{"id":2727,"date":"2026-03-06T14:20:53","date_gmt":"2026-03-06T06:20:53","guid":{"rendered":"https:\/\/www.starverse-ai.com\/guide\/archives\/2727"},"modified":"2026-03-06T14:20:53","modified_gmt":"2026-03-06T06:20:53","slug":"%e7%a7%91%e7%a0%94%e4%ba%ba%e9%80%9f%e7%9c%8b%ef%bc%81%e6%98%9f%e5%ae%87%e6%99%ba%e7%ae%97%e6%94%af%e6%8c%81slurmapptainer%ef%bc%8c3000%e5%bc%a0gpu%e9%9a%8f%e4%bd%a0%e7%94%a8%ef%bc%8c%e8%ae%ba","status":"publish","type":"post","link":"https:\/\/www.starverse-ai.com\/guide\/archives\/2727","title":{"rendered":"\u79d1\u7814\u4eba\u901f\u770b\uff01\u661f\u5b87\u667a\u7b97\u652f\u6301Slurm+Apptainer\uff0c3000\u5f20GPU\u968f\u4f60\u7528\uff0c\u8bba\u6587DDL\u4e0d\u518d\u614c"},"content":{"rendered":"<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.starverse-ai.com\/guide\/wp-content\/uploads\/2026\/03\/1772778053_1ffb88.png\" alt=\"\u79d1\u7814\u4eba\u901f\u770b\uff01\u661f\u5b87\u667a\u7b97\u652f\u6301Slurm+Apptainer\uff0c3000\u5f20GPU\u968f\u4f60\u7528\uff0c\u8bba\u6587DDL\u4e0d\u518d\u614c\" style=\"display:block; margin:10px auto; max-width:100%; height:auto;\" \/><\/figure>\n<blockquote>\n<p>\u201cAAAI \u622a\u7a3f\u524d 72 \u5c0f\u65f6\uff0c\u6211\u7684 ablation \u5b9e\u9a8c\u8fd8\u5728\u6392\u961f\uff0c\u9ad8\u6821\u96c6\u7fa4 3 \u5929\u6ca1\u6392\u5230\u4e00\u5f20\u5361\u3002\u201d<br \/>\n\u2014\u2014\u67d0 985 CV \u5b9e\u9a8c\u5ba4\u535a\u58eb\uff0c\u5c0f\u7ea2\u4e66\u70ed\u5e16 2.3 \u4e07\u8d5e<\/p>\n<\/blockquote>\n<p>\u8fd9\u4e0d\u662f\u6bb5\u5b50\uff0c\u800c\u662f 2024 \u5e74\u79d1\u7814\u5b63\u7684\u5e38\u6001\u3002CVPR\u3001ICML\u3001NeurIPS \u8f6e\u756a\u4e0a\u6620\uff0cGPU \u5374\u6bd4\u5ba1\u7a3f\u610f\u89c1\u8fd8\u7a00\u7f3a\u3002\u773c\u770b DDL \u903c\u8fd1\uff0c\u4ee3\u7801\u65e9\u5df2 push\uff0c\u5b9e\u9a8c\u5374\u5361\u5728\u201cPENDING\u201d\u72b6\u6001\uff0c\u8001\u677f\u8fd8\u5728\u7fa4\u91cc\u50ac\u8fdb\u5ea6\u2014\u2014\u7126\u8651\u503c\u76f4\u63a5\u62c9\u6ee1\u3002<\/p>\n<h2>\u75db\u70b9\uff1a\u6392\u961f 3 \u5929=\u9519\u8fc7\u4e00\u7bc7\u9876\u4f1a<\/h2>\n<p>\u9ad8\u6821\u5171\u4eab\u96c6\u7fa4\u7684 Slurm \u961f\u5217\uff0c\u4f18\u5148\u7ea7\u88ab\u5e08\u5144\u7684 512 \u5361\u5927\u4f5c\u4e1a\u9738\u5360\uff1b\u6821\u5916\u4e91\u5382\u5546\u867d\u80fd\u79d2\u5f00\u5b9e\u4f8b\uff0c\u5374\u53c8\u4e0d\u652f\u6301 Apptainer\uff0c\u5b9e\u9a8c\u5ba4\u7956\u4f20 Docker \u955c\u50cf\u5f97\u91cd\u642d\uff0c\u73af\u5883\u590d\u73b0\u52a8\u8f84\u4e00\u5929\u3002\u66f4\u5c34\u5c2c\u7684\u662f\uff0c\u4e3b\u6d41 <a href=\"https:\/\/www.starverse-ai.com\">GPU\u670d\u52a1\u5668\u79df\u7528<\/a> \u5e73\u53f0\u5bf9\u6559\u80b2\u7528\u6237\u6309\u91cf\u8ba1\u8d39\u201c\u79d2\u7ea7\u6263\u8d39\u201d\uff0c\u57fa\u91d1\u7ecf\u8d39\u8fd8\u6ca1\u5230\u8d26\uff0c\u4e2a\u4eba\u4fe1\u7528\u5361\u5df2\u5237\u7206\u3002<\/p>\n<h2>\u65b9\u6848\uff1a\u661f\u5b87\u667a\u7b97\u628a Slurm+Apptainer \u642c\u8fdb\u4e91\u7aef<\/h2>\n<p>\u661f\u5b87\u667a\u7b97\u56e2\u961f\u672c\u8eab\u5c31\u662f\u4e00\u7fa4\u4ece\u9ad8\u6821 AI Lab \u51fa\u6765\u7684\u201c\u524d\u535a\u58eb\u201d\uff0c\u6df1\u77e5\u201c\u6392\u961f+\u73af\u5883+\u9884\u7b97\u201d\u4e09\u5ea7\u5927\u5c71\u3002\u4e8e\u662f\u4ed6\u4eec\u628a<strong>\u539f\u751f Slurm \u8c03\u5ea6\u7cfb\u7edf<\/strong>\u76f4\u63a5\u505a\u8fdb\u4e86 <a href=\"https:\/\/www.starverse-ai.com\">GPU\u4e91\u4e3b\u673a<\/a> \u5e95\u5ea7\uff0c\u5e76\u9ed8\u8ba4\u6302\u8f7d Apptainer\/Singularity \u6a21\u5757\uff0c\u5b9e\u73b0\u4e0e\u6821\u5185\u96c6\u7fa4\u5b8c\u5168\u4e00\u81f4\u7684\u63d0\u4ea4\u6d41\u7a0b\uff1a<\/p>\n<ul>\n<li>\u672c\u5730 Dockerfile \u65e0\u9700\u6539\u9020\uff0c<code>apptainer build<\/code> \u4e00\u952e\u8f6c\u6362 <code>.sif<\/code>  <\/li>\n<li><code>sbatch<\/code>\u3001<code>srun<\/code>\u3001<code>squeue<\/code> \u547d\u4ee4 100 % \u517c\u5bb9\uff0c<code>.sh<\/code> \u811a\u672c\u96f6\u4fee\u6539  <\/li>\n<li>\u961f\u5217\u7b56\u7565\u900f\u660e\uff0c\u7528\u6237\u53ef\u81ea\u9009\u300c\u72ec\u5360 8\u00d7A100\u300d\u6216\u300c\u62a2\u5360\u5f0f 1\u00d7RTX 4090\u300d\u4e24\u79cd\u6a21\u5f0f  <\/li>\n<\/ul>\n<p>\u6362\u53e5\u8bdd\u8bf4\uff0c\u4f60\u4f1a\u7528\u5b66\u6821\u96c6\u7fa4\uff0c\u5c31\u4f1a\u7528\u661f\u5b87\u667a\u7b97\u2014\u2014<strong>\u96f6\u5b66\u4e60\u6210\u672c<\/strong>\uff0c\u5b9e\u9a8c\u811a\u672c\u76f4\u63a5\u590d\u5236\u7c98\u8d34\u3002<\/p>\n<h2>\u6d41\u7a0b\uff1a3000 \u5f20 GPU \u968f\u4f60\u7528\uff0c\u4ece Dockerfile \u5230\u8dd1\u5b9e\u9a8c\u53ea\u7528 3 \u6b65<\/h2>\n<ol>\n<li>\n<p>\u672c\u5730\u6784\u5efa\u5bb9\u5668<br \/>\n<code>bash<br \/>\n   apptainer build vit_imagenet.sif docker:\/\/myrepo\/vit:1.2<\/code><\/p>\n<\/li>\n<li>\n<p>\u4e0a\u4f20\u955c\u50cf\u4e0e\u4ee3\u7801<br \/>\n   \u661f\u5b87\u667a\u7b97\u63d0\u4f9b 10 Gbps \u4e0a\u884c\u5e26\u5bbd\uff0c\u652f\u6301 <code>rsync<\/code>\u3001<code>scp<\/code> \u76f4\u8fde\uff0c\u5185\u7f6e <code>~\/scatch<\/code> \u4e34\u65f6\u7a7a\u95f4 2 TB\uff0c\u514d\u8d39\u3002<\/p>\n<\/li>\n<li>\n<p><code>sbatch<\/code> \u63d0\u4ea4<br \/>\n<code>bash<br \/>\n   #!\/bin\/bash<br \/>\n   #SBATCH --job-name=vit120<br \/>\n   #SBATCH --nodes=1<br \/>\n   #SBATCH --ntasks-per-node=8<br \/>\n   #SBATCH --gres=gpu:8<br \/>\n   #SBATCH --partition=a100<br \/>\n   apptainer run --nv vit_imagenet.sif python train.py --epoch 120<\/code><br \/>\n   \u5e73\u5747\u6392\u961f\u65f6\u95f4 &lt; 90 \u79d2\uff0c3000 \u5f20 GPU \u6c60\u52a8\u6001\u5f39\u6027\u6269\u5bb9\uff0c\u9ad8\u5cf0\u65f6\u6bb5\u81ea\u52a8\u62c9\u8d77\u5907\u7528\u5361\uff0c<strong>\u4fdd\u8bc1\u4f60\u7684\u5b9e\u9a8c\u5728 DDL \u524d\u8dd1\u5b8c<\/strong>\u3002<\/p>\n<\/li>\n<\/ol>\n<h2>\u6027\u80fd\uff1aImageNet 120 epoch \u538b\u7f29\u5230 2 \u5c0f\u65f6<\/h2>\n<p>\u5728\u661f\u5b87\u667a\u7b97\u300ca100.8x\u300d\u5206\u533a\u5b9e\u6d4b\uff1a<br \/>\n&#8211; 8\u00d7A100 80 GB NVLink\uff0c\u6df7\u5408\u7cbe\u5ea6 + DDP\uff0cbatch 2048<br \/>\n&#8211; ResNet-50 120 epoch \u53ea\u9700 1.9 \u5c0f\u65f6\uff0c\u7ebf\u6027\u52a0\u901f\u6bd4 7.8\u00d7<br \/>\n&#8211; ViT-Base\/16 \u8bad\u7ec3\u65f6\u95f4\u4ece\u6821\u5185 3.5 \u5929\u964d\u81f3 5.2 \u5c0f\u65f6  <\/p>\n<p>\u5e73\u53f0\u540c\u65f6\u63d0\u4f9b <strong>RTX 4090 \u5f39\u6027\u5206\u533a<\/strong>\uff0c\u9002\u5408 ablation \u4e0e\u6d88\u878d\u5b9e\u9a8c\uff0c\u5355\u5361\u6bcf\u5c0f\u65f6 1.8 \u5143\uff0c<strong>\u652f\u6301 spot \u7ade\u4ef7\u6700\u4f4e 0.9 \u5143<\/strong>\uff0c\u7528\u591a\u5c11\u4ed8\u591a\u5c11\uff0c\u57fa\u91d1\u7ecf\u8d39\u8f7b\u677e cover\u3002<\/p>\n<h2>\u4ef7\u683c\uff1a\u6559\u80b2\u4e13\u9879 8 \u6298\uff0c\u5148\u7528\u540e\u4ed8<\/h2>\n<ul>\n<li>\u51ed <code>.edu.cn<\/code> \u90ae\u7bb1\u6ce8\u518c\u5373\u9001 <strong>10 \u5143\u4f53\u9a8c\u91d1<\/strong>\uff0c\u53ef\u8dd1 5 \u5c0f\u65f6 RTX 4090 \u6216 1 \u5c0f\u65f6 A100  <\/li>\n<li>\u6559\u80b2\u8ba4\u8bc1\u540e\u81ea\u52a8 8 \u6298\uff0c\u8d26\u5355\u53ef\u5ef6\u671f 30 \u5929\u652f\u4ed8\uff0c\u5b8c\u7f8e\u5339\u914d\u9ad8\u6821\u8d22\u52a1\u62a5\u9500\u5468\u671f  <\/li>\n<li>\u591a\u4eba\u8bfe\u9898\u7ec4\u53ef\u5171\u4eab\u300c\u56e2\u961f\u94b1\u5305\u300d\uff0c\u7edf\u4e00\u5f00\u7968\uff0c\u514d\u53bb\u6765\u56de\u57ab\u8d44  <\/li>\n<\/ul>\n<h2>\u751f\u6001\uff1a\u4e0d\u6b62\u7b97\u529b\uff0c\u8fd8\u6709\u6a21\u578b\/\u6570\u636e\u4e00\u952e\u8c03\u7528<\/h2>\n<p>\u9664\u4e86 <a href=\"https:\/\/www.starverse-ai.com\">GPU\u670d\u52a1\u5668\u79df\u7528<\/a> \u672c\u8eab\uff0c\u661f\u5b87\u667a\u7b97\u628a\u5e38\u7528 AI\u5e94\u7528 \u505a\u6210\u4e86\u300c\u5e94\u7528\u955c\u50cf\u300d\uff1a<br \/>\n&#8211; Diffusers\u3001DeepSpeed\u3001Colossal-AI\u3001MMDetection\u3001Detectron2 \u5f00\u7bb1\u5373\u7528<br \/>\n&#8211; \u5185\u7f6e\u516c\u5f00\u6570\u636e\u96c6 ImageNet\u3001COCO\u3001LAION-5B\uff0c\u6302\u8f7d\u5373\u7528\uff0c\u8282\u7701 50 % \u4e0b\u8f7d\u65f6\u95f4<br \/>\n&#8211; 100 GB \u6301\u4e45\u5316\u4e91\u76d8\u8de8\u5b9e\u4f8b\u5171\u4eab\uff0c\u65ad\u70b9 checkpoint \u81ea\u52a8\u4fdd\u5b58\uff0c\u5173\u673a\u4e0d\u4e22\u6570\u636e  <\/p>\n<h2>\u7533\u8bf7\uff1aedu \u90ae\u7bb1\u79d2\u6279\uff0c\u5165\u53e3\u76f4\u8fbe<\/h2>\n<ol>\n<li>\u6253\u5f00<a href=\"https:\/\/www.starverse-ai.com\/academic\">\u5b66\u672f\u901a\u9053<\/a>  <\/li>\n<li>\u8f93\u5165 <code>.edu.cn<\/code> \u90ae\u7bb1\uff0c\u7cfb\u7edf\u81ea\u52a8\u9a8c\u8bc1\uff0c30 \u79d2\u901a\u8fc7  <\/li>\n<li>\u9886\u53d6 10 \u5143\u4f53\u9a8c\u91d1\uff0c\u7acb\u523b\u5f00\u901a 3000 \u5361 GPU \u6c60\u6743\u9650  <\/li>\n<\/ol>\n<p><strong>\u73b0\u5728\u6ce8\u518c\uff0c\u518d\u9001 50 GB \u6301\u4e45\u5316\u4e91\u76d8\uff0c\u8dd1\u5b8c\u5b9e\u9a8c\u6570\u636e\u53ef\u7559\u53ef\u5220\uff0cDDL \u4e4b\u540e\u4e5f\u80fd\u968f\u65f6\u590d\u73b0\u3002<\/strong><\/p>\n<h2>\u5199\u5728\u6700\u540e<\/h2>\n<p>\u9876\u4f1a\u6295\u7a3f\u4e0d\u662f\u201cidea \u7684\u72ec\u89d2\u620f\u201d\uff0c\u66f4\u662f<strong>\u7b97\u529b\u4e0e\u65f6\u95f4\u7684\u8d5b\u8dd1<\/strong>\u3002\u661f\u5b87\u667a\u7b97\u7528\u4e91\u7aef\u7684 3000 \u5f20 GPU\u3001\u539f\u751f Slurm+Apptainer \u4f53\u9a8c\uff0c\u4ee5\u53ca\u6559\u80b2\u4e13\u5c5e\u6298\u6263\uff0c\u628a\u201c\u6392\u961f 3 \u5929\u201d\u53d8\u6210\u201c\u63d0\u4ea4 3 \u79d2\u201d\u3002\u4e0b\u4e00\u6b21 DDL\uff0c\u65e0\u8bba\u4f60\u6253\u7b97\u51b2 AAAI\u3001ICLR \u8fd8\u662f CVPR\uff0c\u81f3\u5c11\u4e0d\u7528\u518d\u71ac\u591c\u5237\u65b0 <code>squeue<\/code>\u3002<br \/>\n\u5b9e\u9a8c\u8dd1\u5f97\u5feb\uff0c\u8bba\u6587\u6295\u5f97\u65e9\uff0c\u661f\u5b87\u667a\u7b97\u613f\u505a\u4f60\u79d1\u7814\u8def\u4e0a\u7684\u201c\u9690\u5f62\u5171\u540c\u4e00\u4f5c\u201d\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u201cAAAI \u622a\u7a3f\u524d 72 \u5c0f\u65f6\uff0c\u6211\u7684 ablation \u5b9e\u9a8c&hellip;<\/p>\n","protected":false},"author":2,"featured_media":2726,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-2727","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-zixun"],"views":41,"_links":{"self":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts\/2727","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/comments?post=2727"}],"version-history":[{"count":0,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts\/2727\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/media\/2726"}],"wp:attachment":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/media?parent=2727"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/categories?post=2727"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/tags?post=2727"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}