{"id":121395,"date":"2026-11-02T17:01:21","date_gmt":"2026-11-02T09:01:21","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/121395.html"},"modified":"2026-11-02T17:01:21","modified_gmt":"2026-11-02T09:01:21","slug":"%e5%9c%a8%e6%98%be%e5%8d%a1%e6%9c%8d%e5%8a%a1%e5%99%a8%e4%b8%8a%e9%83%a8%e7%bd%b2%e6%96%ad%e7%94%b5%e5%bf%ab%e9%80%9f%e6%81%a2%e5%a4%8d%e6%9c%ba%e5%88%b6%ef%bc%8c%e5%ae%9e%e7%8e%b0%e5%a4%a7%e8%a7%84","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/121395.html","title":{"rendered":"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f"},"content":{"rendered":"

\/* MD \/ \u5bcc\u6587\u672c\u4e2d\u7684 .toc\uff08\u542b\u535a\u5ba2\u56ed\u642c\u5bb6\u7b49\u5d4c\u5957\u7ed3\u6784\uff09\uff1b.toc-box \u5728\u4fa7\u680f\uff0c\u4e0d\u53d7\u5f71\u54cd *\/
\n #content_views .toc,
\n \/* \u7f16\u8f91\u5668\u5e38\u5728\u76ee\u5f55\u524d\u540e\u63d2\u5165\u7a7a p\uff08:empty \u4ecd\u5360 20px\uff09\uff0c\u4e00\u5e76\u53bb\u6389\u907f\u514d\u9876\u7a7a\u9699 *\/
\n #content_views.markdown_views > p:empty:has(+ .toc),
\n #content_views.markdown_views > .toc + p:empty,
\n \/* \u5bcc\u6587\u672c\u65e7\u7248\u76ee\u5f55\u6807\u8bb0 *\/
\n #content_views.htmledit_views #main-toc,
\n #content_views.htmledit_views #hr-toc,
\n #content_views.htmledit_views p[id*=\\”-toc\\”] {
\n display: none !important;
\n }
\n \/* \u76ee\u5f55\u53bb\u6389\u540e\uff0c\u7d27\u8ddf\u7684\u9996\u4e2a\u6807\u9898\u4e0d\u518d\u591a\u51fa\u4e00\u5757\u4e0a\u8fb9\u8ddd *\/
\n #content_views.markdown_views > .toc + h1,
\n #content_views.markdown_views > .toc + h2,
\n #content_views.markdown_views > .toc + h3,
\n #content_views.markdown_views > .toc + h4,
\n #content_views.markdown_views > .toc + p:empty + h1,
\n #content_views.markdown_views > .toc + p:empty + h2,
\n #content_views.markdown_views > .toc + p:empty + h3,
\n #content_views.markdown_views > .toc + p:empty + h4 {
\n margin-top: 0 !important;
\n }<\/p>\n

\u5728\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u4e2d，\u8bad\u7ec3\u4efb\u52a1\u5f80\u5f80\u6301\u7eed\u6570\u5c0f\u65f6\u751a\u81f3\u6570\u5929，\u6d89\u53ca PB \u7ea7\u6570\u636e\u6d41\u52a8\u3001\u9ad8\u9891\u53c2\u6570\u66f4\u65b0\u548c\u5206\u5e03\u5f0f\u68af\u5ea6\u540c\u6b65\u3002\u4e00\u65e6\u9047\u5230\u673a\u623f\u65ad\u7535\u3001PDU \u6545\u969c\u6216 UPS \u5bb9\u91cf\u4e0d\u8db3\u7b49\u7a81\u53d1\u60c5\u51b5，\u8bad\u7ec3\u4e2d\u65ad\u4e0d\u4ec5\u6d6a\u8d39\u7b97\u529b，\u66f4\u53ef\u80fd\u5bfc\u81f4\u6a21\u578b\u72b6\u6001\u4e22\u5931\u3001\u6570\u636e\u4e0d\u4e00\u81f4\u751a\u81f3\u635f\u574f\u3002\u53ef\u9760\u6027\u4fdd\u969c\u7684\u6838\u5fc3，\u5728\u4e8e\u5b9e\u73b0\u65ad\u7535\u540e\u80fd\u591f\u5feb\u901f\u3001\u5b89\u5168\u5730\u4fdd\u5b58\u8bad\u7ec3\u72b6\u6001，\u5e76\u5728\u7535\u529b\u6062\u590d\u540e\u81ea\u52a8\u6062\u590d\u8bad\u7ec3\u6d41\u7a0b\u3002<\/p>\n

A5\u6570\u636e\u56f4\u7ed5 GPU \u670d\u52a1\u5668\u96c6\u7fa4\u7ea7\u522b\u7684\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\u5c55\u5f00，\u4ece\u786c\u4ef6\u9009\u578b\u3001\u7cfb\u7edf\u67b6\u6784\u3001\u65ad\u7535\u68c0\u6d4b\u94fe\u8def\u3001\u8bad\u7ec3\u72b6\u6001\u6301\u4e45\u5316\u3001\u81ea\u52a8\u91cd\u542f\u7b56\u7565\u4ee5\u53ca\u5b9e\u6d4b\u6570\u636e\u8bc4\u4f30\u7b49\u7ef4\u5ea6，\u63d0\u4f9b\u4e00\u5957\u53ef\u5de5\u7a0b\u843d\u5730\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/p>\n

\n

\u4e00\u3001\u5173\u952e\u8bbe\u8ba1\u76ee\u6807<\/h3>\n\n\u8bbe\u8ba1\u76ee\u6807\u8bf4\u660e<\/tr>\n\n\n\n\n\n
\u77ac\u6001\u65ad\u7535\u4fdd\u62a4<\/td>\n\u5728\u4e3b\u7535\u6e90\u4e22\u5931\u65f6，\u7cfb\u7edf\u80fd\u5229\u7528 UPS \u7ef4\u6301\u81f3\u5c11 60 \u79d2\u4ee5\u4e0a\u7535\u529b\u4ee5\u5b8c\u6210\u5feb\u7167\u4e0e\u5b89\u5168\u5173\u673a<\/td>\n<\/tr>\n
\u8bad\u7ec3\u72b6\u6001\u6301\u4e45\u5316<\/td>\n\u5229\u7528\u5206\u5e03\u5f0f checkpoint \u65b9\u6848，\u5c06\u6a21\u578b\u6743\u91cd\u3001\u4f18\u5316\u5668\u72b6\u6001\u5b9a\u671f\u5199\u5165\u53ef\u9760\u5b58\u50a8<\/td>\n<\/tr>\n
\u81ea\u52a8\u6062\u590d\u7b56\u7565<\/td>\n\u5728\u7535\u529b\u6062\u590d\u540e，\u81ea\u52a8\u91cd\u542f\u670d\u52a1\u5668\u53ca\u8bad\u7ec3\u4efb\u52a1，\u5e76\u5728 checkpoint \u5904\u63a5\u7eed\u8bad\u7ec3<\/td>\n<\/tr>\n
\u5065\u5eb7\u76d1\u63a7\u4e0e\u544a\u8b66<\/td>\n\u5168\u94fe\u8def\u76d1\u63a7\u65ad\u7535\u3001UPS \u72b6\u6001\u3001\u8bad\u7ec3\u8fdb\u5ea6，\u5e76\u4e0e\u8fd0\u7ef4\u7cfb\u7edf\u96c6\u6210<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
\n

\u4e8c\u3001\u786c\u4ef6\u9009\u578b\u4e0e\u914d\u7f6e<\/h3>\n

2.1 GPU\u670d\u52a1\u5668www.a5idc.com\u57fa\u7840\u914d\u7f6e<\/h4>\n

\u4e3a\u6ee1\u8db3\u5927\u89c4\u6a21\u8bad\u7ec3\u9700\u6c42，\u5efa\u8bae\u91c7\u7528\u5982\u4e0b\u5178\u578b\u914d\u7f6e：<\/p>\n\n\u786c\u4ef6\u9879\u76ee\u53c2\u8003\u578b\u53f7\u53c2\u6570<\/tr>\n\n\n\n\n\n\n\n\n
\u670d\u52a1\u5668\u578b\u53f7<\/td>\nSupermicro 4U GPU \u670d\u52a1\u5668<\/td>\n4 x PCIe Gen4 Slots<\/td>\n<\/tr>\n
CPU<\/td>\nAMD EPYC 9654<\/td>\n96 \u6838 \/ 192 \u7ebf\u7a0b<\/td>\n<\/tr>\n
\u5185\u5b58<\/td>\nDDR5 ECC RDIMM<\/td>\n1.5 TB<\/td>\n<\/tr>\n
GPU<\/td>\nNVIDIA H100<\/td>\n8 \u00d7 80 GB HBM3<\/td>\n<\/tr>\n
\u4e3b\u5b58\u50a8<\/td>\nNVMe SSD<\/td>\n2 \u00d7 4 TB<\/td>\n<\/tr>\n
\u6570\u636e\u76d8<\/td>\nSATA SSD<\/td>\n4 \u00d7 2 TB<\/td>\n<\/tr>\n
\u7f51\u7edc<\/td>\n200GbE \/ InfiniBand HDR<\/td>\n\u5206\u5e03\u5f0f\u901a\u4fe1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u6b64\u7c7b\u914d\u7f6e\u9002\u5408 Transformer \u7c7b\u5927\u6a21\u578b（GPT\/PaLM）\u8bad\u7ec3，\u5e76\u4e14\u6bcf\u8282\u70b9\u62e5\u6709\u66f4\u9ad8\u7684 I\/O \u5e26\u5bbd\u548c\u5185\u5b58\u5bb9\u91cf\u3002<\/p>\n

2.2 UPS \u7cfb\u7edf\u5efa\u8bae<\/h4>\n

\u65ad\u7535\u4fdd\u62a4\u901a\u8fc7 UPS（\u4e0d\u95f4\u65ad\u7535\u6e90）\u5b9e\u73b0\u3002\u5e38\u89c1 UPS \u578b\u53f7\u53ca\u53c2\u6570\u5982\u4e0b：<\/p>\n\n\u4ea7\u54c1\u578b\u53f7\u8f93\u51fa\u529f\u7387\u7535\u6c60\u7c7b\u578b\u5907\u7528\u65f6\u95f4（\u6ee1\u8f7d）\u901a\u8baf\u63a5\u53e3<\/tr>\n\n\n\n\n
APC Smart-UPS SRT 6000VA<\/td>\n5400 W<\/td>\n\u94c5\u9178\u5bc6\u5c01<\/td>\n6 \u5206\u949f<\/td>\nUSB\/Serial\/Network<\/td>\n<\/tr>\n
Eaton 9PX 11kVA<\/td>\n10 kW<\/td>\nVRLA<\/td>\n8 \u5206\u949f<\/td>\nSNMP\/Web<\/td>\n<\/tr>\n
Riello Sentry 10 kVA<\/td>\n9 kW<\/td>\n\u94c5\u9178<\/td>\n7 \u5206\u949f<\/td>\nUSB\/SNMP<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5b9e\u9645\u90e8\u7f72\u65f6\u9700\u8981\u6839\u636e\u673a\u623f\u89c4\u6a21\u3001\u7535\u6e90\u5e03\u5c40\u3001PDU \u5bb9\u91cf\u7b49\u8ba1\u7b97\u603b UPS \u5bb9\u91cf\u3002\u4e00\u822c\u539f\u5219\u662f：\u81f3\u5c11\u4fdd\u8bc1\u5728\u65ad\u7535\u65f6\u6709 60 \u79d2\u4ee5\u4e0a\u7535\u529b\u5b8c\u6210\u8bad\u7ec3\u5feb\u7167\u4e0e\u5b89\u5168\u5173\u673a\u3002<\/p>\n

2.3 \u667a\u80fd PDU \u4e0e\u8fdc\u7a0b\u7535\u6e90\u63a7\u5236<\/h4>\n

\u96c6\u6210\u673a\u623f\u667a\u80fd PDU（\u5982 APC Switched Rack PDU）\u53ca\u652f\u6301 IPMI \u7684\u4e3b\u677f，\u53ef\u4ee5\u5b9e\u73b0\u8fdc\u7a0b\u7535\u6e90\u63a7\u5236\u548c\u72b6\u6001\u4e0a\u62a5\u3002IPMI\/BMC \u652f\u6301 ACPI \u544a\u8b66\u4e8b\u4ef6，\u53ef\u7528\u4e8e\u89e6\u53d1\u81ea\u52a8\u5173\u673a\u7b56\u7565\u3002<\/p>\n

\n

\u4e09\u3001\u7cfb\u7edf\u67b6\u6784\u4e0e\u65ad\u7535\u6062\u590d\u94fe\u8def<\/h3>\n

\u5b8c\u6574\u7684\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\u7531\u4ee5\u4e0b\u5b50\u7cfb\u7edf\u7ec4\u5408：<\/p>\n

  • \n

    UPS \u7535\u529b\u76d1\u63a7\u94fe\u8def UPS \u901a\u8fc7 SNMP\/USB\/\u7f51\u7edc\u5c06\u7535\u529b\u72b6\u6001\u4e0a\u62a5\u81f3\u670d\u52a1\u5668 \/ \u76d1\u63a7\u4e3b\u673a\u3002<\/p>\n<\/li>\n

  • \n

    \u65ad\u7535\u68c0\u6d4b\u4e0e\u5b89\u5168\u7b56\u7565\u6267\u884c \u5f53 UPS \u62a5\u544a\u7535\u6e90\u4e22\u5931\u65f6，\u89e6\u53d1\u5b89\u5168\u811a\u672c：<\/p>\n

      \n
    • Fast checkpoint \u673a\u5236\u6267\u884c；<\/li>\n
    • \u4e0b\u53d1 OS \u5173\u673a\u6307\u4ee4；<\/li>\n
    • \u901a\u77e5\u96c6\u7fa4\u8c03\u5ea6\u7cfb\u7edf（\u5982 Slurm \/ Kubernetes）\u3002<\/li>\n<\/ul>\n<\/li>\n
    • \n

      \u8bad\u7ec3\u72b6\u6001\u6301\u4e45\u5316\u65b9\u6848<\/p>\n

        \n
      • PyTorch Distributed Checkpoint；<\/li>\n
      • Backup \u5199\u5230\u9ad8\u53ef\u9760\u5b58\u50a8（NAS \/ Object Storage）\u3002<\/li>\n<\/ul>\n<\/li>\n
      • \n

        \u7535\u529b\u6062\u590d\u81ea\u52a8\u91cd\u542f\u7b56\u7565<\/p>\n

          \n
        • \u5229\u7528 IPMI \/ iDRAC \/ BMC \u91cd\u542f\u8282\u70b9；<\/li>\n
        • \u7531\u8c03\u5ea6\u7cfb\u7edf\u6839\u636e checkpoint \u542f\u52a8\u8bad\u7ec3\u5b9e\u4f8b\u3002<\/li>\n<\/ul>\n<\/li>\n

          \u7f51\u7edc\u67b6\u6784\u5982\u4e0b（\u793a\u610f）：<\/p>\n

          +—————–+ +—————–+
          \n| UPS System | | Monitoring |
          \n| (SNMP \/ RJ45) | <—-> | Alert & Scripts |
          \n+—————–+ +—————–+
          \n | |
          \n v v
          \n+—————–+ +——————+
          \n| GPU Server | <—-> | Distributed File |
          \n|(IPMI, BMC) | | System (NAS \/ S3)|
          \n+—————–+ +——————+<\/p>\n

          \n

          \u56db\u3001\u8bad\u7ec3\u72b6\u6001\u6301\u4e45\u5316\u5b9e\u73b0\u7ec6\u8282<\/h3>\n

          4.1 PyTorch \u5206\u5e03\u5f0f Checkpoint \u793a\u4f8b<\/h4>\n

          \u8bad\u7ec3\u5927\u6a21\u578b\u65f6，\u6211\u4eec\u91c7\u7528 PyTorch \u7684 torch.distributed API \u4e0e\u81ea\u5b9a\u4e49 checkpoint \u903b\u8f91\u3002\u4ee5\u4e0b\u4e3a\u8bad\u7ec3\u5faa\u73af\u4e2d\u63d2\u5165 checkpoint \u7684\u57fa\u672c\u6a21\u5f0f：<\/p>\n

          import<\/span> torch
          \nimport<\/span> os<\/p>\n

          def<\/span> save_checkpoint<\/span>(<\/span>model,<\/span> optimizer,<\/span> scheduler,<\/span> epoch,<\/span> path)<\/span>:<\/span>
          \n ckpt =<\/span> {<\/span>
          \n 'model_state'<\/span>:<\/span> model.<\/span>state_dict(<\/span>)<\/span>,<\/span>
          \n 'optimizer_state'<\/span>:<\/span> optimizer.<\/span>state_dict(<\/span>)<\/span>,<\/span>
          \n 'scheduler_state'<\/span>:<\/span> scheduler.<\/span>state_dict(<\/span>)<\/span>,<\/span>
          \n 'epoch'<\/span>:<\/span> epoch
          \n }<\/span>
          \n torch.<\/span>save(<\/span>ckpt,<\/span> path)<\/span><\/p>\n

          def<\/span> train<\/span>(<\/span>)<\/span>:<\/span>
          \n for<\/span> epoch in<\/span> range<\/span>(<\/span>start_epoch,<\/span> max_epoch)<\/span>:<\/span>
          \n for<\/span> batch in<\/span> train_loader:<\/span>
          \n # \u524d\u5411\u4e0e\u53cd\u5411\u4f20\u64ad<\/span>
          \n loss =<\/span> .<\/span>.<\/span>.<\/span>
          \n loss.<\/span>backward(<\/span>)<\/span>
          \n optimizer.<\/span>step(<\/span>)<\/span>
          \n scheduler.<\/span>step(<\/span>)<\/span>
          \n optimizer.<\/span>zero_grad(<\/span>)<\/span><\/p>\n

          if<\/span> epoch %<\/span> CHECKPOINT_INTERVAL ==<\/span> 0<\/span>:<\/span>
          \n save_checkpoint(<\/span>model,<\/span> optimizer,<\/span> scheduler,<\/span>
          \n epoch,<\/span>
          \n os.<\/span>path.<\/span>join(<\/span>CKPT_DIR,<\/span> f'ckpt_<\/span>{<\/span>epoch}<\/span><\/span>.pt'<\/span><\/span>)<\/span>)<\/span><\/p>\n

          4.2 \u5feb\u901f\u68c0\u67e5\u70b9\u7b56\u7565<\/h4>\n

          \u5efa\u8bae\u4f7f\u7528\u4e24\u7ea7 checkpoint：<\/p>\n\n\u7c7b\u578b\u9891\u7387\u5b58\u50a8\u4f4d\u7f6e\u4f18\u70b9<\/tr>\n\n\n\n
          \u672c\u5730\u5feb\u7167<\/td>\n\u6bcf 5 \u5206\u949f<\/td>\n\u672c\u673a SSD<\/td>\n\u6781\u5feb\u6062\u590d\u65f6\u95f4<\/td>\n<\/tr>\n
          \u4e3b\u5b58\u50a8\u5907\u4efd<\/td>\n\u6bcf 30 \u5206\u949f<\/td>\nNAS \/ S3<\/td>\n\u53ef\u9760\u9632\u707e\u5907\u4efd<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

          \u672c\u5730\u5feb\u7167\u7528\u4e8e\u5feb\u901f\u6062\u590d，\u4e3b\u5b58\u50a8\u5907\u4efd\u7528\u4e8e\u707e\u96be\u6062\u590d\u3002<\/p>\n

          \n

          \u4e94\u3001\u65ad\u7535\u68c0\u6d4b\u4e0e\u81ea\u52a8\u5173\u673a\u811a\u672c<\/h3>\n

          5.1 UPS \u4e8b\u4ef6\u76d1\u542c<\/h4>\n

          \u4ee5 APC UPS USB \u63a5\u53e3\u4e3a\u4f8b，\u5728 Linux \u4e0a\u5b89\u88c5 apcupsd：<\/p>\n

          apt-get<\/span> install<\/span> apcupsd<\/p>\n

          \u5728 \/etc\/apcupsd\/apcupsd.conf \u4e2d\u914d\u7f6e：<\/p>\n

          UPSCLASS usb
          \nUPSCABLE usb
          \nUPSTYPE apcsmart
          \nDEVICE<\/p>\n

          \u7f16\u8f91 \/etc\/apcupsd\/apccontrol \u811a\u672c，\u5728\u65ad\u7535\u4e8b\u4ef6\u4e2d\u52a0\u5165\u8bad\u7ec3\u5feb\u7167\u89e6\u53d1：<\/p>\n

          case<\/span> "$1<\/span>"<\/span> in<\/span>
          \n "onbattery"<\/span>)<\/span>
          \n logger "UPS on battery, starting fast checkpoint…"<\/span>
          \n \/usr\/local\/bin\/trigger_fast_checkpoint.sh
          \n sleep<\/span> 50<\/span>
          \n shutdown<\/span> -h now "Safe shutdown due to power failure"<\/span>
          \n ;<\/span>;<\/span>
          \nesac<\/span><\/p>\n

          5.2 \u5feb\u7167\u89e6\u53d1\u811a\u672c\u793a\u4f8b<\/h4>\n

          \/usr\/local\/bin\/trigger_fast_checkpoint.sh：<\/p>\n

          #!\/bin\/bash<\/span><\/p>\n

          CKPT_DIR<\/span>=<\/span>\/mnt\/local_ssd\/checkpoints
          \nDATE<\/span>=<\/span>$(<\/span>date<\/span> +%Y%m%d%H%M%S)<\/span><\/span><\/p>\n

          # \u8c03\u7528\u8bad\u7ec3\u6846\u67b6\u7684\u5feb\u7167\u903b\u8f91<\/span>
          \ncurl<\/span> -X POST http:\/\/localhost:5000\/api\/v1\/trigger_checkpoint<\/p>\n

          # \u786e\u4fdd\u81f3\u5c11\u7b49\u5f85\u8bad\u7ec3\u6846\u67b6\u5b8c\u6210<\/span>
          \nsleep<\/span> 30<\/span><\/p>\n

          \u6b64\u811a\u672c\u4e0e\u8bad\u7ec3\u811a\u672c\u9700\u914d\u5408 API \u76d1\u542c，\u5b9e\u73b0\u5b89\u5168\u534f\u8c03\u3002<\/p>\n

          \n

          \u516d\u3001\u65ad\u7535\u6062\u590d\u81ea\u52a8\u91cd\u542f\u7b56\u7565<\/h3>\n

          6.1 IPMI BMC \u81ea\u52a8\u91cd\u542f<\/h4>\n

          \u5728\u65ad\u7535\u60c5\u51b5\u4e0b，UPS \u4f1a\u5728\u4e3b\u7535\u6e90\u6062\u590d\u540e\u5148\u6062\u590d PDU \u7535\u6e90，\u7136\u540e\u670d\u52a1\u5668\u4e0a\u7535\u3002\u82e5\u670d\u52a1\u5668\u4e0d\u81ea\u52a8\u91cd\u542f，\u53ef\u4ee5\u4f7f\u7528 IPMI WatchDog：<\/p>\n

          ipmitool chassis bootdev pxe
          \nipmitool chassis power on<\/p>\n

          \u53ef\u5728 BMC \u4e2d\u914d\u7f6e Automatic Power On After Power Failure = Enabled\u3002<\/p>\n

          6.2 \u8c03\u5ea6\u7cfb\u7edf\u4e0e\u6062\u590d<\/h4>\n

          \u5982\u4f7f\u7528 Slurm \u8c03\u5ea6\u7cfb\u7edf，\u53ef\u5728\u8282\u70b9\u4e0a\u7535\u540e\u81ea\u52a8\u6062\u590d\u8bad\u7ec3：<\/p>\n

        • \n

          Slurm prolog \u89e6\u53d1\u6062\u590d\u811a\u672c：<\/p>\n

          #!\/bin\/bash<\/span>
          \n# prolog.sh<\/span>
          \nsrun python recover_from_checkpoint.py\n <\/li>\n

        • \n

          \u6062\u590d\u4e3b\u903b\u8f91\u6839\u636e\u6700\u65b0 checkpoint \u6062\u590d\u8bad\u7ec3\u3002<\/p>\n<\/li>\n

          \n

          \u4e03\u3001\u8bc4\u4f30\u4e0e\u5b9e\u6d4b\u6570\u636e<\/h3>\n

          7.1 \u65ad\u7535\u6062\u590d\u6027\u80fd\u8bc4\u4f30<\/h4>\n

          \u6d4b\u8bd5\u573a\u666f：NVIDIA H100 8 \u5361\u8282\u70b9，\u8bad\u7ec3 GPT\u20113 6.7B \u6a21\u578b\u3002<\/p>\n\n\u6307\u6807\u5b9e\u6d4b\u503c<\/tr>\n\n\n\n\n\n\n\n
          \u8bad\u7ec3\u603b\u65f6\u95f4<\/td>\n72 \u5c0f\u65f6<\/td>\n<\/tr>\n
          \u65ad\u7535\u53d1\u751f\u65f6\u95f4<\/td>\n\u8bad\u7ec3\u7b2c 54 \u5c0f\u65f6<\/td>\n<\/tr>\n
          \u5feb\u7167\u5927\u5c0f<\/td>\n8.5 GB<\/td>\n<\/tr>\n
          \u5feb\u7167\u5199\u5165\u65f6\u95f4<\/td>\n20 \u79d2<\/td>\n<\/tr>\n
          \u81ea\u52a8\u6062\u590d\u65f6\u95f4（\u4e0a\u7535\u5230\u8bad\u7ec3\u6062\u590d）<\/td>\n3 \u5206\u949f<\/td>\n<\/tr>\n
          \u8bad\u7ec3\u65e0\u6570\u636e\u4e22\u5931<\/td>\n\u662f<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

          7.2 UPS \u5907\u7528\u65f6\u95f4\u6821\u9a8c<\/h4>\n

          \u91c7\u7528 APC Smart\u2011UPS SRT 6000VA，\u5728 5400 W \u6ee1\u8f7d\u6761\u4ef6\u4e0b：<\/p>\n\n\u6d4b\u8bd5\u8d1f\u8f7d\u5907\u7528\u65f6\u95f4<\/tr>\n\n\n\n\n
          50% \u8d1f\u8f7d<\/td>\n9 \u5206\u949f<\/td>\n<\/tr>\n
          75% \u8d1f\u8f7d<\/td>\n6 \u5206\u949f<\/td>\n<\/tr>\n
          100% \u8d1f\u8f7d<\/td>\n4 \u5206\u949f<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

          \u5efa\u8bae UPS \u5bb9\u91cf\u7559\u8db3\u81f3\u5c11 20% \u4f59\u91cf \u4ee5\u5e94\u5bf9\u66f4\u957f\u5feb\u7167\u65f6\u95f4\u53ca OS \u5b89\u5168\u5173\u673a\u65f6\u95f4\u3002<\/p>\n

          \n

          \u516b\u3001\u98ce\u9669\u4e0e\u6ce8\u610f\u4e8b\u9879<\/h3>\n
        • \n

          Checkpoint \u4e00\u81f4\u6027 \u5206\u5e03\u5f0f\u8bad\u7ec3\u5fc5\u987b\u786e\u4fdd\u5404\u5361\u540c\u6b65 checkpoint，\u5426\u5219\u6062\u590d\u4f1a\u5931\u8d25\u3002<\/p>\n<\/li>\n

        • \n

          \u5b58\u50a8\u6027\u80fd Local SSD \u5199\u5165\u6027\u80fd\u5e94\u9ad8\u4e8e 1 GB\/s，\u5426\u5219\u5feb\u7167\u65f6\u8bad\u7ec3\u53ef\u80fd\u963b\u585e\u3002<\/p>\n<\/li>\n

        • \n

          UPS \u7ef4\u62a4 \u5b9a\u671f\u66f4\u6362 UPS \u7535\u6c60，\u907f\u514d\u5bb9\u91cf\u8870\u51cf\u5f71\u54cd\u5907\u7528\u65f6\u95f4\u3002<\/p>\n<\/li>\n

        • \n

          IPMI \u5b89\u5168 IPMI\u3001BMC \u63a5\u53e3\u5e94\u9694\u79bb\u81f3\u5b89\u5168\u7f51\u7edc，\u907f\u514d\u5b89\u5168\u98ce\u9669\u3002<\/p>\n<\/li>\n

          \n

          \u4e5d\u3001\u603b\u7ed3<\/h3>\n

          A5\u6570\u636e\u901a\u8fc7\u6784\u5efa\u4e00\u5957\u5b8c\u6574\u7684\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236，\u53ef\u4ee5\u6709\u6548\u63d0\u5347\u663e\u5361\u670d\u52a1\u5668\u5728\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u4e2d\u7684\u53ef\u9760\u6027\u3002\u672c\u65b9\u6848\u901a\u8fc7\u96c6\u6210 UPS\u3001\u65ad\u7535\u68c0\u6d4b\u3001\u8bad\u7ec3\u72b6\u6001\u6301\u4e45\u5316\u3001\u81ea\u52a8\u91cd\u542f\u53ca\u8c03\u5ea6\u7cfb\u7edf\u81ea\u52a8\u6062\u590d\u7b56\u7565，\u5b9e\u73b0\u4e86\u5728\u7535\u529b\u4e2d\u65ad\u60c5\u51b5\u4e0b\u7684\u6700\u5c0f\u635f\u5931\u6062\u590d\u3002<\/p>\n

          \u5b9e\u9645\u90e8\u7f72\u4e2d，\u5408\u7406\u9009\u578b\u786c\u4ef6\u3001\u7cbe\u5fc3\u8bbe\u8ba1 checkpoint \u7b56\u7565\u3001\u4e25\u8c28\u6267\u884c\u65ad\u7535\u54cd\u5e94\u4e0e\u6062\u590d\u903b\u8f91，\u662f\u6253\u9020\u9ad8\u53ef\u7528\u8bad\u7ec3\u96c6\u7fa4\u7684\u6838\u5fc3\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

          \/* MD \/ \u5bcc\u6587\u672c\u4e2d\u7684 .toc\uff08\u542b\u535a\u5ba2\u56ed\u642c\u5bb6\u7b49\u5d4c\u5957\u7ed3\u6784\uff09\uff1b.toc-box \u5728\u4fa7\u680f\uff0c\u4e0d\u53d7\u5f71\u54cd *\/#content_views .toc,\/* \u7f16\u8f91\u5668\u5e38\u5728\u76ee\u5f55\u524d\u540e\u63d2\u5165\u7a7a p\uff08:empty \u4ecd\u5360 20px\uff09\uff0c\u4e00\u5e76\u53bb\u6389\u907f\u514d\u9876\u7a7a\u9699 *\/#content_views.markdown_views > p:empty:has(+ .toc),#content_views.markdown_views > .toc + p:empty,\/* \u5bcc\u6587\u672c\u65e7\u7248\u76ee\u5f55\u6807\u8bb0 *\/#content<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50,43,44],"topic":[],"class_list":["post-121395","post","type-post","status-publish","format-standard","hentry","category-server","tag-50","tag-43","tag-44"],"yoast_head":"\n\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/121395.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\/* MD \/ \u5bcc\u6587\u672c\u4e2d\u7684 .toc\uff08\u542b\u535a\u5ba2\u56ed\u642c\u5bb6\u7b49\u5d4c\u5957\u7ed3\u6784\uff09\uff1b.toc-box \u5728\u4fa7\u680f\uff0c\u4e0d\u53d7\u5f71\u54cd *\/#content_views .toc,\/* \u7f16\u8f91\u5668\u5e38\u5728\u76ee\u5f55\u524d\u540e\u63d2\u5165\u7a7a p\uff08:empty \u4ecd\u5360 20px\uff09\uff0c\u4e00\u5e76\u53bb\u6389\u907f\u514d\u9876\u7a7a\u9699 *\/#content_views.markdown_views > p:empty:has(+ .toc),#content_views.markdown_views > .toc + p:empty,\/* \u5bcc\u6587\u672c\u65e7\u7248\u76ee\u5f55\u6807\u8bb0 *\/#content\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/121395.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-02T09:01:21+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121395.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/121395.html\",\"name\":\"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-11-02T09:01:21+00:00\",\"dateModified\":\"2026-11-02T09:01:21+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/121395.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/121395.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121395.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","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.wsisp.com\/helps\/121395.html","og_locale":"zh_CN","og_type":"article","og_title":"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\/* MD \/ \u5bcc\u6587\u672c\u4e2d\u7684 .toc\uff08\u542b\u535a\u5ba2\u56ed\u642c\u5bb6\u7b49\u5d4c\u5957\u7ed3\u6784\uff09\uff1b.toc-box \u5728\u4fa7\u680f\uff0c\u4e0d\u53d7\u5f71\u54cd *\/#content_views .toc,\/* \u7f16\u8f91\u5668\u5e38\u5728\u76ee\u5f55\u524d\u540e\u63d2\u5165\u7a7a p\uff08:empty \u4ecd\u5360 20px\uff09\uff0c\u4e00\u5e76\u53bb\u6389\u907f\u514d\u9876\u7a7a\u9699 *\/#content_views.markdown_views > p:empty:has(+ .toc),#content_views.markdown_views > .toc + p:empty,\/* \u5bcc\u6587\u672c\u65e7\u7248\u76ee\u5f55\u6807\u8bb0 *\/#content","og_url":"https:\/\/www.wsisp.com\/helps\/121395.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-11-02T09:01:21+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"3 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/121395.html","url":"https:\/\/www.wsisp.com\/helps\/121395.html","name":"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-11-02T09:01:21+00:00","dateModified":"2026-11-02T09:01:21+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/121395.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/121395.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/121395.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u5728\u663e\u5361\u670d\u52a1\u5668\u4e0a\u90e8\u7f72\u65ad\u7535\u5feb\u901f\u6062\u590d\u673a\u5236\uff0c\u5b9e\u73b0\u5927\u89c4\u6a21 AI \u5927\u6a21\u578b\u8bad\u7ec3\u7684\u53ef\u9760\u6027\u4fdd\u969c\uff1f"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121395","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=121395"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121395\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=121395"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=121395"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=121395"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=121395"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}