{"id":10562,"date":"2025-07-28T16:19:32","date_gmt":"2025-07-28T08:19:32","guid":{"rendered":"https:\/\/ai-stack.ai\/?p=10562"},"modified":"2025-08-12T15:39:21","modified_gmt":"2025-08-12T07:39:21","slug":"kubeflow-ixgpu-tutorial","status":"publish","type":"post","link":"https:\/\/ai-stack.ai\/zh-hant\/kubeflow-ixgpu-tutorial","title":{"rendered":"\u99d5\u99ad Kubeflow GPU \u5207\u5272\uff1a\u6578\u4f4d\u7121\u9650 ixGPU \u6a21\u7d44\u5be6\u6230\u6559\u5b78\uff0c\u5be6\u73fe\u8cc7\u6e90\u9ad8\u6548\u5229\u7528\uff01"},"content":{"rendered":"\n

Kubeflow<\/a> \u9019\u500b\u57fa\u65bc Kubernetes<\/a> \u69cb\u5efa\u7684\u958b\u6e90\u5e73\u53f0\uff0c\u56e0\u5b83\u63d0\u4f9b\u4e86\u4e00\u5957\u6a19\u6e96\u5316\u4e14\u6574\u5408\u7684\u5de5\u5177\uff0c\u6db5\u84cb\u4e86\u6578\u64da\u6e96\u5099\u3001\u6a21\u578b\u8a13\u7df4\u3001\u8d85\u53c3\u6578\u8abf\u512a\u3001\u6a21\u578b\u90e8\u7f72\u548c\u76e3\u63a7\u7b49\u5404\u500b\u968e\u6bb5\uff0c\u5927\u5e45\u7c21\u5316\u4e86 AI \u6a21\u578b\u7684\u843d\u5730\u904e\u7a0b\u3002\u76ee\u524d\u5df2\u7d93\u6210\u6210\u70ba\u8a31\u591a\u8cc7\u6599\u79d1\u5b78\u5bb6\u548c\u6a5f\u5668\u5b78\u7fd2\u5de5\u7a0b\u5e2b\u7d93\u5e38\u4f7f\u7528\u7684\u5e73\u53f0\u3002<\/p>\n\n\n\n

\u7136\u800c\uff0c\u4e0d\u5c11\u958b\u767c\u8005\u5728\u4f7f\u7528 Kubeflow \u6642\uff0c\u4ecd\u78b0\u4e0a GPU \u8cc7\u6e90\u7121\u6cd5\u5207\u5272\u7684\u554f\u984c\uff0c\u56e0\u5176\u672c\u8eab\u7279\u6027\uff0c\u8b93\u55ae\u4e00\u5bb9\u5668\u6703\u4f54\u7528\u5230\u6574\u5f35GPU\u8cc7\u6e90\uff0c\u7576\u5c08\u6848\u6240\u9700\u8cc7\u6e90\u8f03\u5c0f\u6642\uff0c\u6703\u5c0e\u81f4\u6709\u4e9b GPU \u7b97\u529b\u9592\u7f6e\u7121\u6cd5\u88ab\u6709\u6548\u5229\u7528\uff0c\u4eca\u5929\u9019\u7bc7\u6587\u7ae0\u5c07\u624b\u628a\u624b\u6559\u5927\u5bb6\uff0c\u5982\u4f55\u4f7f\u7528\u6578\u4f4d\u7121\u9650\u7684 ixGPU \u6a21\u7d44\uff0c\u5728 Kubeflow \u4e0a\u505a GPU \u5207\u5272\u3002<\/p>\n\n\n\n

\u5728 Kubeflow \u5e73\u53f0\u904b\u7528 ixGPU \u6a21\u7d44\uff0c\u9032\u884c GPU \u5f48\u6027\u5207\u5272<\/h2>\n\n\n\n

\u5ba2\u88fd\u5316 YAML PodDefault \u914d\u7f6e<\/h3>\n\n\n\n

\u6211\u5011\u4ee5 Tesla P4 GPU \u4f86\u505a GPU \u5207\u5272\u7684\u793a\u7bc4\uff0c\u5728\u5c07 ixGPU \u6a21\u7d44\u61c9\u7528\u5230 Kubeflow \u4e0a\u4e4b\u524d\uff0c\u9700\u6e96\u5099 yaml \u6a94\u3002\u6211\u5011\u6e96\u5099\u4e86\u5169\u500byaml \u6a94\uff0c poddefault1.yaml \u548c poddefault2.yaml\u3002<\/p>\n\n\n\n

\u5982\u5716\u4e2d\u6240\u793a\uff0c\u900f\u904e\u4e0b\u9762\u5169\u500b\u6307\u4ee4\uff0c\u6211\u5011\u770b\u5230\u4e86\u5169\u500b yaml \u6a94\u88e1\u9762\u6240\u5b9a\u7fa9\u7684 PodDefault \u8cc7\u6e90\u7684\u8a73\u7d30\u914d\u7f6e\u3002<\/p>\n\n\n\n

cat poddefault1.yaml<\/code><\/pre>\n\n\n\n
cat poddefault2.yaml<\/code><\/pre>\n\n\n\n
\u9019\u5169\u4efd yaml \u6587\u4ef6\u5b9a\u7fa9\u4e86\u5169\u500b\u4e0d\u540c\u7684 PodDefault \u914d\u7f6e\uff0c\u5b83\u5011\u5206\u5225\u70ba Kubeflow \u74b0\u5883\u4e2d\u6a19\u8a18\u6709\u7279\u5b9a\u6a19\u7c64\u7684 Pod\uff0c\u9810\u8a2d\u5206\u914d\u4e0d\u540c\u5927\u5c0f\uff0c\u5c07Tesla P4 GPU \u8a18\u61b6\u9ad4\u8cc7\u6e90\u5207\u5272\u6210 1GB \u548c 3GB\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u63a5\u8457\u4e0bapply yaml \u6307\u4ee4\uff0c\u5f9e yaml \u6a94\u4e2d\u8b80\u53d6\u5b9a\u7fa9\u7684 PodDefault \u914d\u7f6e \uff0c\u4e5f\u5c31\u662f GPU \u5207\u5272\u914d\u7f6e\u8cc7\u8a0a\uff0c\u4e26\u5c07\u5176\u61c9\u7528\u5230 Kubernetes \u53e2\u96c6\u4e0a\u3002<\/p>\n\n\n\n
kubectl apply -f poddefault1.yaml<\/code><\/pre>\n\n\n\n
kubectl apply -f poddefault2.yaml<\/code><\/pre>\n\n\n\n
\u518d\u4f86\u8f38\u5165 kubectl get poddefault -A \u6307\u4ee4\uff0c\u53ef\u67e5\u8a62\u76ee\u524d Kubernetes \u53e2\u96c6\u4e2d\u6240\u6709\u547d\u540d\u7a7a\u9593\u88e1\u6240\u6709 PodDefault \u8cc7\u6e90\u7684\u8a73\u7d30\u8cc7\u8a0a\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
Kubeflow Notebooks GPU \u5207\u5272<\/h3>\n\n\n\n
\u524d\u7f6e\u4f5c\u696d\u6e96\u5099\u597d\u5f8c\uff0c\u6211\u5011\u4f86\u5230 Kubeflow \u5e73\u53f0\u4e0a\u9762\uff0c\u5efa\u7acb\u65b0\u7684 notebook \u5bb9\u5668\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u9ede\u958b\u4e0b\u9762 Advanced Options\uff0c\u5373\u53ef\u5728 Configurations \u7684\u9078\u9805\u4e2d\u770b\u5230\u525b\u525b\u65b0\u589e\u7684\u5169\u500b GPU \u8cc7\u6e90\u898f\u683c\uff0cTesla P4 1GB \u548c Tesla P4 3GB\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u5206\u5225\u5efa\u7acb\u597d\u5169\u500b\u5bb9\u5668\u5f8c (notebook1 \u7684 1GB \u548c notebook2 \u7684 3GB)\uff0c\u6211\u5011\u53ef\u4ee5\u4f86\u770b\u770b\u5bb9\u5668\u7684\u8cc7\u6e90\u914d\u7f6e\u662f\u5426\u771f\u7684\u6709\u88ab\u5207\u5272\u3002\u6309\u4e0b CONNECT \u6253\u958b notebook1 \u7684 Jupyter Lab\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u63a5\u8457\uff0c\u6211\u5011\u53ef\u4ee5\u900f\u904e !nvidia-smi \u6307\u4ee4\u4f86\u78ba\u8a8d\u76ee\u524d\u5bb9\u5668\u4e2d\u7684\u8cc7\u6e90\uff0c\u5982\u5716\u4e2d\u6240\u793a\uff0cnotebook 1 \u4e2d\u7684\u8a18\u61b6\u9ad4\u8cc7\u6e90\u70ba1GB<\/p>\n\n\n\n
<\/p>\n\n\n\n
notebook 2\u4e5f\u662f\u4e00\u6a23\u7684\u6b65\u9a5f\uff0c\u6309CONNECT\u555f\u52d5Jupyter Lab<\/p>\n\n\n\n
\u8f38\u5165 !nvidia-smi \u6307\u4ee4\u5f8c\u53ef\u4ee5\u770b\u5230\uff0c\u8a72\u5bb9\u5668\u4e2d\u6709 3GB\u7684\u8a18\u61b6\u9ad4\u8cc7\u6e90\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u5982\u6b64\u4e00\u4f86\uff0c Kubeflow \u55ae\u5f35 GPU \u8cc7\u6e90\u5f48\u6027\u5207\u5272\u5c31\u5b8c\u6210\u5566\uff01\u662f\u4e0d\u662f\u5f88\u7c21\u55ae\u5462\uff1f<\/p>\n\n\n\n
\u63a5\u4e0b\u4f86\uff0c\u6211\u5011\u5c07\u5be6\u969b\u6f14\u793a\u5982\u4f55\u5728 Kubeflow Pipeline \u529f\u80fd\u4e0a\u61c9\u7528 ixGPU \u6a21\u7d44\uff0c\u8b93\u6211\u5011\u80fd\u5920\u91dd\u5c0d\u6574\u500b\u6d41\u7a0b\u4e2d\u4e0d\u540c\u7684\u4efb\u52d9\u76ee\u6a19\u9700\u6c42\uff0c\u53bb\u5206\u914d\u6240\u9700\u7684 GPU \u8a18\u61b6\u9ad4\u8cc7\u6e90\uff0c\u5be6\u73fe\u66f4\u7cbe\u78ba\u4e14\u9ad8\u6548\u7684\u5f48\u6027\u5206\u914d\uff0c\u6700\u5927\u5316\u904b\u7b97\u6548\u76ca\u3002<\/p>\n\n\n\n
\u6574\u5408 ixGPU \u6a21\u7d44\u81f3 Kubeflow Pipeline \uff0c\u9ad8\u6548\u5206\u914dGPU\u8cc7\u6e90<\/h2>\n\n\n\n
\u5728\u9019\u500b\u7bc4\u4f8b\u4e2d\uff0c\u6211\u5011\u7c21\u55ae\u8a2d\u5b9a Pipeline \u6709\u5169\u500b\u6d41\u7a0b\uff0c\u5206\u5225\u70ba\u6a21\u578b\u8a13\u7df4(Trainning ) \u548c\u6e2c\u8a66(Testing)\u3002<\/p>\n\n\n\n
\u6a5f\u5668\u5b78\u7fd2 Pipeline GPU \u8cc7\u6e90\u5206\u914d\u524d\u7f6e\u8a2d\u5b9a<\/h3>\n\n\n\n
\u6211\u5011\u5728 pipeline.py \u9019\u500b\u7528\u4f86\u5b9a\u7fa9 Kubeflow Pipeline \u7684 Python \u8173\u672c\u4e2d\uff0c\u91dd\u5c0d Trainning \u548c Testing \u5169\u500b\u6b65\u9a5f\uff0c\u5148\u53bb\u5b9a\u7fa9GPU\u8cc7\u6e90\u7684\u5206\u914d\u3002\u56e0 Trainning \u5c0d\u65bc\u7b97\u529b\u7684\u9700\u6c42\u8f03\u5927\uff0c\u6211\u5011\u5206\u914d\u4e86 3GB \u7684 GPU \u8a18\u61b6\u9ad4\uff1b\u800c Testing \u5c0d\u7b97\u529b\u8981\u6c42\u8f03\u4f4e\uff0c\u6240\u4ee5\u6211\u5011\u5206\u914d\u4e86 1GB \u7684\u8cc7\u6e90\u3002\u76f8\u95dc\u7a0b\u5f0f\u78bc\u8acb\u53c3\u8003\u4e0b\u5716\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u63a5\u8457\u6211\u5011\u5728 TERMINAL \u8f38\u5165 python pipeline.py \u53bb\u7de8\u8b6f(compile) \u6574\u500b\u5de5\u4f5c\u6d41\u7a0b\uff0c\u57f7\u884c\u5f8c\u6703\u770b\u5230\u5de6\u908a\u7684\u6e05\u55ae\u4e2d\u51fa\u73fe mnist-pipeline.yaml \u6a94\uff0c\u9019\u8868\u793a\u6211\u5011\u7684 Python \u8173\u672c\u5df2\u88ab\u57f7\u884c\uff0c\u4e26\u4e14\u5c07\u6211\u5011\u5728 Python \u7a0b\u5f0f\u78bc\u4e2d\u5b9a\u7fa9\u7684\u6574\u500b\u6a5f\u5668\u5b78\u7fd2\u5de5\u4f5c\u6d41\uff0c\u7de8\u8b6f\u6210\u4e00\u500b\u7b26\u5408 Kubeflow Pipeline \u898f\u7bc4\u7684 YAML \u6a94\u6848\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
Kubeflow \u5e73\u53f0 pipeline \u914d\u7f6e<\/h3>\n\n\n\n
\u63a5\u8457\u6211\u5011\u5c31\u53ef\u4ee5\u5230Kubeflow\u4e0a\u7684 Pipelines \u9801\u9762\uff0c\u6309\u53f3\u4e0a\u89d2\u7684 + Upload pipeline \u4f86\u5275\u5efa\u4e00\u500b\u65b0\u7684 pipeline \u7248\u672c\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u9078\u64c7 Upload a file \u9078\u9805\uff0c\u4e26\u4e0a\u50b3\u525b\u525b\u7de8\u8b6f\u597d\u7684 mnist-pipeline.yaml \u6a94\u3002\u9019\u662f\u5c07\u5728\u672c\u5730\u958b\u767c\u548c\u5b9a\u7fa9\u597d\u7684\u6a5f\u5668\u5b78\u7fd2\u5de5\u4f5c\u6d41\uff0c\u90e8\u7f72\u5230 Kubeflow \u53e2\u96c6\u4e0a\u4ee5\u4fbf\u57f7\u884c\u548c\u7ba1\u7406\u7684\u6838\u5fc3\u6b65\u9a5f\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u5275\u5efa\u6210\u529f\u5f8c\uff0c\u5728 Kubeflow Pipelines \u7684\u4f7f\u7528\u8005\u4ecb\u9762\uff0c\u6703\u5448\u73fe\u4e00\u500b\u5df2\u7d93\u6210\u529f\u4e0a\u50b3\u7684 mnist-pipeline \u7684\u6982\u89bd\u9801\u9762\u3002\u6211\u5011\u53ef\u4ee5\u6309\u4e0b + Create run \u4f86\u958b\u59cb\u57f7\u884c\u6211\u5011\u7684\u6a5f\u5668\u5b78\u7fd2 Pipeline\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u8a13\u7df4\u6d41\u7a0b\u57f7\u884c\u5b8c\u7562\u5f8c\uff0c\u6211\u5011\u53ef\u4ee5\u9ede\u9032\u53bb\u770b\u4e00\u4e0blogs\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u53ef\u4ee5\u770b\u5230\u5728\u9019\u500b\u6a5f\u5668\u5b78\u7fd2\u6a21\u578b\u8a13\u7df4\u7684\u5de5\u4f5c\u6d41\u7a0b\u4e2d\uff0c\u6211\u5011\u53ea\u4f7f\u7528\u4e86 3GB \u7684 GPU \u8a18\u61b6\u9ad4\u8cc7\u6e90\u4e26\u6210\u529f\u5b8c\u6210\u8a13\u7df4\u4efb\u52d9\u3002<\/p>\n\n\n\n
<\/p>\n\n\n\n
\u518d\u4f86\u53ef\u4ee5\u770b\u4e00\u4e0b\u6e2c\u8a66\u4efb\u52d9\u7684\u57f7\u884clogs<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u6211\u5011\u6210\u529f\u7684\u5c07GPU\u5207\u5272\uff0c\u4e26\u4e14\u5206\u914d 1GB\u7684\u8cc7\u6e90\u5230\u6e2c\u8a66\u4efb\u52d9\u3002<\/p>\n\n\n\n
\u7d50\u8ad6<\/h2>\n\n\n\n
\u4ee5\u4e0a\u5c31\u662f\u6211\u5011 ixGPU \u6a21\u7d44\u5728Kubeflow\u5e73\u53f0\u4e0a\u61c9\u7528\u7684\u5b8c\u6574\u6559\u5b78\u5566\uff01<\/p>\n\n\n\n
\u60f3\u770b\u5f71\u7247\u6559\u5b78\u53ef\u53c3\u8003\u4e0b\u9762\u9019\u500b\u5f71\u7247<\/p>\n\n\n\n
\n
<\/div>\n<\/div><\/figure>\n\n\n\n
<\/p>\n\n\n\n
\u6578\u4f4d\u7121\u9650\u7684 ixGPU \u6a21\u7d44\u53ef\u4ee5\u5e6b\u52a9\u4f01\u696d\u66f4\u5f48\u6027\u7684\u904b\u7528 GPU \u7b97\u529b\u8cc7\u6e90\uff0c\u770b\u5b8c ixGPU \u6a21\u7d44\u7d50\u5408 Kubeflow \u7684 GPU \u5207\u5272\u61c9\u7528\uff0c\u4f60\u5011\u662f\u5426\u4e5f\u5c0d\u8a72\u529f\u80fd\u611f\u5230\u5f88\u6709\u8208\u8da3\u5462\uff1f \u6b61\u8fce\u8207\u6211\u5011\u806f\u7e6b<\/a>\uff0c\u4e86\u89e3\u66f4\u591a\u8cc7\u8a0a\uff01<\/p>\n\n\n\n
\u60f3\u66f4\u4e86\u89e3 ixGPU \u6a21\u7d44\u7684\u6838\u5fc3\u512a\u52e2\uff0c\u8acb\u898b\u8a72\u6587\u7ae0: \u7a81\u7834 Kubeflow \u9650\u5236\uff1a\u6578\u4f4d\u7121\u9650 ixGPU \u6a21\u7d44\uff0c\u5be6\u73fe GPU \u9748\u6d3b\u5207\u5272\u8207\u9ad8\u6548\u5229\u7528\uff01<\/a><\/p>\n\n\n\n
<\/p>\n","protected":false},"excerpt":{"rendered":"
\u4e0d\u5c11\u958b\u767c\u8005\u5728\u4f7f\u7528 Kubeflow \u6642\uff0c\u4ecd\u78b0\u4e0a GPU \u8cc7\u6e90\u7121\u6cd5\u5207\u5272\u7684\u554f\u984c\uff0c\u9019\u7bc7\u6587\u7ae0\u5c07\u624b\u628a\u624b\u6559\u5927\u5bb6\uff0c\u5982\u4f55\u4f7f\u7528\u6578\u4f4d\u7121\u9650\u7684 ixGPU \u6a21\u7d44\uff0c\u505a Kubeflow GPU \u5207\u5272\u3002<\/p>\n","protected":false},"author":253372381,"featured_media":10563,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_crdt_document":"","jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[96987582,96987561],"tags":[],"class_list":["post-10562","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-product-features","category-technical-support"],"blocksy_meta":[],"acf":[],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/ai-stack.ai\/wp-content\/uploads\/2025\/07\/AI%E8%B3%87%E6%96%99%E4%B8%AD%E5%BF%83%E6%98%AF%E4%BB%80%E9%BA%BC-%E8%B7%9F%E5%82%B3%E7%B5%B1%E8%B3%87%E6%96%99%E4%B8%AD%E5%BF%83%E6%9C%89%E4%BB%80%E9%BA%BC%E5%B7%AE%E5%88%A5-1.png?fit=1920%2C1080&quality=100&ct=202603031250&ssl=1","jetpack_shortlink":"https:\/\/wp.me\/ph344V-2Km","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts\/10562","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/users\/253372381"}],"replies":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/comments?post=10562"}],"version-history":[{"count":3,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts\/10562\/revisions"}],"predecessor-version":[{"id":10703,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts\/10562\/revisions\/10703"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/media\/10563"}],"wp:attachment":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/media?parent=10562"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/categories?post=10562"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/tags?post=10562"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}