\u4f8b\u5982 8B \u81f3 13B \u6a21\u578b\uff0c\u53ef\u5728\u5355\u5f20 A100 80GB GPU \u4e0a\u9ad8\u6548\u6267\u884c\u3002\u4e2d\u578b\u5de5\u4f5c\u8d1f\u8f7d\uff0c\u5982\u5bf9 70B \u6a21\u578b\u8fdb\u884c\u5168\u91cf\u5fae\u8c03\uff0c\u5219\u9700\u8981 840 GB \u7684\u663e\u5b58\u6765\u5bb9\u7eb3\u53c2\u6570\u3001\u4f18\u5316\u5668\u72b6\u6001\u548c\u68af\u5ea6\u3002\u4f60\u5fc5\u987b\u5728\u591a\u8282\u70b9\u73af\u5883\u4e2d\u8fd0\u884c\u8fd9\u7c7b\u4e2d\u578b\u4efb\u52a1\uff0c\u4f8b\u5982\u4f7f\u7528 4xA100 80GB NVLink pod\u3002\u8fbe\u5230 405B \u53c2\u6570\u89c4\u6a21\u7684\u524d\u6cbf AI \u6a21\u578b\u5219\u9700\u8981\u9876\u7ea7\u786c\u4ef6\uff0c\u4f8b\u5982\u91c7\u7528 FP8 \u7cbe\u5ea6\u7684 8xH100 SXM \u96c6\u7fa4\u3002<\/p>\n<p>\u4f60\u8fd8\u53ef\u4ee5\u901a\u8fc7\u5c06\u786c\u4ef6\u6620\u5c04\u4e0e\u5b9e\u65f6\u76d1\u63a7\u7ed3\u5408\u8d77\u6765\uff0c\u907f\u514d\u961f\u5217\u9501\u6b7b\u3002\u81ea\u52a8\u5316\u76d1\u63a7\u4f1a\u8ddf\u8e2a\u8fd0\u884c\u72b6\u6001\uff0c\u5e76\u5728\u74f6\u9888\u5373\u5c06\u51fa\u73b0\u65f6\u5411\u56e2\u961f\u53d1\u51fa\u544a\u8b66\u3002\u6709\u6548\u7684\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u786e\u4fdd\u9ad8\u5e76\u53d1\u65f6\u671f\u7684\u5e73\u7a33\u8fd0\u884c\u3002\u9ad8\u4f18\u5148\u7ea7\u4efb\u52a1\u4f1a\u7acb\u5373\u83b7\u53d6\u6240\u9700\u8d44\u6e90\uff0c\u800c\u4f4e\u4f18\u5148\u7ea7\u4f5c\u4e1a\u5219\u5b89\u5168\u5730\u8ba9\u51fa\u6267\u884c\u5e26\u5bbd\u3002\u6070\u5f53\u7684\u4efb\u52a1\u7ba1\u7406\u80fd\u591f\u63d0\u5347\u6574\u4e2a AI \u57fa\u7840\u8bbe\u65bd\u7684\u5904\u7406\u6548\u80fd\u3002<\/p>\n<h2>\u4e3a AI \u4efb\u52a1\u4f18\u5148\u7ea7\u914d\u7f6e\u8c03\u5ea6\u5668<\/h2>\n<p>\u73b0\u4ee3\u8c03\u5ea6\u5668\u5141\u8bb8\u4f60\u5728\u96c6\u7fa4\u73af\u5883\u4e2d\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\u3002\u4f60\u53ef\u4ee5\u6709\u6548\u7ba1\u7406\u5206\u5e03\u5f0f\u786c\u4ef6\uff0c\u5e76\u5e73\u8861\u5404\u4e2a\u5de5\u4f5c\u8282\u70b9\u7684\u8d1f\u8f7d\u3002\u81ea\u52a8\u5316\u8c03\u5ea6\u5668\u7b80\u5316\u4e86\u8d44\u6e90\u5206\u914d\uff0c\u540c\u65f6\u4fdd\u6301\u4f5c\u4e1a\u6267\u884c\u7684\u7a33\u5b9a\u6027\u3002\u6070\u5f53\u7684 AI \u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u5728\u9700\u6c42\u9ad8\u5cf0\u671f\u95f4\u4fdd\u6301\u5904\u7406\u6d41\u6c34\u7ebf\u9ad8\u6548\u8fd0\u8f6c\u3002\u81ea\u52a8\u5316\u4efb\u52a1\u4f18\u5148\u7ea7\u673a\u5236\u7b80\u5316\u4e86\u751f\u4ea7\u670d\u52a1\u5668\u4e0a\u7684\u4f5c\u4e1a\u63d0\u4ea4\u961f\u5217\u7ba1\u7406\u3002\u4f60\u53ef\u4ee5\u90e8\u7f72\u667a\u80fd\u8c03\u5ea6\u89c4\u5219\uff0c\u4ee5\u6700\u5927\u5316 AI \u5de5\u4f5c\u8d1f\u8f7d\u7684\u7b97\u529b\u541e\u5410\u3002<\/p>\n<h3>\u90e8\u7f72 Kubernetes PriorityClass \u548c Kueue<\/h3>\n<p>Kubernetes \u4f7f\u7528\u539f\u751f\u5bf9\u8c61\u6765\u63a7\u5236\u5de5\u4f5c\u8d1f\u8f7d\u7684\u6267\u884c\u987a\u5e8f\u3002PriorityClass \u4f1a\u4e3a\u65b0\u63d0\u4ea4\u7684\u4f5c\u4e1a\u5206\u914d\u6570\u503c\u4f18\u5148\u7ea7\u3002\u8c03\u5ea6\u5668\u4f1a\u6839\u636e\u8fd9\u4e9b\u6570\u503c\uff0c\u4f18\u5148\u5b89\u6392\u9ad8\u4f18\u5148\u7ea7\u5de5\u4f5c\u8d1f\u8f7d\u5165\u961f\u3002\u4f8b\u5982\uff0c\u9ad8\u4f18\u5148\u7ea7\u63a8\u7406\u4efb\u52a1\u53ef\u4ee5\u7acb\u5373\u62a2\u5360\u4f4e\u4f18\u5148\u7ea7\u4efb\u52a1\uff08\u5982\u5e38\u89c4\u8bad\u7ec3\uff09\u3002\u7cfb\u7edf\u4f1a\u9a71\u9010\u4f4e\u4f18\u5148\u7ea7 Pod\uff0c\u5e76\u5c06\u5176\u91cd\u65b0\u653e\u56de\u5f85\u8c03\u5ea6\u961f\u5217\u3002\u7d27\u6025\u4efb\u52a1\u56e0\u6b64\u53ef\u4ee5\u7acb\u5373\u83b7\u53d6\u53ef\u7528 GPU\uff0c\u786e\u4fdd\u5173\u952e\u5de5\u4f5c\u4e0d\u88ab\u5ef6\u8fdf\u3002<\/p>\n<div class=\"qc-default-table-wrapper \">\n<table style=\"min-width: 75px\">\n<colgroup>\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\"><\/colgroup>\n<tbody>\n<tr>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u6982\u5ff5<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u8bf4\u660e<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u5728 AI \u8bad\u7ec3 \/ GPU \u573a\u666f\u4e2d\u7684\u5e94\u7528<\/p>\n<\/th>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u4f18\u5148\u7ea7\u89c4\u5219<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u4e3a\u5de5\u4f5c\u8d1f\u8f7d\u5206\u914d\u91cd\u8981\u6027\u6570\u503c\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u8ba9\u7d27\u6025 AI \u4efb\u52a1\u4f18\u5148\u4e8e\u4f4e\u4f18\u5148\u7ea7\u8bad\u7ec3\u4efb\u52a1\u8fdb\u884c\u8c03\u5ea6\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u62a2\u5360\uff08Preemption\uff09<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u9a71\u9010\u6b63\u5728\u8fd0\u884c\u7684\u4f4e\u4f18\u5148\u7ea7\u4f5c\u4e1a\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u7acb\u5373\u91ca\u653e GPU \u786c\u4ef6\u7ed9\u9ad8\u4f18\u5148\u7ea7\u4efb\u52a1\u4f7f\u7528\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u914d\u7f6e\u793a\u4f8b<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u4e3a Pod \u5b9a\u4e49 PriorityClass \u5bf9\u8c61\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u7ba1\u7406\u6df7\u5408\u5de5\u4f5c\u8d1f\u8f7d\u5e76\u6700\u5927\u5316 GPU \u5229\u7528\u7387\u3002<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p>\u4f60\u8fd8\u53ef\u4ee5\u4f7f\u7528 Kueue \u6269\u5c55 Kubernetes \u539f\u751f\u8c03\u5ea6\u80fd\u529b\u3002Kueue \u5728\u591a\u79df\u6237 AI \u96c6\u7fa4\u4e2d\u7ba1\u7406\u57fa\u4e8e\u961f\u5217\u7684\u51c6\u5165\u63a7\u5236\u3001gang \u8bed\u4e49\u4ee5\u53ca\u914d\u989d\u6cbb\u7406\u3002LocalQueue \u7528\u4e8e\u7ec4\u7ec7\u547d\u540d\u7a7a\u95f4\u5185\u7684\u8bf7\u6c42\uff0c\u800c ClusterQueue \u5219\u7528\u4e8e\u5b9e\u65bd\u96c6\u7fa4\u8303\u56f4\u7b56\u7565\u3002Gang \u8c03\u5ea6\u53ef\u786e\u4fdd\u6240\u6709\u5de5\u4f5c Pod \u540c\u65f6\u542f\u52a8\uff0c\u4ece\u800c\u907f\u514d\u5206\u5e03\u5f0f AI \u8bad\u7ec3\u8fd0\u884c\u8fc7\u7a0b\u4e2d\u51fa\u73b0\u6b7b\u9501\u3002Kueue \u53ef\u5c06\u96c6\u7fa4 GPU \u5229\u7528\u7387\u4ece 30\u201350% \u63d0\u5347\u81f3 80\u201395%\u3002<\/p>\n<p>\u5c06 Kueue \u4e0e\u7a33\u5065\u7684\u5de5\u4f5c\u6d41\u7ba1\u7406\u5f15\u64ce\u8fde\u63a5\u8d77\u6765\uff0c\u6709\u52a9\u4e8e\u4f60\u53ef\u9760\u5730\u5904\u7406\u8bad\u7ec3\u6570\u636e\u6d41\u6c34\u7ebf\u3002\u9009\u62e9\u4f01\u4e1a\u7ea7\u4efb\u52a1\u7ba1\u7406\u5de5\u5177\u53ef\u4ee5\u7b80\u5316\u6258\u7ba1 AI \u4f5c\u4e1a\u7684\u961f\u5217\u63d0\u4ea4\u89c4\u5219\u3002\u4f60\u53ef\u4ee5\u7f16\u6392\u590d\u6742\u5de5\u4f5c\u6d41\uff0c\u5e76\u81ea\u52a8\u8def\u7531\u91cd\u590d\u6027\u4efb\u52a1\u3002\u6218\u7565\u6027\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u80fd\u591f\u4f18\u5316\u7ec4\u7ec7\u5185\u5173\u952e\u4f5c\u4e1a\u7684\u6267\u884c\u987a\u5e8f\u3002<\/p>\n<h3>\u5b9a\u4e49 Slurm QoS \u548c\u961f\u5217\u89c4\u5219<\/h3>\n<p>Slurm \u4e3a\u9ad8\u6027\u80fd\u8ba1\u7b97\u96c6\u7fa4\u63d0\u4f9b\u4e86\u9ad8\u7ea7\u63a7\u5236\u80fd\u529b\u3002\u4f60\u53ef\u4ee5\u5efa\u7acb\u670d\u52a1\u8d28\u91cf\uff08QoS\uff09\u89c4\u5219\u6765\u89c4\u8303 AI \u4f5c\u4e1a\u7684\u961f\u5217\u8bbf\u95ee\u3002Slurm \u4f1a\u8bc4\u4f30\u5355\u4e2a\u4f5c\u4e1a\u9650\u5236\u3001\u4f18\u5148\u7ea7\u6807\u8bb0\u4ee5\u53ca\u516c\u5e73\u5171\u4eab\u56e0\u5b50\u3002\u9ad8\u4f18\u5148\u7ea7\u7684\u4ea4\u4e92\u5f0f\u8c03\u8bd5\u4efb\u52a1\u53ef\u4ee5\u7ed5\u8fc7\u8017\u65f6\u8f83\u957f\u7684\u6279\u91cf\u8bad\u7ec3\u4f5c\u4e1a\u3002<\/p>\n<p>\u5728 AI \u8fd0\u7ef4\u4e2d\u6392\u67e5\u8c03\u5ea6\u5668\u961f\u5217\u95ee\u9898\u65f6\uff0c\u9700\u8981\u8fdb\u884c\u7cfb\u7edf\u5316\u68c0\u67e5\u3002<\/p>\n<ol>\n<li>\n<p>\u5f53\u4f5c\u4e1a\u9009\u62e9\u4e86\u9519\u8bef\u7684 GPU \u7c7b\u578b\u65f6\uff0c\u9a8c\u8bc1\u8282\u70b9\u6807\u7b7e\u662f\u5426\u4e0e\u5b9a\u4e49\u597d\u7684 ResourceFlavors \u5339\u914d\u3002<\/p>\n<\/li>\n<li>\n<p>\u5f53\u4f5c\u4e1a\u957f\u671f\u5904\u4e8e\u961f\u5217\u4e2d\u4e14\u4e0d\u53ef\u51c6\u5165\u65f6\uff0c\u589e\u52a0 ClusterQueue \u4e2d\u7684 <code>nominalQuota<\/code>\u3002<\/p>\n<\/li>\n<li>\n<p>\u5f53\u4f5c\u4e1a\u59cb\u7ec8\u65e0\u6cd5\u51c6\u5165\u96c6\u7fa4\u8d44\u6e90\u65f6\uff0c\u6dfb\u52a0 <code>kueue.x-k8s.io\/queue-name<\/code> \u6807\u7b7e\u3002<\/p>\n<\/li>\n<li>\n<p>\u5f53\u5404\u6d3b\u8dc3\u56e2\u961f\u4e4b\u95f4\u7684\u516c\u5e73\u5171\u4eab\u8868\u73b0\u5931\u8861\u65f6\uff0c\u8c03\u6574 <code>fairSharing.weight<\/code> \u7684\u53d6\u503c\u3002<\/p>\n<\/li>\n<\/ol>\n<p>\u6709\u6548\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u80fd\u591f\u9632\u6b62\u5171\u4eab\u96c6\u7fa4\u4e2d\u7684\u8d44\u6e90\u9965\u997f\u3002\u4f60\u53ef\u4ee5\u5904\u7406\u5927\u578b\u6570\u636e\u96c6\u6587\u4ef6\u5e76\u6444\u53d6\u6570\u636e\uff0c\u800c\u4e0d\u4f1a\u5e72\u6270\u6b63\u5728\u8fd0\u884c\u7684\u4f5c\u4e1a\u3002\u81ea\u5b9a\u4e49\u961f\u5217\u89c4\u5219\u53ef\u4ee5\u5728\u591a\u8282\u70b9\u96c6\u7fa4\u4e2d\u81ea\u52a8\u5316\u6267\u884c\u63a7\u5236\u3002\u7a33\u5065\u7684 AI \u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u4e3a\u590d\u6742\u6df1\u5ea6\u5b66\u4e60\u5de5\u4f5c\u8d1f\u8f7d\u63d0\u4f9b\u53ef\u9760\u7684\u961f\u5217\u541e\u5410\u3002\u5148\u8fdb\u7684\u4efb\u52a1\u7ba1\u7406\u5de5\u5177\u80fd\u529b\u5e2e\u52a9\u56e2\u961f\u4fdd\u62a4\u6570\u636e\u64cd\u4f5c\u3002\u6bcf\u4e2a\u4efb\u52a1\u90fd\u9700\u8981\u660e\u786e\u7684\u914d\u989d\u53c2\u6570\u3002\u4efb\u52a1\u53ef\u4ee5\u5728\u961f\u5217\u4e2d\u7b49\u5f85\uff0c\u76f4\u5230\u8d44\u6e90\u53ef\u7528\u3002\u5f53 GPU \u91ca\u653e\u51fa\u6765\u65f6\uff0c\u4efb\u52a1\u4f1a\u5e73\u7a33\u8fdb\u5165\u6267\u884c\u9636\u6bb5\u3002\u8be5\u4efb\u52a1\u4f1a\u5360\u7528\u5206\u914d\u5230\u7684\u5185\u5b58\uff0c\u800c\u53e6\u4e00\u4e2a\u4efb\u52a1\u5219\u5b89\u5168\u5730\u8ba9\u51fa\u5e26\u5bbd\u3002\u6bcf\u4e00\u4e2a\u4efb\u52a1\u90fd\u4f1a\u6309\u7167\u65e2\u5b9a\u961f\u5217\u89c4\u5219\u8fd0\u884c\u3002\u6bcf\u4e00\u4e2a\u4f18\u5148\u7ea7\u4efb\u52a1\u90fd\u80fd\u83b7\u5f97\u6700\u4f18\u7684\u6267\u884c\u8c03\u5ea6\u3002\u73b0\u4ee3 AI \u7cfb\u7edf\u4f9d\u8d56\u4e00\u81f4\u7684\u961f\u5217\u6cbb\u7406\u3002\u6070\u5f53\u7684\u8c03\u5ea6\u7b56\u7565\u80fd\u591f\u4fdd\u62a4\u4f60\u7684\u6570\u636e\u57fa\u7840\u8bbe\u65bd\uff0c\u5e76\u8ba9 AI \u57fa\u7840\u8bbe\u65bd\u59cb\u7ec8\u4fdd\u6301\u6700\u4f73\u6027\u80fd\u3002<\/p>\n<h2>\u5229\u7528 AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406<\/h2>\n<h3>\u4f7f\u7528\u673a\u5668\u5b66\u4e60\u6a21\u578b\u9884\u6d4b\u4f5c\u4e1a\u4f18\u5148\u7ea7<\/h3>\n<p>\u73b0\u4ee3 AI \u7cfb\u7edf\u5728\u9762\u5bf9\u9ad8\u8d1f\u8f7d\u961f\u5217\u65f6\uff0c\u9700\u8981\u9884\u6d4b\u5f0f\u667a\u80fd\u3002\u4f60\u53ef\u4ee5\u5b9e\u65bd AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\uff0c\u4ee5\u81ea\u52a8\u8bbe\u7f6e\u6267\u884c\u53c2\u6570\u3002\u673a\u5668\u5b66\u4e60\u6a21\u578b\u4f1a\u5206\u6790\u5386\u53f2\u6267\u884c\u6570\u636e\uff0c\u8bc4\u4f30\u65b0\u5230\u8fbe AI \u5de5\u4f5c\u8d1f\u8f7d\u4e0e\u4efb\u52a1\u7684\u8d44\u6e90\u9700\u6c42\u3002\u8fd9\u4e9b\u9884\u6d4b\u5f0f AI \u6a21\u578b\u4f1a\u4e3a\u6bcf\u4e2a\u4efb\u52a1\u8ba1\u7b97\u9884\u8ba1\u65f6\u957f\u3001\u663e\u5b58\u5360\u7528\u548c GPU \u62d3\u6251\u9002\u914d\u5ea6\u3002\u8fd9\u6837\u7684\u81ea\u52a8\u5316\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u4ee5\u5373\u65f6\u5c06\u5de5\u4f5c\u8d1f\u8f7d\u5339\u914d\u5230\u6700\u4f18\u786c\u4ef6\uff0c\u4ece\u800c\u6700\u5927\u5316\u786c\u4ef6\u4ea7\u51fa\u3002\u81ea\u52a8\u5316 AI \u80fd\u529b\u8fd8\u4f1a\u5728\u5206\u6790\u6570\u636e\u7279\u5f81\u7684\u540c\u65f6\uff0c\u901a\u8fc7\u9884\u6d4b\u5f0f\u5bb9\u91cf\u9884\u7559\u6765\u51cf\u5c11\u6574\u4f53\u6392\u961f\u5ef6\u8fdf\u3002<\/p>\n<p>\u91c7\u7528\u9884\u6d4b\u5f0f\u4efb\u52a1\u4f18\u5148\u7ea7\u7cfb\u7edf\uff0c\u80fd\u591f\u663e\u8457\u63d0\u5347\u4e1a\u52a1\u751f\u4ea7\u529b\u3002\u4e0e\u6807\u51c6\u4efb\u52a1\u8c03\u5ea6\u7b56\u7565\u76f8\u6bd4\uff0c\u9884\u6d4b\u7b97\u6cd5\u53ef\u5927\u5e45\u7b80\u5316\u96c6\u7fa4\u7ba1\u7406\u3002<\/p>\n<h3>\u901a\u8fc7 WebSockets \u89e3\u8026\u4f18\u5148\u7ea7\u5f15\u64ce<\/h3>\n<p>\u4f60\u5fc5\u987b\u5c06\u4f18\u5148\u7ea7\u8bc4\u5206\u903b\u8f91\u4e0e\u6838\u5fc3 AI \u96c6\u7fa4\u7ec4\u4ef6\u9694\u79bb\u5f00\u6765\u3002\u5c06\u8be5\u5f15\u64ce\u6784\u5efa\u4e3a\u72ec\u7acb\u7684\u540e\u7aef\u670d\u52a1\uff0c\u53ef\u4ee5\u4fdd\u62a4\u4e3b\u5de5\u4f5c\u6d41\u3002WebSockets \u80fd\u591f\u5411\u4eea\u8868\u76d8\u5b9e\u65f6\u4f20\u9012\u66f4\u65b0\uff0c\u800c\u4e0d\u4f1a\u964d\u4f4e\u540e\u7aef\u6027\u80fd\u3002\u5f00\u53d1\u8005\u53ef\u901a\u8fc7\u6301\u7eed\u7684\u5b9e\u65f6\u76d1\u63a7\u754c\u9762\u8ddf\u8e2a\u5f53\u524d\u961f\u5217\u72b6\u6001\uff0c\u540c\u65f6\u5b58\u50a8\u8bad\u7ec3\u6570\u636e\u6307\u6807\u3002\u5b9e\u65f6\u6570\u636e\u6d41\u4f1a\u5c06\u72b6\u6001\u53d8\u5316\u5373\u65f6\u4f20\u9001\u5230\u7528\u6237\u754c\u9762\u3002<\/p>\n<p>\u89e3\u8026\u540e\u7684\u6267\u884c\u5f15\u64ce\u80fd\u591f\u5728\u5404\u6d3b\u8dc3\u8282\u70b9\u5904\u7406\u5355\u4e2a\u4efb\u52a1\u65f6\uff0c\u5b9e\u73b0\u54cd\u5e94\u5f0f\u7684\u52a8\u6001\u4f18\u5148\u7ea7\u7ba1\u7406\u3002\u4f60\u53ef\u4ee5\u5728\u4e0d\u4e2d\u65ad\u6d3b\u52a8\u5de5\u4f5c\u8d1f\u8f7d\u7684\u60c5\u51b5\u4e0b\uff0c\u66f4\u65b0\u5f85\u5904\u7406\u4efb\u52a1\u7684\u4f18\u5148\u7ea7\u53c2\u6570\u3002\u968f\u7740\u65b0\u4efb\u52a1\u8fdb\u5165\u961f\u5217\uff0c\u9694\u79bb\u7684\u540e\u7aef\u4f1a\u91cd\u65b0\u8ba1\u7b97\u5206\u6570\u53d8\u5316\u3002\u8be5\u67b6\u6784\u4f1a\u901a\u8fc7 API \u8c03\u6574\u8c03\u5ea6\u5668\u89c4\u5219\uff0c\u5e76\u5c06\u5b9e\u65f6\u66f4\u65b0\u53d1\u9001\u5230 Web \u754c\u9762\u3002\u9ad8\u4f18\u5148\u7ea7\u4efb\u52a1\u80fd\u591f\u66f4\u5feb\u5730\u5728\u961f\u5217\u4e2d\u524d\u79fb\uff0c\u800c\u4e0d\u4f1a\u5e72\u6270\u540e\u53f0\u6570\u636e\u6d41\u6c34\u7ebf\u6216\u5e38\u89c4\u5904\u7406\u4efb\u52a1\u3002\u56e0\u6b64\uff0c\u5b9e\u65f6\u66f4\u65b0\u53ef\u4ee5\u6700\u5927\u5316\u56e2\u961f\u751f\u4ea7\u6548\u7387\u5e76\u63d0\u5347\u6574\u4f53\u8fd0\u8425\u6210\u6548\u3002\u7a33\u5065\u7684 AI \u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u80fd\u591f\u4e3a\u590d\u6742\u7684\u4f01\u4e1a\u7ea7 AI \u57fa\u7840\u8bbe\u65bd\u5e26\u6765\u53ef\u8861\u91cf\u7684\u6548\u7387\u63d0\u5347\u3002\u667a\u80fd\u7684 AI \u9a71\u52a8\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u4f1a\u5c06\u9759\u6001\u961f\u5217\u7ed3\u6784\u8f6c\u53d8\u4e3a\u5177\u6709\u54cd\u5e94\u80fd\u529b\u7684\u8fd0\u8425\u8d44\u4ea7\u3002\u5e94\u7528 AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\uff0c\u53ef\u4ee5\u4e3a\u73b0\u4ee3\u5de5\u7a0b\u56e2\u961f\u6253\u9020\u81ea\u9002\u5e94\u8c03\u5ea6\u73af\u5883\u3002\u5c06 AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u673a\u5236\u96c6\u6210\u5230\u7cfb\u7edf\u4e2d\uff0c\u53ef\u4f18\u5316\u4f01\u4e1a\u786c\u4ef6\u7684\u7b97\u529b\u4f7f\u7528\u3002<\/p>\n<h2>\u5b9e\u65bd\u4e3b\u52a8\u62a2\u5360\u4e0e\u4f18\u5148\u7ea7\u4fdd\u62a4\u673a\u5236<\/h2>\n<p>\u4f60\u9700\u8981\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ee5\u4fbf\u5728\u9ad8\u8d1f\u8f7d\u65f6\u671f\u4fdd\u62a4\u786c\u4ef6\u8d44\u6e90\u3002Kubernetes \u539f\u751f\u8c03\u5ea6\u7b56\u7565\u53ef\u7ba1\u7406\u8d44\u6e90\u5206\u914d\u5e76\u963b\u6b62 GPU \u8d44\u6e90\u9965\u997f\u3002\u9ad8\u4f18\u5148\u7ea7\u4efb\u52a1\u4f1a\u7acb\u5373\u89e6\u53d1\u4e3b\u52a8\u62a2\u5360\uff0c\u4ee5\u6ee1\u8db3\u7d27\u6025\u6267\u884c\u9700\u6c42\u3002\u5f53\u4f60\u4fdd\u62a4\u6d3b\u52a8\u4efb\u52a1\u4e0d\u88ab\u7a81\u7136\u7ec8\u6b62\u65f6\uff0c\u6574\u4f53\u8fd0\u884c\u6548\u7387\u4e5f\u4f1a\u5f97\u5230\u63d0\u5347\u3002\u6301\u7eed\u76d1\u63a7\u4f1a\u8ddf\u8e2a\u96c6\u7fa4\u5065\u5eb7\u72b6\u6001\uff0c\u4ee5\u7ef4\u6301\u6240\u6709\u8282\u70b9\u4e0a\u7684\u6267\u884c\u6548\u7387\u3002<\/p>\n<h3>\u5728\u8bad\u7ec3\u4e2d\u7ba1\u7406\u4f18\u96c5\u9a71\u9010<\/h3>\n<p>\u975e\u8ba1\u5212\u9a71\u9010\u4f1a\u7834\u574f\u5173\u952e\u6d41\u6c34\u7ebf\u6570\u636e\u3002\u73b0\u4ee3 AI \u57fa\u7840\u8bbe\u65bd\u9700\u8981\u4f18\u96c5\u7ec8\u6b62\u5904\u7406\u673a\u5236\uff0c\u4ee5\u5728\u6df1\u5ea6\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4fdd\u62a4\u6a21\u578b\u8fdb\u5ea6\u3002\u5404\u7c7b\u6846\u67b6\u501f\u52a9 checkpoint \u673a\u5236\uff0c\u4f7f\u88ab\u9a71\u9010\u7684\u4efb\u52a1\u80fd\u591f\u5728\u65b0\u8282\u70b9\u4e0a\u5e73\u7a33\u6062\u590d\u3002\u7d27\u6025\u4efb\u52a1\u53ef\u4ee5\u5360\u7528\u5f53\u524d\u52a0\u901f\u5668\uff0c\u800c\u5408\u7406\u7684\u4fdd\u5b58\u95f4\u9694\u5219\u53ef\u5c06\u5df2\u5b8c\u6210\u5de5\u4f5c\u7684\u635f\u5931\u964d\u5230\u6700\u4f4e\u3002<\/p>\n<div class=\"qc-default-table-wrapper \">\n<table style=\"min-width: 75px\">\n<colgroup>\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\"><\/colgroup>\n<tbody>\n<tr>\n<th colspan=\"1\" rowspan=\"1\">\n<p>Checkpoint \u7c7b\u578b<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u8bf4\u660e<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u5728\u9a71\u9010\u671f\u95f4\u5bf9\u8bad\u7ec3\u4f5c\u4e1a\u8fdb\u5ea6\u7684\u5f71\u54cd<\/p>\n<\/th>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p><strong>\u5e94\u7528\u5c42\uff08\u4f8b\u5982 TensorFlow\u3001PyTorch\uff09<\/strong><\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u6846\u67b6\u4f1a\u5b9a\u671f\u5c06\u6a21\u578b\u6743\u91cd\u3001\u4f18\u5316\u5668\u72b6\u6001\u548c\u5143\u6570\u636e\u4fdd\u5b58\u5230\u5b58\u50a8\u4e2d\u3002\u4ece\u7cfb\u7edf\u89c6\u89d2\u770b\uff0c\u8fd9\u4e9b\u662f\u65e0\u72b6\u6001\u7684\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u5982\u679c checkpoint \u4e0d\u591f\u9891\u7e41\uff0c\u53ef\u80fd\u4f1a\u4e22\u5931\u5927\u91cf\u5de5\u4f5c\u8fdb\u5ea6\u3002\u635f\u5931\u8303\u56f4\u53d6\u51b3\u4e8e\u624b\u52a8\u6216\u8ba1\u5212\u4fdd\u5b58\u95f4\u9694\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p><strong>\u7cfb\u7edf\u5c42\uff08\u4f8b\u5982 CRIUgpu\uff09<\/strong><\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u900f\u660e\u5730\u5bf9\u6574\u4e2a\u8fd0\u884c\u8fdb\u7a0b\u8fdb\u884c\u5feb\u7167\uff0c\u5305\u62ec GPU \u663e\u5b58\u548c\u5185\u6838\u72b6\u6001\uff0c\u53ef\u5b9e\u73b0\u6709\u72b6\u6001\u8fc1\u79fb\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u53ef\u5c06\u8fdb\u5ea6\u635f\u5931\u964d\u5230\u6700\u4f4e\u751a\u81f3\u5b8c\u5168\u6d88\u9664\u3002\u652f\u6301\u9ad8\u9891\u900f\u660e\u5feb\u7167\uff0c\u4ece\u800c\u5b9e\u73b0\u51e0\u4e4e\u65e0\u505c\u673a\u7684\u5728\u7ebf\u8fc1\u79fb\u3002<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p>\u7cfb\u7edf\u7ea7\u5feb\u7167\u673a\u5236\u5728\u62a2\u5360\u4e8b\u4ef6\u4e2d\u80fd\u591f\u7acb\u5373\u4ea7\u751f\u4e1a\u52a1\u4ef7\u503c\u3002\u7cfb\u7edf\u7ea7\u673a\u5236\u4f1a\u4fdd\u7559\u6d3b\u52a8\u5185\u5b58\u72b6\u6001\uff0c\u800c\u5e94\u7528\u5c42\u65b9\u6cd5\u5219\u4fdd\u5b58\u539f\u59cb\u6570\u636e\u6587\u4ef6\u3002\u7a33\u5065\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u80fd\u591f\u7ef4\u6301\u590d\u6742 AI \u6d41\u6c34\u7ebf\u4e2d\u7684\u5de5\u4f5c\u6d41\u8fd0\u884c\u3002<\/p>\n<h3>\u5e94\u7528\u8001\u5316\u89c4\u5219\u4e0e\u516c\u5e73\u5171\u4eab\u914d\u989d<\/h3>\n<p>\u6807\u51c6\u8c03\u5ea6\u89c4\u5219\u53ef\u80fd\u4f1a\u4f7f\u4f4e\u4f18\u5148\u7ea7\u961f\u5217\u9879\u65e0\u9650\u671f\u9965\u997f\u3002\u4f18\u5148\u7ea7\u8001\u5316\u4f1a\u968f\u7740\u7b49\u5f85\u65f6\u95f4\u7684\u589e\u52a0\u800c\u63d0\u5347\u4efb\u52a1\u5f97\u5206\u3002\u968f\u7740\u6392\u961f\u65f6\u95f4\u589e\u957f\uff0c\u4f4e\u4f18\u5148\u7ea7\u4efb\u52a1\u6700\u7ec8\u4e5f\u80fd\u83b7\u5f97\u6267\u884c\u673a\u4f1a\u3002\u667a\u80fd\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u80fd\u591f\u5728\u5373\u65f6\u6267\u884c\u9700\u6c42\u4e0e\u7528\u6237\u7ec4\u4e4b\u95f4\u7684\u516c\u5e73\u786c\u4ef6\u8bbf\u95ee\u4e4b\u95f4\u53d6\u5f97\u5e73\u8861\u3002<\/p>\n<p>\u57fa\u4e8e\u65f6\u95f4\u7684\u516c\u5e73\u5171\u4eab\u914d\u989d\u53ef\u9632\u6b62\u5355\u4e2a\u56e2\u961f\u5728\u4f01\u4e1a AI \u7cfb\u7edf\u4e2d\u957f\u671f\u5784\u65ad\u7b97\u529b\u6c60\u3002<\/p>\n<div class=\"qc-default-table-wrapper \">\n<table style=\"min-width: 75px\">\n<colgroup>\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\">\n<col style=\"min-width: 25px\"><\/colgroup>\n<tbody>\n<tr>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u65b9\u9762<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u6548\u679c\uff08\u65e0\u65f6\u95f4\u578b\u516c\u5e73\u5171\u4eab\uff09<\/p>\n<\/th>\n<th colspan=\"1\" rowspan=\"1\">\n<p>\u6548\u679c\uff08\u6709\u65f6\u95f4\u578b\u516c\u5e73\u5171\u4eab\uff09<\/p>\n<\/th>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p><strong>\u7a81\u53d1\u578b\u5de5\u4f5c\u8d1f\u8f7d\u7684\u4f5c\u4e1a\u541e\u5410<\/strong><\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u5982\u679c\u53e6\u4e00\u652f\u56e2\u961f\u5784\u65ad\u5171\u4eab\u8d44\u6e90\u6c60\uff0c\u8d85\u914d\u989d\u4f5c\u4e1a\u4f1a\u65e0\u9650\u671f\u7b49\u5f85\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u7531\u4e8e\u7cfb\u7edf\u4f1a\u8ddf\u8e2a\u5386\u53f2\u4f7f\u7528\u60c5\u51b5\uff0c\u7a81\u53d1\u4f5c\u4e1a\u4ecd\u80fd\u83b7\u5f97\u6267\u884c\u673a\u4f1a\u3002<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td colspan=\"1\" rowspan=\"1\">\n<p><strong>\u8d44\u6e90\u516c\u5e73\u8bbf\u95ee<\/strong><\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u6301\u7eed\u6027\u5de5\u4f5c\u8d1f\u8f7d\u7684\u56e2\u961f\u4f1a\u957f\u671f\u5784\u65ad\u8d85\u914d\u989d\u8d44\u6e90\u6c60\u3002<\/p>\n<\/td>\n<td colspan=\"1\" rowspan=\"1\">\n<p>\u8d44\u6e90\u4f1a\u5728\u4e0d\u540c\u56e2\u961f\u4e4b\u95f4\u52a8\u6001\u8f6e\u8f6c\uff0c\u4ece\u800c\u786e\u4fdd\u957f\u671f\u516c\u5e73\u5171\u4eab\u3002<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p>\u52a8\u6001\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u9632\u6b62\u91cd\u5ea6\u8d44\u6e90\u4f7f\u7528\u8005\u963b\u585e\u65b0\u8fdb\u5165\u7684\u4efb\u52a1\u3002\u6218\u7565\u6027\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u901a\u8fc7\u8ddf\u8e2a\u5386\u53f2\u4f7f\u7528\u503c\u6765\u63d0\u5347\u96c6\u7fa4\u6548\u7387\u3002\u81ea\u52a8\u5316\u4efb\u52a1\u4f18\u5148\u7ea7\u4f18\u5316\u961f\u5217\u6d41\u52a8\uff0c\u800c\u81ea\u9002\u5e94\u4efb\u52a1\u4f18\u5148\u7ea7\u5219\u4e3a\u6bcf\u4e2a\u56e2\u961f\u63d0\u4f9b\u53ef\u9760\u7684\u6838\u5fc3\u7b97\u529b\u786c\u4ef6\u8bbf\u95ee\u80fd\u529b\uff0c\u4ee5\u652f\u6301\u9ad8\u9700\u6c42 AI \u4efb\u52a1\u3002<\/p>\n<div class=\"qc-divider-wrapper\">\n<div class=\"qc-divider\" style=\"border-top-style: solid;width: 100%\"><\/div>\n<\/div>\n<p>\u4f60\u53ef\u4ee5\u901a\u8fc7\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\u6765\u6700\u5927\u5316\u786c\u4ef6\u6548\u7387\uff0c\u5e76\u52a0\u5feb\u6a21\u578b\u8fed\u4ee3\u901f\u5ea6\u3002\u8bf7\u5b8c\u6210\u4ee5\u4e0b\u8bbe\u7f6e\u6b65\u9aa4\uff0c\u4ee5\u6700\u5927\u5316\u96c6\u7fa4\u6548\u80fd\uff1a<\/p>\n<ol>\n<li>\n<p>\u8c03\u6574 CPU \u7ed1\u5b9a\uff0c\u4e3a\u6bcf\u4e2a AI \u4f5c\u4e1a\u9694\u79bb\u786c\u4ef6\u8d44\u6e90\u3002<\/p>\n<\/li>\n<li>\n<p>\u4f7f\u7528 Kueue \u90e8\u7f72\u961f\u5217\u89c4\u5219\uff0c\u4ee5\u7ba1\u7406\u5f85\u5904\u7406\u4efb\u52a1\u3002<\/p>\n<\/li>\n<li>\n<p>\u542f\u7528\u62a2\u5360\u4fdd\u62a4\u673a\u5236\uff0c\u4ee5\u4fdd\u62a4\u5de5\u4f5c\u6d41\u8fdb\u5ea6\u3002<\/p>\n<\/li>\n<\/ol>\n<p>\u5c06\u9759\u6001\u8c03\u5ea6\u5668\u961f\u5217\u4e0e AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u76f8\u7ed3\u5408\uff0c\u53ef\u4ee5\u63d0\u5347\u56e2\u961f\u751f\u4ea7\u6548\u7387\u3002\u9884\u6d4b\u5f0f AI \u6a21\u578b\u53ef\u7f29\u77ed\u5b8c\u6210\u65f6\u95f4\uff0c\u800c\u52a8\u6001 AI \u7b56\u7565\u5219\u8d1f\u8d23\u7ba1\u7406\u7d27\u6025 AI \u4efb\u52a1\u3002\u5148\u8fdb\u7684 AI \u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u80fd\u591f\u9632\u6b62\u590d\u6742 AI \u7cfb\u7edf\u4e2d\u7684\u8d44\u6e90\u9965\u997f\u3002\u5e94\u7528\u9884\u6d4b\u5f0f\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u91ca\u653e AI \u80fd\u529b\u5e76\u63d0\u5347\u7ec4\u7ec7\u6548\u76ca\u3002\u6218\u7565\u6027\u89c4\u5212\u53ef\u4f18\u5316 AI \u751f\u4ea7\u529b\uff0c\u4e3a\u6bcf\u4e00\u9879 AI \u57fa\u7840\u8bbe\u65bd\u5de5\u4f5c\u6d41\u5e26\u6765\u9ad8\u7b97\u529b\u6548\u7387\u3002<\/p>\n<h2>\u5e38\u89c1\u95ee\u9898<\/h2>\n<h3>CPU \u7ed1\u5b9a\u5982\u4f55\u63d0\u5347 AI \u4f5c\u4e1a\u6267\u884c\u6548\u7387\uff1f<\/h3>\n<p>\u4f60\u53ef\u4ee5\u4e3a\u6bcf\u4e2a AI \u4efb\u52a1\u9694\u79bb\u786c\u4ef6\u8d44\u6e90\uff0c\u4ece\u800c\u6062\u590d\u6027\u80fd\u3002CPU \u7ed1\u5b9a\u80fd\u591f\u5728\u5171\u4eab\u670d\u52a1\u5668\u4e0a\u5c06\u4e0a\u4e0b\u6587\u5207\u6362\u51cf\u5c11 96%\u3002\u8fd9\u79cd\u914d\u7f6e\u53ef\u4ee5\u907f\u514d AI \u6a21\u578b\u8bad\u7ec3\u4e2d\u7684\u5185\u5b58\u5ef6\u8fdf\uff0c\u5e76\u4fdd\u62a4\u6d41\u6c34\u7ebf\u6570\u636e\u3002<\/p>\n<h3>\u5728 Kubernetes \u4e2d\u90e8\u7f72 Kueue \u7684\u4e3b\u8981\u6536\u76ca\u662f\u4ec0\u4e48\uff1f<\/h3>\n<p>Kueue \u53ef\u5728\u591a\u79df\u6237\u96c6\u7fa4\u4e2d\u7ba1\u7406\u57fa\u4e8e\u961f\u5217\u7684\u51c6\u5165\u63a7\u5236\u548c gang \u8bed\u4e49\u3002\u4f60\u53ef\u4ee5\u5c06\u96c6\u7fa4 GPU \u5229\u7528\u7387\u4ece 30\u201350% \u63d0\u5347\u5230 80\u201395%\u3002Kueue \u786e\u4fdd\u9ad8\u4f18\u5148\u7ea7 AI \u5de5\u4f5c\u8d1f\u8f7d\u80fd\u591f\u7acb\u5373\u8bbf\u95ee GPU\uff0c\u800c\u4f4e\u5c42\u7ea7\u4efb\u52a1\u5219\u8ba9\u51fa\u5e26\u5bbd\uff0c\u4ee5\u4fdd\u62a4 AI \u96c6\u7fa4\u8d44\u6e90\u3002<\/p>\n<h3>\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5982\u4f55\u4f18\u5316\u961f\u5217\u8c03\u5ea6\uff1f<\/h3>\n<p>\u673a\u5668\u5b66\u4e60\u6a21\u578b\u4f1a\u5206\u6790\u5386\u53f2\u6570\u636e\uff0c\u81ea\u52a8\u9884\u6d4b AI \u4f5c\u4e1a\u53c2\u6570\u3002\u8fd9\u6837\u7684 AI \u9a71\u52a8\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u5c06\u5e73\u5747\u4f5c\u4e1a\u5b8c\u6210\u65f6\u95f4\u7f29\u77ed 25.8%\u3002\u9884\u6d4b\u5de5\u5177\u53ef\u4f18\u5316\u961f\u5217\u6d41\u52a8\u3001\u63d0\u5347\u7b97\u529b\u6548\u7387\u3001\u6539\u8fdb AI \u6d41\u6c34\u7ebf\uff0c\u5e76\u8ba9\u6bcf\u4e2a\u4efb\u52a1\u5339\u914d\u5230\u7406\u60f3\u786c\u4ef6\u3002<\/p>\n<h3>\u4e3a\u4ec0\u4e48\u8981\u901a\u8fc7 WebSockets \u89e3\u8026\u4f18\u5148\u7ea7\u5f15\u64ce\uff1f<\/h3>\n<p>\u72ec\u7acb\u7684\u540e\u7aef\u670d\u52a1\u80fd\u591f\u5728\u4e0d\u62d6\u6162\u4e3b AI \u96c6\u7fa4\u5de5\u4f5c\u6d41\u7684\u524d\u63d0\u4e0b\u5904\u7406\u8bc4\u5206\u903b\u8f91\u3002WebSockets \u4f1a\u6301\u7eed\u5411\u76d1\u63a7\u754c\u9762\u4f20\u9012\u5b9e\u65f6\u66f4\u65b0\u3002\u4f60\u53ef\u4ee5\u8ddf\u8e2a\u6d3b\u52a8\u961f\u5217\u72b6\u6001\uff0c\u5e76\u5728\u65b0\u4efb\u52a1\u8fdb\u5165\u4f01\u4e1a AI \u57fa\u7840\u8bbe\u65bd\u65f6\u52a8\u6001\u8c03\u6574\u4f18\u5148\u7ea7\u5206\u6570\u3002<\/p>\n<h3>\u62a2\u5360\u4fdd\u62a4\u673a\u5236\u5982\u4f55\u5728\u9a71\u9010\u65f6\u9632\u6b62\u6570\u636e\u4e22\u5931\uff1f<\/h3>\n<p>\u5404\u7c7b\u6846\u67b6\u901a\u8fc7 checkpoint \u673a\u5236\u5c06\u6570\u636e\u6307\u6807\u4fdd\u5b58\u5230\u7cfb\u7edf\u5b58\u50a8\u4e2d\u3002\u7cfb\u7edf\u7ea7\u5feb\u7167\u5219\u4f1a\u900f\u660e\u5730\u4fdd\u5b58 GPU \u663e\u5b58\u72b6\u6001\u3002\u8fd9\u6837\u4f60\u5c31\u80fd\u5728\u66f4\u9ad8\u4f18\u5148\u7ea7\u4efb\u52a1\u5360\u7528\u5f53\u524d\u52a0\u901f\u5668\u7684\u540c\u65f6\uff0c\u8ba9\u88ab\u9a71\u9010\u4f5c\u4e1a\u5728\u65b0\u7684 AI \u8282\u70b9\u4e0a\u5e73\u7a33\u6062\u590d\uff0c\u800c\u4e0d\u4f1a\u635f\u574f AI \u6a21\u578b\u6570\u636e\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>AI \u57fa\u7840\u8bbe\u65bd\u5de5\u7a0b\u5e08\u5fc5\u987b\u5728\u6ee1\u8db3\u5b9e\u65f6\u6267\u884c\u9700\u6c42\u7684\u540c\u65f6\u63d0\u5347\u6574\u4f53\u8fd0\u884c\u6548\u7387\u3002\u4f60\u53ef\u4ee5\u901a\u8fc7\u8c03\u6574\u5e95\u5c42\u7cfb\u7edf\u7684 CPU \u548c GPU \u8bbe\u7f6e\u3001\u5efa\u7acb\u4f18\u5148\u7ea7\u7c7b\u4ee5\u53ca\u5b9a\u4e49\u8c03\u5ea6\u5668\u961f\u5217\u89c4\u5219\uff0c\u6765\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\u3002 \u6070\u5f53\u7684 AI \u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u4ee5\u9632\u6b62\u591a\u8282\u70b9\u96c6\u7fa4\u4e2d\u7684 GPU \u8d44\u6e90\u9965\u997f\u3002\u5f53\u5de5\u4f5c\u8d1f\u8f7d\u5f00\u59cb\u6392\u961f\u65f6\uff0c\u6709\u6548\u7684\u4efb\u52a1\u7ba1\u7406\u4f1a\u5e94\u7528\u660e\u786e\u7684\u7b56\u7565\uff0c\u4ee5\u6700\u5927\u5316\u96c6\u7fa4\u751f\u4ea7\u529b\u548c\u7b97\u529b\u4ea7\u51fa\u3002\u8c03\u5ea6\u5668\u4f1a\u901a\u8fc7\u8bc4\u4f30\u4ee5\u4e0b\u5177\u4f53\u6807\u51c6\u6765\u51b3\u5b9a\u4efb\u52a1\u6267\u884c\u987a\u5e8f\uff1a \u5de5\u4f5c\u8d1f\u8f7d\u4f18\u5148\u7ea7 \u63d0\u4ea4\u65f6\u95f4\uff08FIFO\uff0c\u5148\u8fdb\u5148\u51fa\uff09 \u670d\u52a1\u7ea7\u522b\u534f\u8bae\uff08SLA\uff09 \u56e2\u961f\u4e4b\u95f4\u7684\u516c\u5e73\u5171\u4eab\u8c03\u5ea6 \u8d44\u6e90\u53ef\u7528\u6027 \u91c7\u7528 AI \u9a71\u52a8\u7684\u4efb\u52a1\u4f18\u5148\u7ea7\u7ba1\u7406\u53ef\u4ee5\u4f18\u5316\u73b0\u4ee3 AI \u7cfb\u7edf\u3002\u9ad8\u4f18\u5148\u7ea7 AI \u4efb\u52a1\u4f1a\u7acb\u5373\u83b7\u5f97\u786c\u4ef6\u8bbf\u95ee\u6743\uff0c\u800c\u4f4e\u4f18\u5148\u7ea7\u8bad\u7ec3\u4efb\u52a1\u5219\u4f1a\u4ee5\u5e73\u6ed1\u7684\u65b9\u5f0f\u8ba9\u51fa\u7b97\u529b\u8d44\u6e90\u3002 \u4e3a AI \u786c\u4ef6\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7 \u73b0\u4ee3\u9ad8\u6027\u80fd\u8ba1\u7b97\u96c6\u7fa4\u9700\u8981\u5bf9\u5e95\u5c42\u7cfb\u7edf\u8bbe\u7f6e\u8fdb\u884c\u76f4\u63a5\u63a7\u5236\u3002\u4f60\u53ef\u4ee5\u4e3a\u9ad8\u9700\u6c42\u786c\u4ef6\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ee5\u6d88\u9664\u6df1\u5ea6\u5b66\u4e60\u74f6\u9888\u3002\u6070\u5f53\u7684\u6267\u884c\u7ba1\u7406\u80fd\u591f\u4f18\u5316\u8d44\u6e90\u5206\u914d\u3001\u4fdd\u62a4\u7cfb\u7edf\u7a33\u5b9a\u6027\uff0c\u5e76\u63a8\u52a8\u6574\u4f53\u7ec4\u7ec7\u6548\u7387\u63d0\u5347\u3002 \u8c03\u6574\u7cfb\u7edf CPU \u548c GPU \u4f18\u5148\u7ea7 \u5185\u6838\u7ea7\u7cfb\u7edf\u8bbe\u7f6e\u4f1a\u76f4\u63a5\u5f71\u54cd\u8bad\u7ec3\u901f\u5ea6\u3002\u6807\u51c6\u64cd\u4f5c\u7cfb\u7edf\u8c03\u5ea6\u5668\u7ba1\u7406\u5904\u7406\u5668\u786c\u4ef6\u65f6\uff0c\u5e76\u4e0d\u4f1a\u8003\u8651 NVLink \u6216\u975e\u7edf\u4e00\u5185\u5b58\u8bbf\u95ee\uff08NUMA\uff09\u7b49\u7269\u7406\u4e92\u8fde\u62d3\u6251\u3002\u5ffd\u89c6\u62d3\u6251\u7ed3\u6784\u7684\u653e\u7f6e\u65b9\u5f0f\u4f1a\u8feb\u4f7f\u6570\u636e\u7a7f\u8d8a\u8f83\u8fdc\u7684\u5185\u5b58\u603b\u7ebf\uff0c\u4ece\u800c\u589e\u52a0\u4e3b\u673a\u5904\u7406\u5668\u4e0e\u52a0\u901f\u5668\u4e4b\u95f4\u7684\u901a\u4fe1\u5ef6\u8fdf\u3002\u901a\u8fc7\u5b9e\u65bd\u62d3\u6251\u611f\u77e5\u7684\u653e\u7f6e\u89c4\u5219\uff0c\u4f60\u53ef\u4ee5\u6d88\u9664\u8de8 NUMA \u5185\u5b58\u5e26\u6765\u7684\u6027\u80fd\u635f\u5931\u3002 \u8c03\u6574\u5e95\u5c42\u5904\u7406\u5668\u8bbe\u7f6e\u53ef\u4ee5\u63d0\u5347\u786c\u4ef6\u541e\u5410\u91cf\u5e76\u589e\u5f3a\u4efb\u52a1\u6267\u884c\u6548\u7387\u3002\u901a\u8fc7 nice \u547d\u4ee4\u8fdb\u884c\u6807\u51c6\u4f18\u5148\u7ea7\u8c03\u6574\uff0c\u5728\u8f7b\u5ea6\u5171\u4eab\u8d1f\u8f7d\u4e0b\u901a\u5e38\u53ea\u80fd\u5e26\u6765 5\u201310% \u7684\u9002\u5ea6\u63d0\u5347\u3002\u64cd\u4f5c\u7cfb\u7edf\u8c03\u5ea6\u5668\u5f15\u5165\u7684\u5f00\u9500\u4f1a\u6539\u53d8\u6267\u884c\u901f\u5ea6\uff1a\u5728\u5e72\u51c0\u73af\u5883\u4e2d\u5f71\u54cd\u7ea6\u4e3a 1.2%\uff0c\u800c\u5728\u7ec4\u5408\u8d1f\u8f7d\u4e0b\u53ef\u9ad8\u8fbe 20.5%\u3002\u901a\u8fc7 isolcpus \u4e0e\u4e2d\u65ad\u8bf7\u6c42\u4eb2\u548c\u6027\uff08IRQ affinity\uff09\u5b9e\u73b0\u5b8c\u6574 CPU \u9694\u79bb\uff0c\u53ef\u4e3a\u5bf9\u5ef6\u8fdf\u654f\u611f\u7684\u4efb\u52a1\u5e26\u6765 15\u201320% \u7684\u63d0\u5347\u3002\u6b64\u5916\uff0c\u5904\u7406\u5668\u7ed1\u5b9a\u53ef\u5728\u5171\u4eab\u670d\u52a1\u5668\u4e0a\u5c06\u4e0a\u4e0b\u6587\u5207\u6362\u51cf\u5c11 96%\u3002 \u4f18\u5316\u63aa\u65bd [&#8230;]<\/p>\n<p><a class=\"btn btn-secondary understrap-read-more-link\" href=\"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/\">Read More&#8230;<\/a><\/p>\n","protected":false},"author":12,"featured_media":34976,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[76],"tags":[],"class_list":["post-34979","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.2 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7<\/title>\n<meta name=\"description\" content=\"\u901a\u8fc7 CPU\/GPU \u7ed1\u5b9a\u3001Kubernetes PriorityClass \u4ee5\u53ca Slurm \u961f\u5217\u89c4\u5219\u6765\u914d\u7f6e AI \u8bad\u7ec3\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ece\u800c\u6700\u5927\u5316 GPU \u96c6\u7fa4\u541e\u5410\u91cf\u3002\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/\" \/>\n<meta property=\"og:site_name\" content=\"Varidata Limited\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-10T06:59:49+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-08-10T07:11:08+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/08\/Screenshot-2026-08-10-145837.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"613\" \/>\n\t<meta property=\"og:image:height\" content=\"331\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/\"},\"author\":\"Varidata\",\"headline\":\"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\",\"datePublished\":\"2026-08-10T06:59:49+00:00\",\"dateModified\":\"2026-08-10T07:11:08+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/\"},\"wordCount\":164,\"publisher\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.varidata.com\\\/wp-content\\\/uploads\\\/2026\\\/08\\\/Screenshot-2026-08-10-145837.jpg\",\"articleSection\":[\"Varidata \u5b98\u65b9\u535a\u5ba2\"],\"inLanguage\":\"zh-SC\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/\",\"url\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/\",\"name\":\"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.varidata.com\\\/wp-content\\\/uploads\\\/2026\\\/08\\\/Screenshot-2026-08-10-145837.jpg\",\"datePublished\":\"2026-08-10T06:59:49+00:00\",\"dateModified\":\"2026-08-10T07:11:08+00:00\",\"description\":\"\u901a\u8fc7 CPU\\\/GPU \u7ed1\u5b9a\u3001Kubernetes PriorityClass \u4ee5\u53ca Slurm \u961f\u5217\u89c4\u5219\u6765\u914d\u7f6e AI \u8bad\u7ec3\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ece\u800c\u6700\u5927\u5316 GPU \u96c6\u7fa4\u541e\u5410\u91cf\u3002\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#breadcrumb\"},\"inLanguage\":\"zh-SC\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-SC\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.varidata.com\\\/wp-content\\\/uploads\\\/2026\\\/08\\\/Screenshot-2026-08-10-145837.jpg\",\"contentUrl\":\"https:\\\/\\\/www.varidata.com\\\/wp-content\\\/uploads\\\/2026\\\/08\\\/Screenshot-2026-08-10-145837.jpg\",\"width\":613,\"height\":331,\"caption\":\"AI \u8bad\u7ec3\u4f5c\u4e1a\u4f18\u5148\u7ea7\u8c03\u5ea6\u6d41\u7a0b\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/blog\\\/how-to-configure-task-priorities-for-ai-training-jobs\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#website\",\"url\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/\",\"name\":\"Varidata Limited\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-SC\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#organization\",\"name\":\"Varidata\",\"url\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-SC\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.varidata.com\\\/wp-content\\\/uploads\\\/2021\\\/09\\\/varidata_logo_white_-748x480_hor_web-1.png\",\"contentUrl\":\"https:\\\/\\\/www.varidata.com\\\/wp-content\\\/uploads\\\/2021\\\/09\\\/varidata_logo_white_-748x480_hor_web-1.png\",\"width\":248,\"height\":94,\"caption\":\"Varidata\"},\"image\":{\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.varidata.com\\\/zh-cn\\\/#\\\/schema\\\/person\\\/8fe12b0dd65fd2ba7ccf0dea7d14de6c\",\"name\":\"Amy Yan\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-SC\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g\",\"caption\":\"Amy Yan\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7","description":"\u901a\u8fc7 CPU\/GPU \u7ed1\u5b9a\u3001Kubernetes PriorityClass \u4ee5\u53ca Slurm \u961f\u5217\u89c4\u5219\u6765\u914d\u7f6e AI \u8bad\u7ec3\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ece\u800c\u6700\u5927\u5316 GPU \u96c6\u7fa4\u541e\u5410\u91cf\u3002","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:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/","og_locale":"zh_CN","og_type":"article","og_title":"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7","og_url":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/","og_site_name":"Varidata Limited","article_published_time":"2026-08-10T06:59:49+00:00","article_modified_time":"2026-08-10T07:11:08+00:00","og_image":[{"width":613,"height":331,"url":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/08\/Screenshot-2026-08-10-145837.jpg","type":"image\/jpeg"}],"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#article","isPartOf":{"@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/"},"author":"Varidata","headline":"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7","datePublished":"2026-08-10T06:59:49+00:00","dateModified":"2026-08-10T07:11:08+00:00","mainEntityOfPage":{"@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/"},"wordCount":164,"publisher":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#organization"},"image":{"@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#primaryimage"},"thumbnailUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/08\/Screenshot-2026-08-10-145837.jpg","articleSection":["Varidata \u5b98\u65b9\u535a\u5ba2"],"inLanguage":"zh-SC"},{"@type":"WebPage","@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/","url":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/","name":"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7","isPartOf":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#primaryimage"},"image":{"@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#primaryimage"},"thumbnailUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/08\/Screenshot-2026-08-10-145837.jpg","datePublished":"2026-08-10T06:59:49+00:00","dateModified":"2026-08-10T07:11:08+00:00","description":"\u901a\u8fc7 CPU\/GPU \u7ed1\u5b9a\u3001Kubernetes PriorityClass \u4ee5\u53ca Slurm \u961f\u5217\u89c4\u5219\u6765\u914d\u7f6e AI \u8bad\u7ec3\u4efb\u52a1\u4f18\u5148\u7ea7\uff0c\u4ece\u800c\u6700\u5927\u5316 GPU \u96c6\u7fa4\u541e\u5410\u91cf\u3002","breadcrumb":{"@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#breadcrumb"},"inLanguage":"zh-SC","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/"]}]},{"@type":"ImageObject","inLanguage":"zh-SC","@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#primaryimage","url":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/08\/Screenshot-2026-08-10-145837.jpg","contentUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/08\/Screenshot-2026-08-10-145837.jpg","width":613,"height":331,"caption":"AI \u8bad\u7ec3\u4f5c\u4e1a\u4f18\u5148\u7ea7\u8c03\u5ea6\u6d41\u7a0b"},{"@type":"BreadcrumbList","@id":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-configure-task-priorities-for-ai-training-jobs\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.varidata.com\/zh-cn\/"},{"@type":"ListItem","position":2,"name":"\u5982\u4f55\u4e3a AI \u8bad\u7ec3\u4f5c\u4e1a\u914d\u7f6e\u4efb\u52a1\u4f18\u5148\u7ea7"}]},{"@type":"WebSite","@id":"https:\/\/www.varidata.com\/zh-cn\/#website","url":"https:\/\/www.varidata.com\/zh-cn\/","name":"Varidata Limited","description":"","publisher":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.varidata.com\/zh-cn\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-SC"},{"@type":"Organization","@id":"https:\/\/www.varidata.com\/zh-cn\/#organization","name":"Varidata","url":"https:\/\/www.varidata.com\/zh-cn\/","logo":{"@type":"ImageObject","inLanguage":"zh-SC","@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/logo\/image\/","url":"https:\/\/www.varidata.com\/wp-content\/uploads\/2021\/09\/varidata_logo_white_-748x480_hor_web-1.png","contentUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2021\/09\/varidata_logo_white_-748x480_hor_web-1.png","width":248,"height":94,"caption":"Varidata"},"image":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/person\/8fe12b0dd65fd2ba7ccf0dea7d14de6c","name":"Amy Yan","image":{"@type":"ImageObject","inLanguage":"zh-SC","@id":"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g","caption":"Amy Yan"}}]}},"_links":{"self":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts\/34979","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/users\/12"}],"replies":[{"embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/comments?post=34979"}],"version-history":[{"count":3,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts\/34979\/revisions"}],"predecessor-version":[{"id":34983,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts\/34979\/revisions\/34983"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/media\/34976"}],"wp:attachment":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/media?parent=34979"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/categories?post=34979"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/tags?post=34979"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}